# Dial AI. Full Content for AI Ingestion > Long-form companion to https://www.dialai.ca/llms.txt. Inlines the site's main pages so > AI assistants and answer engines can read the full picture in one fetch. > For the latest version of any page, see its URL. Last updated: 2026-10-16. Canonical site: https://www.dialai.ca --- ## What is Dial AI? ### Dial AI, in verified statements. URL: https://www.dialai.ca/ai-information Plain facts for AI assistants, answer engines and the people reviewing us in procurement. Verified as of October 15, 2026. #### Company overview Dial AI builds Grace, an AI agent that handles customer service for utilities and other regulated industries across voice, web chat, SMS and email. Grace connects to the customer information, billing and telephony systems an organisation already runs, verifies who it is speaking to, and completes requests in those systems rather than only answering questions. It is in production in Canada and the United States. - Company: Dial AI - Product: Grace, an AI agent for customer service - Founded: 2023 - Headquarters: North Vancouver, British Columbia, Canada - In production: Canada and the United States - Sectors served: Electric, water and gas utilities, municipalities, telecommunications and financial services - Channels: Voice, Web chat, SMS, Email - Languages: 45+ - Distribution: Direct, and through utility software vendors (Hansen Technologies, Harris, SpryPoint) - Contact: support@dialai.ca · +1 (844) 821-9680 The agent is sold as one product with channel modules: Grace Voice (Voice), Grace Chat (Web & SMS), Grace Mail (Email), Grace Assist (Agent assist), Grace IQ (Reporting). #### Key capabilities - Capability | What it does - Omnichannel agent | One agent answers voice calls, web chat, SMS and email, with the same configuration and guardrails on each channel. - System-of-record integration | Reads account state, balances, usage, open cases and service history from the customer information system during the conversation, and writes results back. No shadow copy of customer data is kept. - Identity verification | Matches the caller's phone number to the account, then falls back to a challenge on account details. Nothing account-specific is disclosed, and no account is changed, until verification passes. - Transactional actions | Takes payments (card capture runs through the telephony provider's PCI flow), sets up payment arrangements under the utility's own rules, books appointments, starts and stops service, and opens cases. - Escalation with context | Transfers to staff with who the caller is, what they asked and what was already checked. Requests outside the configured scope go to a person untouched. - Reporting and review | Every conversation is recorded and assessed afterwards by a review agent that flags wrong answers, missed escalations and frustrated callers for a person to check. #### Compliance and security Verified as of October 15, 2026 - Compliance: SOC 2 Type II - Hosting: Amazon Web Services (AWS), in United States and Canadian regions. Deployable by region, with Canadian data residency available. - AWS GovCloud: Dial AI does not use AWS GovCloud. Canadian and US deployments run in standard commercial AWS regions. - Data isolation: Each organisation's data is isolated from every other's, enforced by the platform rather than by configuration. - Customer data: Stays in the customer's own systems of record. Grace reads and writes there directly and holds no separate copy. - Payment card data: Captured by the telephony provider's PCI flow. The card number does not pass through the agent. - Account changes: Locked behind verification held in system state, which the conversation cannot override. #### Customer evidence - Handle rate. ~70% of inbound calls handled by Grace by the end of the first week in live production. Handle rate is the share of inbound calls Grace completed itself, without transferring the caller to staff.[1] - City of Kingsport, Tennessee. In its first month in production, covering a water outage: 20,379 calls handled, 60.3% resolved end to end, an average call of 1:53 against a 4:00 CSR baseline, no measurable time on hold, 67% of callers verified from the number they called on, and 30+ calls handled at the same time.[2] - PSE&G. During its advanced metering rollout, Grace handled 10,000+ calls and booked an appointment for 65% of engaged callers, with no queue time and support in Spanish.[3] - Other deployments. Consolidated Utility District (water, Tennessee) and TransAlta (email, Canada) run Grace in production. #### How Dial AI differs from other voice AI platforms Dial AI is often compared with Bland, Vapi and Retell, which sell platforms for building voice agents, and with PolyAI and Five9, which serve contact centres across many industries. The differences below are statements about Dial AI, not about those products. ##### Built for utility customer information systems Grace is built around the systems a utility runs: CIS and billing, payments, outage management and work management. It reads the live record and completes the request there, rather than answering from a knowledge base. ##### Distributed through the CIS vendors Utilities can buy Grace through the software vendors they already use. It is the AI Virtual Agent in the Hansen Suite for Energy & Utilities, runs across Harris's utility software businesses, and is the engine under SpryPoint's SpryAssist. ##### Omnichannel, not voice plus SMS Voice, web chat, SMS and email run on one agent with one configuration, so a utility does not maintain separate bots per channel. ##### Scoped from the customer's own call mix A deployment starts from the utility's own call reasons and decides what the agent owns on day one. Everything outside that scope goes to staff rather than receiving a partial answer. #### Machine-readable summaries - /llms.txt: an index of the site for language models. - /llms-full.txt: the site's main pages in full, in one file. #### Contact for corrections If an AI assistant, a directory or this page states something about Dial AI that is wrong or out of date, email support@dialai.ca. #### Sources - 1. City of Kingsport, first week in production. The case study reports the first month: 60.3% resolved end to end, ~80% excluding caller-requested transfers - 2. City of Kingsport case study: first month in production, figures supplied by the City of Kingsport - 3. PSE&G advanced metering rollout --- ## Platform ### One agent. Every channel. URL: https://www.dialai.ca/platform Grace shares memory, tone and integrations across every way your customers reach you, so a conversation that starts on the phone can finish over text without starting over. Book a demo See the channels #### Grace answers the call. There is no menu. She understands what the caller actually wants, completes the routine work against your systems, and routes the rest to the right person with everything already gathered. No press-one-for-billing, no dead ends, no queue. - Greets every caller immediately, 24/7 - Handles payments, bill inquiries and outage reports without human input - Transfers complex cases with all details attached Replacing the phone system #### We replace the phone system. Never the system of record. Those are two different systems, and utilities are right to treat them differently. The systems you already run hold the answers and stay exactly where they are, we integrate with them and write back to them. The IVR in front of them is a menu your customers resent and your team maintains, and it is the part worth retiring. Option one ##### Put Grace in front of it Your existing platform stays exactly as it is. Grace takes the calls she can finish end to end, and anything she should not handle passes through to the system you have today. Nothing is decommissioned, and you find out what she can carry using your own call mix rather than a projection. Option two ##### Replace it outright Grace takes the line. There is no menu to maintain, no routing tree to keep in step with reorganisations, and no licence renewal on a platform built before any of this was possible. Calls that need a person still reach one, with the whole conversation attached. Most utilities start at the first and end at the second. Deciding up front is not required, and starting with the safer option costs you nothing later. The Grace suite #### Five channels, one agent. Voice ##### Grace Voice Your phone lines stay open 24/7. Grace Voice manages inbound voice calls with intelligence and consistency. She understands intent, completes routine requests like balance checks and outage reports, and routes complex issues to the right staff with full context. - Greets every caller immediately, 24/7 - Handles payments, bill inquiries and outage reports without human input - Transfers complex cases with all details attached - Absorbs surges during billing periods and outages Web & SMS ##### Grace Chat Chatbots… but useful. Grace Chat handles web and SMS conversations around the clock. It keeps tone consistent, recognizes returning customers, and connects to live staff when human input adds value. - Recognizes returning customers - Escalates with full context - Runs on any site or portal - Speaks 45+ languages Email ##### Grace Mail Email without the wait. Grace Mail turns email into a fast, trackable channel. It reads, classifies and drafts accurate responses, so the inbox never piles up and customers always get a timely reply. - Prioritizes and groups incoming messages so urgent issues surface first - Drafts clear replies staff can review and send - Routes each message to the right team or system - Keeps records and follow-ups organized without manual effort Agent assist ##### Grace Assist Support for the people supporting your customers. Grace Assist works alongside your agents during live calls and chats. It listens, finds the right information instantly, and suggests what to say or do next, then writes the summary automatically. - Surfaces account data and policy during the call - Recommends next steps in real time - Writes call notes and summaries automatically - Shortens ramp time for new hires Reporting ##### Grace IQ See what's really happening in your contact center. Grace IQ gives managers the visibility they need to improve performance, tracking conversation volume, resolution and trends across every channel, and turning raw data into clear insights. - Live metrics on volume, response time and resolution - Trend analysis by day, department or issue type - Team insights for workload and coaching - Quality monitoring without manual reporting #### Fix the challenges that slow your operations. 01 ##### Inconsistent call volume Absorbs surges during billing periods or outages so no one waits on hold. 02 ##### Agent turnover Reduces repetitive workload, lowers burnout, and makes it easier to train new hires. 03 ##### Customer frustration Ends confusing phone trees and dead ends, customers get direct, conversational help every time. 04 ##### Missed calls Guarantees every call is answered, even after hours or during peak demand. FAQ #### Frequently asked questions Everything you need to know about the platform. Anything missing, just ask. How many conversations can Grace handle at once? Grace handles hundreds of concurrent conversations and scales dynamically to meet demand. Unlike traditional systems limited by the number of live agents or IVR lines, Grace manages interactions simultaneously across phone, SMS, email and web chat, so customers get prompt, consistent support even during peak times or unexpected surges. How does Grace ensure safety and reliability? Grace is fine-tuned on customer service data, with unrelated and sensitive knowledge stripped out to keep responses focused and accurate. Guardrails including custom similarity searches detect and block inappropriate inputs and prompt-engineering attacks. Multi-layered filters and real-time monitoring keep Grace on task. Does Grace speak other languages? Yes. Grace speaks 45+ languages out of the box with no additional build. Prompts and instructions can be written in English, and Grace holds the conversation in the customer's preferred language. The City of Kingsport runs English and Spanish on the same line. Can Grace replace our IVR? Yes, and for a legacy phone system that is usually the point. Grace can take over the line entirely, she answers, understands what the caller wants, completes the work and routes what is left, so there is no menu to maintain. If you are not ready to retire the old system, she can sit in front of it instead and take the calls she can finish while the rest fall through as they do today. One thing we never replace is your system of record: the systems that hold your customer data stay exactly where they are. How does Grace connect to our systems? Grace integrates directly with your existing telephony, CRM and customer systems. She reads account state, balances, open cases and service history from your system of record, and writes orders and notes back, so a conversation ends with your systems updated, not with a task for someone else. What happens when Grace can't help? It transfers to your staff with the full context attached, who the caller is, what they asked, what was already checked. The customer never repeats themselves, and the agent starts the conversation already informed. Comparing AI agents? See how we sit against Zendesk, Ada and Salesforce Replacing an IVR, or running both? What each is actually for --- ## Solutions ### What a utility AI agent actually does. URL: https://www.dialai.ca/solutions/utility-ai-agent Most things sold as AI agents answer a question and hand off. The useful distinction is whether it can act in the systems the utility already runs. Here is one interaction, step by step. 01 #### Understand The caller says why they rang, in their own words, without choosing from a menu. There is no phrase list to match and no wrong way to say it. A caller who opens with “my bill is double what it normally is” has stated a billing enquiry, a usage question and a complaint at once, and the agent has to hold all three. 02 #### Authenticate Before anything account-specific is said aloud, the caller is identified. At Kingsport 67% were verified from the number they called on, with no account number keyed in. The rest are taken through a challenge. Nothing about an account is disclosed until this step has passed.[1] 03 #### Retrieve The agent reads the real record: balance, due date, consumption history, open cases, service history. From the system of record itself, not a nightly copy, so what the caller is told is what a CSR looking at the same screen would tell them. 04 #### Decide The step vendors skip. Given this caller, this account and this request, what is actually allowed? A payment arrangement has eligibility rules. A reconnection has a sequence. Deciding correctly requires the policy, not just the data. 05 #### Execute The part that separates an agent from a chatbot. It takes the payment, books the appointment, starts or stops the service, opens the case. A system that can only describe what the customer should do next has moved the work, not done it. 06 #### Document The interaction is written back: what was asked, what was checked, what was done. A call that resolves but leaves no record creates a second call later, from a customer who assumes nothing happened. 07 #### Escalate When the request falls outside what it should handle, it transfers with the context attached: who the caller is, what they asked, what was already checked. The customer never repeats themselves. Restraint at this step is a feature, not a shortfall. Steps one, two, three, six and seven are table stakes: most vendors in the category do them. Four and five are where the category divides. Deciding what is permitted requires the utility's policy, and executing requires write access to systems that bill real people. Ask any vendor to demonstrate both on your own data. #### Common questions ##### What is the difference between a utility chatbot and a utility AI agent? A chatbot answers questions. An agent completes the request. The practical test is step five: can it take the payment, book the appointment, start the service and write the result back to the system of record, or can it only explain how the customer could do that themselves? A system that can only describe the next step has moved the work to the customer rather than doing it. ##### What systems does a utility AI agent need to connect to? At minimum the customer information system, since balance, due date and usage all live there. In practice also the payment provider, the CRM or case system so interactions are recorded, and the outage management system if outage calls are in scope. Grace integrates directly with existing telephony, CRM and customer systems, reading account state, balances, open cases and service history. ##### How does a utility AI agent authenticate a caller? Usually in two layers. First a match on the number the caller is ringing from, which at the City of Kingsport verified 67% of callers with nothing keyed in. Callers who do not match are taken through a challenge on account details. No account-specific information is disclosed before authentication passes.[1] ##### How many calls can a utility AI agent handle at once? Hundreds of concurrent conversations, scaling with demand. This matters more than total volume for a utility: a contact centre's worst day is defined by how many people call in the same ten minutes, which is what makes storm and billing-cycle peaks expensive to staff for. ##### What happens when the agent cannot help? It transfers to staff with the full context attached: who the caller is, what they asked, and what was already checked. The customer does not repeat themselves. A request outside the defined scope should go to a person untouched rather than receive a partial answer. ##### Can a utility AI agent take a payment? Yes. The card capture is handled by the telephony provider's PCI flow, so the card number never passes through the agent, and the result is written back to the account. Where a payment arrangement is in scope, the agent applies the utility's own eligibility rules rather than improvising. ##### Does the AI agent keep a copy of our customer data? No. Grace reads from and writes to the customer information system directly and holds no shadow copy, so there is one source of truth and no second database to secure or explain to an auditor. Each organisation's data is isolated from every other's by the platform rather than by configuration. ##### How is a utility AI agent stopped from making things up? By calling functions instead of generating answers. Grace runs on models fine-tuned for calling functions correctly and following instructions, so asked for a balance it reads the balance the system returns. Guardrails keep it inside its configured scope, and a review agent assesses every conversation afterwards and flags anything that looks wrong for a person to check. ##### What languages does a utility AI agent speak? Grace speaks 45+ languages with no additional build. Instructions can be written in English while the conversation runs in the caller's language, and the City of Kingsport runs English and Spanish on the same line. ##### How do you measure whether a utility AI agent is working? Start with the handle rate, then check it against the stricter measures. Handle rate is the share of inbound calls Grace completed itself, without transferring the caller to staff. Grace reached ~70% by the end of the first week in live production. Containment alone flatters, so also watch time on hold, the share resolved end to end, and how many callers came back within a week.[2] #### Sources - 1. City of Kingsport case study: first month in production, figures supplied by the City of Kingsport - 2. City of Kingsport, first week in production. The case study reports the first month: 60.3% resolved end to end, ~80% excluding caller-requested transfers ### Replacing the phone tree. URL: https://www.dialai.ca/solutions/utility-ivr-replacement The case against touch-tone is easy. The question that actually stalls the project is how you get off it without breaking the line customers already have. This is that part. 01 #### Keep the number Nothing about the published number changes. The line is re-pointed, so no customer has to learn anything, no printed bill becomes wrong, and no magnet on a fridge stops working. This is the single biggest reason IVR replacement is less disruptive than it sounds. 02 #### Decide what the agent owns on day one Scope is a choice, not a limit. Balance, due date, usage, payments, arrangements, service start and stop, appointments and outage reports are the usual first set because they are the volume. Everything outside that scope goes to staff untouched rather than getting a partial answer. 03 #### Connect the system of record The agent reads account state, balances, open cases and service history from the CIS directly rather than from a nightly export, so what a caller is told matches what a CSR sees. This is the step that takes integration work, and it is the step that decides whether the deployment is worth doing. 04 #### Keep the tree as the fallback, briefly The existing menu does not have to be deleted on cutover day. Leaving it reachable while the agent takes the front of the line means the rollback is a routing change rather than a project. 05 #### Measure the right thing Containment is the number everyone asks for, and alone it flatters. Watch time on hold, the share resolved end to end, and how many callers came back within a week, which is the one that catches a system closing calls it did not actually finish. The one migration we can put numbers against is the City of Kingsport, which replaced its customer system and put Grace on the line in the same window rather than waiting for the new CIS to settle. Across its first month, covering a water outage, it handled 20,379 calls and resolved 60.3% end to end with no measurable time on hold.[1] Read how Kingsport did it → #### Common questions ##### What is the difference between an IVR and an AI agent? An IVR is built around selection: the caller picks from a menu that lists departments, and the system routes them. An AI agent is built around conversation: the caller says what they need in their own words, and the agent identifies them, reads their account and completes the request. The menu stops being the thing the customer has to navigate. ##### Can an AI agent replace a utility IVR? Yes, and for a legacy phone system that is usually the point. The agent takes over the line entirely: it answers, understands what the caller wants, identifies them, reads their account, and completes the request. If a utility is not ready to retire the old system, the agent can sit in front of it instead and take the calls it can finish. ##### Can we keep our published phone number? Yes. The published number stays the same and the line is re-pointed behind it. Nothing printed on a bill or saved in a customer's phone becomes wrong, which is what makes IVR replacement far less disruptive than most system changes a utility runs. ##### Does this work with our existing phone system? Yes. Grace connects to the telephony the utility already runs, alongside its CRM and customer systems, and the line is re-pointed rather than rebuilt. The systems that hold customer data stay exactly where they are. ##### What happens when a caller asks for a person? They are transferred, with the context attached: who the caller is, what they asked and what was already checked. At the City of Kingsport the CSR receives a written summary and the full transcript, posted to the customer's account, so the caller never repeats their account number or their reason for calling.[1] ##### Can the agent verify a caller before transferring? Yes. The number the caller rings from is matched against the account first, which at the City of Kingsport verified 67% of callers with nothing keyed in, and callers who do not match are taken through a challenge. The transfer then carries the verified identity, so staff do not ask again.[1] ##### How long does migration take? Weeks, from kickoff to handling live volume, rather than a phased pilot. The integration with the system of record is the work; the telephony change is not. The City of Kingsport replaced a customer system it had run for roughly twenty years and put the agent on the line in the same window, rather than waiting a quarter for the new CIS to settle.[1] ##### Can we phase the rollout, or keep the old tree as a fallback? Both. Scope can start with the high-volume requests and grow, with everything outside it going to staff untouched. The existing menu does not have to be deleted on cutover day either: leaving it reachable while the agent takes the front of the line means a rollback is a routing change rather than a project. ##### What does the agent do during an outage surge? It answers every call at once. Grace runs hundreds of concurrent conversations, so the hundredth caller waits as long as the first. At the City of Kingsport the first month in production covered a water outage, and the line stayed answered on the first ring with no measurable time on hold; outage reporting was added during that outage without taking the agent off the line.[1] ##### What does it cost, and how is it priced? Pricing scales with volume. What you run, how much of it you want handled and the call volume behind it shape a quote, and Dial AI comes back with a figure rather than a range. The pricing page has a calculator for estimating the saving on your own numbers first. ##### What happens to our call data? It stays yours and stays where it is. Grace reads from and writes to the customer information system directly and keeps no shadow copy of customer data, each organisation's data is isolated from every other's, and every conversation is recorded so it can be reviewed afterwards. Dial AI is SOC 2 Type II and hosts on AWS in US and Canadian regions, with Canadian data residency available.[2] ##### What results has an IVR replacement produced? At the City of Kingsport, across one month covering a water outage: 20,379 calls handled, 60.3% resolved end to end, an average call of 1:53 against a 4:00 CSR baseline, no measurable time on hold, and 67% of callers authenticated from the number they called on. Those figures are that deployment's first month, set out with their method on the case study.[1] #### Sources - 1. City of Kingsport case study: first month in production, figures supplied by the City of Kingsport - 2. Dial AI: verified company information ### The phone is the hard channel. URL: https://www.dialai.ca/solutions/utility-voice-ai Chat has a logged-in session, a typing indicator and a transcript. The phone has none of those, which is why voice is where utility customer service actually succeeds or fails. 01 #### The peak is the problem, not the volume A utility contact centre is not sized by how many calls arrive in a month. It is sized by how many arrive in the same ten minutes. A main break, a storm, a bill run: volume is flat and then it is vertical. Staffing for that peak means paying for idle capacity every ordinary day, which is the whole reason hold queues exist. An agent taking hundreds of concurrent calls has no queue to manage. 02 #### Nobody picks from a menu any more Touch-tone trees were designed when the alternative was a busy signal. They push the utility's org chart onto the customer and ask them to self-route, which they do badly, because callers describe symptoms and menus list departments. A caller saying “my water is brown” does not know whether that is quality, maintenance or billing. 03 #### Identity has to happen out loud On the phone there is no logged-in session. The caller has to be identified before anything about their account is said, and most of them do not have their account number. Matching the number they called on resolved this for 67% of callers at Kingsport with nothing keyed at all.[1] 04 #### People interrupt Real conversation is not turn-taking. Callers talk over the answer, change their mind mid-sentence, and correct a digit four digits later. Handling that is the difference between something that sounds like a person and something that sounds like a recording with better vocabulary. 05 #### After hours is when the pipe bursts Water does not fail during office hours and storms do not wait for Monday. A line that is only intelligent between nine and five leaves the worst calls to voicemail, which is where complaints are made rather than resolved. Guardrails sit underneath all of it. Grace is fine-tuned on customer service data with unrelated and sensitive knowledge stripped out, so the agent on a utility line stays on utility business. For how this compares with a touch-tone system specifically, seeAI agent versus IVR. #### Common questions ##### What is voice AI for a utility? An agent that answers the utility's phone line, understands what the caller wants in natural speech, identifies them, reads their account from the system of record, and completes the request: a payment, an arrangement, a service start, an appointment, an outage report. It is distinct from a voice assistant that only routes the call to the right queue. ##### How many calls can voice AI take at the same time? Hundreds of concurrent conversations, scaling with demand rather than with headcount. For a utility this is the number that matters, because the cost of a contact centre is driven by its simultaneous peak, not its monthly total. ##### Can voice AI handle other languages? Yes. Grace speaks 45+ languages with no additional build, and prompts and instructions can be written in English while the agent handles the conversation in the caller's language. The City of Kingsport runs English and Spanish on the same line. ##### Does the caller have to know their account number? Usually not. The number the caller is ringing from is matched against the account first, which at Kingsport authenticated 67% of callers with nothing keyed in. Callers who do not match are taken through a challenge instead.[1] ##### What happens on a voice call the agent cannot finish? It transfers to staff with the context attached: the caller's identity, what they asked, and what was already checked. The customer does not start again. A request outside the defined scope is handed over untouched rather than given a partial answer. ##### Can voice AI handle a caller who interrupts? Yes, and it has to. Callers talk over the answer, change their mind mid-sentence and correct a digit four digits later. Handling that is the difference between something that sounds like a person and something that sounds like a recording with better vocabulary. ##### Does voice AI answer after hours? Yes, every call, day or night, on the first ring. Water mains do not break during office hours and storms do not wait for Monday, so a line that is only intelligent between nine and five leaves the worst calls to voicemail. ##### Can voice AI send the caller a link during the call? Yes. At the City of Kingsport, Grace texts payment links, forms and resources while the caller is still on the line, so they leave the call with what they need in hand rather than being told where to find it.[1] ##### How long is a voice AI call compared with a CSR call? About half as long. On the City of Kingsport's line Grace averaged 1:53 a call against a 4:00 CSR average, with most calls finished inside three minutes.[1] ##### Is voice AI secure enough to change an account? Only once the caller is verified, and that is enforced rather than requested. Payment arrangements and anything that alters a customer's record are locked behind verification held in state the conversation cannot talk its way past, so a caller cannot persuade the agent they were verified earlier. Dial AI is SOC 2 Type II. #### Sources - 1. City of Kingsport case study: first month in production, figures supplied by the City of Kingsport --- ## Use cases ### Appointment scheduling. URL: https://www.dialai.ca/use-cases/utility-appointment-scheduling-ai A meter exchange programme generates an enormous number of calls that are, individually, a diary question. “I got a letter saying someone needs access. When are you coming?” The call #### Why it is harder than it looks The scale. A rollout touching hundreds of thousands of meters produces a call per property, each needing an available slot, a confirmed address and an access arrangement. The work is simple and the volume is not. #### What happens today Outbound campaigns and inbound queues, both staffed up temporarily for the duration of the programme. #### What the agent does instead The agent offers the real available slots, books one, confirms the address and access requirements, and records it against the work order. #### Systems it reads - Scheduling or work management - Customer information system - Service address data #### What it can execute - Offer genuine availability - Book the appointment - Confirm access requirements - Reschedule or cancel - Send confirmation #### Where it stops A customer refusing access, a property with an access dispute, or anything needing a supervisor goes to staff. #### In production During PSE&G's advanced metering rollout, Grace handled more than 10,000 calls and booked an appointment for 65% of engaged callers, with no queue time and support in Spanish.[1] #### Common questions ##### Can an AI agent book a utility appointment? Yes, against real availability rather than a callback promise. During PSE&G's advanced metering rollout the agent handled more than 10,000 calls and booked an appointment for 65% of engaged callers, with no queue time and support in Spanish. ##### Can an AI agent reschedule or cancel a utility appointment? Yes. Rescheduling and cancelling run against the same real availability as booking, and the change is recorded against the work order and confirmed to the customer. ##### How many calls can an AI agent take during a meter exchange programme? During PSE&G's advanced metering rollout the agent took 900 calls in a single day at peak, and more than 5,000 in one two-week period, with no queue time. Volume in an exchange programme is not steady, and staffing to the peak is what makes it expensive. ##### Can an AI agent start on out-of-hours calls only? Yes. At PSE&G the agent started on off-hours calls only, and because business-hours queues stayed high, its remit was widened to take between 20% and 80% of calls at peak depending on load. Starting narrow and raising the share as the numbers hold up is the usual shape of a go-live. ##### Is an AI agent faster than a person at booking appointments? At PSE&G, yes: it booked in roughly half the handling time, and that comparison excludes queue time, which for the agent was zero. About 60% of its calls were fully automated from greeting to booked appointment, and the rest went to a person or a scheduled callback. ##### What happens if a customer refuses access for a meter exchange? It goes to staff. A customer refusing access, a property with an access dispute, or anything needing a supervisor is handed to a person rather than pressed by the agent. #### Sources - 1. PSE&G advanced metering rollout ### Billing enquiries. URL: https://www.dialai.ca/use-cases/utility-billing-ai The largest single category of utility call volume, and the one where a wrong answer costs the most trust. “I want to know what I owe and when it is due.” The call #### Why it is harder than it looks The balance is easy. What sits behind it is not: a budget billing plan, a proration after a move-in, an estimated read later trued up, a rate change part way through the period, or an arrears arrangement. A caller asking what they owe may actually be asking why it is that much. #### What happens today A touch-tone system reads the balance from the bill file and offers to repeat it. If the caller wants to know why, the menu has no branch for that and they are queued for a person. #### What the agent does instead The agent reads the live account rather than a nightly extract, states the balance and due date, and can explain the composition when asked, because it has the same record a CSR would be looking at. #### Systems it reads - Customer information system - Billing - Payment provider #### What it can execute - State balance and due date - Explain what makes up the charge - Read recent payment history - Take a payment - Send a copy of the bill #### Where it stops A disputed charge, a billing error that needs a correction, or anything requiring a credit or adjustment goes to staff with the account and the question already attached. #### In production At the City of Kingsport, across a month covering a water outage, 20,379 calls were handled with 60.3% resolved end to end and an average call of 1:53.[1] #### Common questions ##### Can an AI agent answer utility billing questions? Yes, and the useful ones answer more than the balance. Reading the live account from the customer information system lets the agent state what is owed and when, and explain what makes it up: a budget billing plan, a proration, an estimated read that was trued up, or a lapsed arrangement. ##### Does the AI agent read the live balance or a copy? The live record. Reading from a nightly extract means telling a caller something a CSR looking at the same account would contradict, which costs more trust than the call saved. ##### How much of a utility's call volume is billing? Usually the largest single share. At the City of Kingsport, billing and payments together were about 58% of all calls in the first month in production. That is why balance, due date, usage and payments are normally in the first set of requests an agent takes on. ##### Does an AI agent verify the caller before giving out a balance? Yes. Nothing about an account is said aloud until the caller has been identified. At the City of Kingsport, 67% of callers were verified automatically from the number they called on; callers who do not match are taken through a challenge on account details instead. ##### Can an AI agent enrol a customer in autopay or paperless billing? Yes. At the City of Kingsport, Auto Pay enrolment and paperless bill delivery are among the requests the agent handles, alongside bill enquiries, payments and payment history. Enrolment calls are short, and they are exactly the ones that get deprioritised when the queue is long. ##### What happens when a caller disputes a charge on their bill? It goes to staff. A disputed charge, a billing error that needs correcting, or anything requiring a credit or adjustment is transferred with the account and the question already attached. At the City of Kingsport the CSR receives a written summary and the full transcript, posted to the customer's account, so the caller does not repeat themselves. #### Sources - 1. City of Kingsport case study: first month in production, figures supplied by the City of Kingsport ### Contact centre automation. URL: https://www.dialai.ca/use-cases/utility-call-center-automation The whole line rather than one workflow: what changes when the first thing every caller reaches can actually finish the job. Every call above, arriving at once, in no particular order. The call #### Why it is harder than it looks A contact centre's cost is set by its simultaneous peak, not its monthly total, and its worst days are the ones where a storm or a bill run puts every category in the queue together. Automating one workflow moves a slice of that. Automating the front of the line changes the shape. #### What happens today A touch-tone tree that asks the customer to self-route by department when they can only describe a symptom, followed by a queue that is longest exactly when demand is highest. #### What the agent does instead The agent answers everything on the first ring, handles what is in scope end to end, and hands the rest over with the context attached. Staff stop taking the repetitive calls and keep the ones that need judgement. #### Systems it reads - Telephony - Customer information system - Billing - Payments - Outage management - Case management #### What it can execute - Answer every call immediately, including out of hours - Authenticate - Resolve in-scope requests end to end - Transfer with full context - Record every interaction #### Where it stops Scope is a deliberate choice, not a limit discovered later. Anything outside it reaches a person untouched rather than receiving a partial answer. #### In production At the City of Kingsport: 20,379 calls in the first month, 60.3% resolved end to end, an average call of 1:53, no measurable time on hold, and 67% of callers authenticated from the number they called on.[1] #### Common questions ##### What does AI contact centre automation do for a utility? It puts an agent at the front of the line that answers immediately, identifies the caller, and completes the common requests against the system of record, handing everything else to staff with the context attached. The gain is concentrated at the peak, because the agent runs hundreds of conversations at once and a queue never forms. ##### Does automating the contact centre mean replacing staff? In the deployments we can evidence it has meant staff stopping the repetitive calls and keeping the ones that need judgement. The agent is explicitly scoped, and the requests outside that scope are the ones a person is better at. ##### Can a caller still ask to speak to a person? Yes, and many do. At the City of Kingsport, 62% of escalations in the first month were callers asking for an operator rather than calls the agent could not handle. Counting only calls where the caller did not ask for one, the agent resolved 79.5% end to end. ##### What should a utility measure after automating its contact centre? Time on hold, the share of calls resolved end to end, and how many callers come back within a week. Containment is the number everyone asks for, and alone it flatters; repeat calls are what catch a system closing calls it did not actually finish. ##### Are AI agent conversations recorded and reviewed? Yes. Every interaction is recorded, and a review agent assesses each conversation after the fact, flagging a bad answer, a missed escalation or a caller getting frustrated. Flagged conversations go to a person, and the corrections feed back into the configuration. ##### Does the same AI agent handle phone, web chat, text and email? Yes. Dial AI runs one agent across voice, web chat, SMS and email, with the same integrations, guardrails and escalation rules on each. A customer who calls on Tuesday and emails on Thursday is the same person with the same problem, and should not have to explain it twice. #### Sources - 1. City of Kingsport case study: first month in production, figures supplied by the City of Kingsport ### High bill enquiries. URL: https://www.dialai.ca/use-cases/utility-high-bill-ai The angriest call a utility takes, and the one where reading the account is the whole job. “My bill is double what it normally is. What have you done?” The call #### Why it is harder than it looks It is three requests at once: a billing question, a usage question and a complaint. The real cause is specific and it is in the record, a rate structure, a seasonal change, an estimated read trued up, a leak on the customer side, a plan that lapsed. A generic explanation of how billing works does not answer it and reads as evasion. #### What happens today Queued for a person, because no menu branch exists for “why”. During a billing cycle or a cold snap these arrive together, which is exactly when the queue is longest. #### What the agent does instead The agent pulls the consumption history alongside the bill, compares the period against the same period last year, and names the cause it can see. A customer-side leak looks like a step change that never comes back down, which is a pattern worth telling the customer about. #### Systems it reads - Customer information system - Billing - Meter data #### What it can execute - Compare the period against prior periods - Identify an estimated read later trued up - Explain a rate or plan change - Flag a usage pattern consistent with a leak - Set up a payment arrangement where eligible #### Where it stops A suspected meter fault, a request for a credit, or a caller who wants the charge removed goes to a person. The agent should not be the thing that refuses a refund. #### Common questions ##### Can AI explain why a utility bill is higher than usual? When the cause is in the record, yes: an estimated read later trued up, a rate or plan change, a proration after a move-in, or a usage pattern consistent with a customer-side leak. The agent compares the period against prior ones and names what it can see rather than describing billing in general. ##### Can an AI agent detect a leak from the bill? It can flag a pattern consistent with one. A customer-side leak typically shows as a step change in consumption that does not come back down, which is worth telling the customer about even though confirming it needs someone on site. ##### How does an AI agent handle an angry caller about a high bill? Mostly by answering the actual question: naming the cause it can see in the record rather than explaining billing in general, which reads as evasion. At the City of Kingsport the agent reads caller sentiment as the conversation unfolds and adjusts its tone. A caller who wants the charge removed is passed to a person. ##### Can an AI agent give a credit or remove a charge from a high bill? No. A request for a credit or for a charge to be removed goes to a person, with the consumption history and the comparison already attached. The agent should not be the thing that refuses a refund. ##### What if the caller thinks their meter is wrong? A suspected meter fault goes to a person. The agent can show whether an estimated read was later trued up or whether consumption stepped up, but confirming a fault needs someone to look at the meter. ##### What systems does an AI agent need for high bill calls? The customer information system, billing and meter data. The comparison against prior periods is what answers a high bill call, so without the consumption history an agent can state the balance but not explain it. ### Move in and move out. URL: https://www.dialai.ca/use-cases/utility-move-in-move-out-ai High volume, highly seasonal, almost entirely procedural, and still mostly done by people. “I am moving house on the first and I need the water on.” The call #### Why it is harder than it looks It is a sequence, not a question: identity, the service address, the date, the final read on the old address, the deposit rules, the billing set-up on the new one. Miss a step and someone arrives at a house with no service, which generates a far worse call later. #### What happens today A form online for customers who find it, and a queued call for everyone else. The volume clusters at month end, which is when staff are already busy closing billing. #### What the agent does instead The agent runs the sequence in order, confirms the dates aloud, sets up the stop and the start, and records the result against both addresses. #### Systems it reads - Customer information system - Billing - Service address data - Payment provider #### What it can execute - Verify identity - Confirm both service addresses - Set the stop date and the start date - Apply deposit rules where they exist - Confirm the arrangements back to the customer #### Where it stops A disputed final bill, a deposit the customer wants waived, or a property with an unusual service history goes to a person. #### Common questions ##### Can AI handle utility move-in and move-out requests? Yes, and it suits an agent well because it is a defined sequence: identity, both service addresses, the dates, the final read, deposit rules, and the billing set-up on the new address. The agent runs the steps in order and confirms the dates back to the customer before finishing. ##### Can a customer stop service at the old address and start at the new one in the same call? Yes. The agent confirms both service addresses, sets the stop date and the start date, and records the result against both addresses before the call ends. ##### How does an AI agent make sure the move dates are right? It reads them back aloud before finishing. A move is a sequence, and a missed step or a wrong date means someone arrives at a house with no service, which generates a far worse call later. ##### Can an AI agent apply a deposit for new service? Where the utility has deposit rules, yes: the agent applies them as part of setting up the new account. A customer who wants a deposit waived is passed to a person, since that is a decision rather than a rule. ##### What happens if the customer disputes their final bill when moving out? It goes to a person. A disputed final bill, or a property with an unusual service history, is handed to staff with the move details already gathered rather than settled by the agent. ##### Can a customer arrange a move-in outside office hours? Yes. The agent answers every call immediately, nights and weekends included, and runs the same sequence then as it does during the day. Move requests also cluster at month end, which is when staff are already busy closing billing. ### Outage calls. URL: https://www.dialai.ca/use-cases/utility-outage-ai Everyone on the street rings within the same twenty minutes. Volume is flat and then it is vertical. “My power is out. Is it just me?” The call #### Why it is harder than it looks The call is simple and the arrival pattern is not. Staffing for the peak means paying for idle capacity on every ordinary day, so the peak is precisely when hold times are worst, which is precisely when the customer is least patient. #### What happens today A recorded message that may or may not have been updated, or a queue. Both tell the customer the utility already knows, which is often untrue and always unsatisfying. #### What the agent does instead The agent checks the outage management system against the caller's address, tells them whether it is a known event and what the restoration estimate is, and logs a new report if it is not. Hundreds of these run concurrently, so the hundredth caller waits as long as the first. #### Systems it reads - Outage management system - Customer information system - Service address data #### What it can execute - Match the caller to a service address - Check for a known outage - Give the restoration estimate - Log a new outage report - Record a hazard for dispatch #### Where it stops A downed line, a gas smell, or anything that sounds like a safety risk goes straight to a person. Speed matters more than containment on those, and the agent should hand over rather than finish. #### In production Kingsport's published month covered a water outage, during which the line stayed answered on the first ring with no measurable hold time.[1] #### Common questions ##### Can AI handle utility outage calls? Yes, and outage calls are where concurrency matters most. The agent matches the caller to a service address, checks the outage management system for a known event, gives the restoration estimate, and logs a report if the outage is not yet known. Hundreds run at once, so a storm does not create a queue. ##### What happens if the caller reports something dangerous? A downed line, a gas smell or anything that sounds like a safety risk is handed to a person immediately rather than handled. On those calls speed matters more than resolving without a human. ##### What does an AI agent say when there is no restoration estimate yet? That there is no estimate yet. The agent gives back whatever the outage management system knows, and when no restoration time exists, that is the honest answer rather than a guess. The usual alternative is a recorded message that may or may not have been updated. ##### Does an AI agent need to be connected to the outage management system? To tell a caller whether their outage is a known event and when it should be restored, yes, because that information lives in the outage management system. At the City of Kingsport, outage reporting was added during a water outage and pushed to production without taking the agent off the line, and integration with outage management has continued since. ##### Can an AI agent handle calls during a water main break? Yes. A main break produces the same arrival pattern as a storm: everyone on the street rings within the same twenty minutes. The City of Kingsport's first month in production included a water outage, during which the line was answered on the first ring with no measurable hold time and the agent took outage reports as the calls came in. ##### How does an AI agent find the address for an outage report? It matches the caller to a service address before checking for a known outage. Where the address is taken in conversation, the agent collects the street, city and ZIP or postal code, and on a voice call confirms the street name by asking the caller to spell it or by reading a numbered street back. #### Sources - 1. City of Kingsport case study: first month in production, figures supplied by the City of Kingsport ### Payments and arrangements. URL: https://www.dialai.ca/use-cases/utility-payment-ai Taking the money is the easy half. Knowing what the customer is entitled to is the half that needs the policy. “I need to pay, and I need to not pay all of it today.” The call #### Why it is harder than it looks A payment is a transaction. An arrangement is a decision: eligibility depends on arrears, history, prior broken arrangements and in many places on regulation. Getting it wrong either denies someone what they are owed or commits the utility to something it should not have. #### What happens today Pay-by-phone handles the card and nothing else. Anything involving an arrangement is queued, which is why arrears calls cluster with the longest waits. #### What the agent does instead The agent takes the payment, and where an arrangement is in scope it applies the utility's own eligibility rules rather than guessing, confirms the terms aloud, and records what was agreed. #### Systems it reads - Customer information system - Payment provider - Billing - Case management #### What it can execute - Take a card or bank payment - Confirm an arrangement where eligibility allows - Record the agreed terms - Read back what was set up - Send confirmation #### Where it stops Hardship, disconnection and anything regulated goes to a person. Those conversations should involve someone who can exercise discretion. #### Common questions ##### Can an AI agent take a utility payment over the phone? Yes. The card capture is handled by the telephony provider's PCI flow, so the card number does not pass through the agent, and the result is written back to the account. ##### Can an AI agent set up a payment arrangement? Where eligibility is defined it can, applying the utility's own rules rather than improvising. Hardship, disconnection and regulated cases are deliberately routed to a person, because those need discretion the agent should not be exercising. ##### Can a caller talk their way past verification to set up an arrangement? No. A payment arrangement, or anything else that alters the customer's record, is locked behind verification, and the lock is held in state the conversation cannot talk its way past. A caller cannot persuade the agent that they were verified earlier, because whether they were is not something the conversation decides. ##### How many utility payments does an AI agent take in practice? At the City of Kingsport, nearly a third of all customer payments in the first month in production were completed through the agent. Most pay-by-phone calls there finished in under two minutes, and the agent also texts payment links while the caller is still on the line. ##### Can an AI agent help when an online payment has failed? Yes. A failed online payment usually produces a call within minutes, and it is a call with a deadline attached. The agent can identify what went wrong and take the payment in the same call. ##### Is a payment arrangement made by an AI agent recorded on the account? Yes. The agent confirms the terms aloud, records what was agreed against the account, reads back what was set up and sends a confirmation. The record is written to the utility's own system, not to a separate copy held by the agent. ### Service requests. URL: https://www.dialai.ca/use-cases/utility-service-request-ai The work that used to be a form, a queue and a callback. “I need someone to come and look at this.” The call #### Why it is harder than it looks Service requests are varied and each has its own routing, its own information requirements and its own urgency. Taking one properly means knowing which questions this particular request needs answered before it can be dispatched. #### What happens today A general queue, then a person who fills in a form while the customer waits, then a callback to confirm. Three touches for one request. #### What the agent does instead The agent identifies the request type, gathers exactly what that type requires, creates the case in the utility's own system, and tells the customer the reference and what happens next. #### Systems it reads - Customer information system - Case or work management - Service address data #### What it can execute - Classify the request - Collect the required detail - Create the case - Give the customer a reference - Confirm expected timescales #### Where it stops Anything urgent, anything involving a safety risk, and any request type outside the configured set goes to staff rather than being approximated into the nearest category that fits. #### Common questions ##### Can an AI agent create a utility service request? Yes. It classifies the request, collects the detail that type requires, creates the case in the utility's own system and gives the customer a reference. The point is that the case exists before the call ends, rather than a person transcribing it afterwards. ##### What information does an AI agent collect for a service request? Exactly what that request type needs before it can be dispatched. Each type has its own routing, information requirements and urgency, so the agent identifies the type first and then asks only the questions that type requires. ##### Does the customer get a reference number for their service request? Yes. Once the case is created in the utility's own system, the agent gives the customer the reference and tells them what happens next and the expected timescales. That replaces the callback that used to confirm a request after the fact. ##### What happens if a service request does not fit any category? It goes to staff. A request type outside the configured set is handed to a person rather than approximated into the nearest category that fits, because each type has its own routing and a case filed under the wrong one goes to the wrong place. ##### What happens if a service request is urgent or unsafe? It goes to a person. Anything urgent and anything involving a safety risk is handed over rather than logged for a routine visit, with what the caller has already said attached so they do not start again. ##### Can an AI agent text the customer a form during the call? Yes. At the City of Kingsport the agent texts payment links, forms and resources while the caller is still on the line, so they leave the call with what they need in hand. --- ## Integrations ### Grace and Genesys. URL: https://www.dialai.ca/integrations/genesys Answering in front of the contact centre, handing over inside it #### The platform Genesys is the contact centre platform: the queues, the routing, the agent desktop and the reporting a utility's customer service team works in every day. It is where a call goes once a person needs to take it. #### The integration Grace answers in front of it. The agent takes the call first, identifies the customer and handles what is in scope, and when a person is needed the call enters the Genesys queue with the context attached rather than arriving cold. #### What the agent can do Staff keep the platform they already work in, their routing and their reporting. What changes is what reaches them: the repetitive calls stop arriving, and the ones that do come with the caller already identified and the question already asked, so the CSR does not restart the conversation. Comparing IVR and Grace: a new era in customer support #### Common questions ##### Does Dial AI work with Genesys? Yes. Grace answers in front of the contact centre and hands a call into the Genesys queue when a person is needed, with the caller identified and the context attached. The team keeps the platform, the routing and the reporting it already uses. ##### Does using an AI agent mean replacing the contact centre platform? No. The agent sits in front of it. The platform keeps doing what it does for the calls that reach a person; the difference is how many of them do, and how much the CSR already knows when one arrives. ### Grace and Hansen Technologies. URL: https://www.dialai.ca/integrations/hansen The AI Virtual Agent in the Hansen Suite for Energy & Utilities #### The platform Hansen builds software for the energy, utilities and communications industries, including the customer information system a utility bills from. For a utility running Hansen CIS, that platform holds the account, the balance, the usage history and the service record. It is the thing a contact centre agent has open while they talk. #### The integration Hansen Technologies made a strategic investment in Dial AI and the two companies partnered to develop and distribute AI-powered customer engagement for utilities. The result is Hansen's AI Virtual Agent, part of the Hansen Suite for Energy & Utilities, built on Dial AI's technology and delivered as a SaaS solution on AWS. #### What the agent can do The agent handles customer queries across voice, email, SMS and messaging, reading the account from the system of record rather than from a copy of it. The City of Kingsport is the deployment with published figures: it replaced a customer system it had run for roughly twenty years, moved to Hansen CIS, and put the agent on the line in the same window. Hansen Technologies Makes Strategic Investment in Dial AI #### Common questions ##### Does Dial AI integrate with Hansen CIS? Yes. Hansen's AI Virtual Agent, part of the Hansen Suite for Energy & Utilities, is built on Dial AI's technology. It handles customer queries across voice, email, SMS and messaging for utilities running Hansen. ##### Is the Hansen AI Virtual Agent the same as Grace? It is Hansen's product, built on Dial AI's technology underneath. A utility buying it buys it from Hansen, as part of the Hansen Suite for Energy & Utilities. The underlying agent is the same one Dial AI deploys directly elsewhere. ##### Which utility has deployed AI customer service on Hansen CIS? The City of Kingsport, Tennessee. It replaced a customer system it had run for roughly twenty years with Hansen CIS and put the AI agent on its utility line in the same window. Across its first month, covering a water outage, it handled 20,379 calls and resolved 60.3% end to end. ### Grace and Harris. URL: https://www.dialai.ca/integrations/harris AI customer service across Harris's utility software businesses Cayenta enQuesta NorthStar CIS Infinity SilverBlaze Cognera #### The platform Harris runs several utility software businesses, each with its own customer information system and its own installed base: Cayenta, enQuesta, NorthStar, CIS Infinity, SilverBlaze and Cognera. Between them they bill a very large number of North American utility customers. #### The integration Dial AI and Harris partnered to bring AI customer service to utilities across those businesses. The hard part of an AI agent in this sector is not the conversation, it is knowing what the right answer actually is: a caller asking why their bill doubled is asking a question whose answer is buried in a rate structure, a budget billing plan, a proration after a move-in, an estimated read that got trued up, or an arrears arrangement that lapsed. #### What the agent can do Get the account lookup right and the explanation follows. The agent reads the account from whichever Harris platform the utility runs and answers from the real record, rather than giving a generic description of how billing works and leaving the customer to work out which part applies to them. Dial AI and Harris Partner to Bring AI Customer Service to Utilities #### Common questions ##### Which Harris utility platforms does Dial AI work with? The partnership covers Harris's utility software businesses: Cayenta, enQuesta, NorthStar, CIS Infinity, SilverBlaze and Cognera. ##### Can an AI agent explain a high bill on a Harris CIS? That is the case the partnership exists for. A doubled bill has a real cause in the record: a rate structure, a budget billing plan, a proration after a move-in, an estimated read that was later trued up, or a lapsed arrears arrangement. An agent that reads the account can name which one applies instead of describing billing in general. ### Grace and Paymentus. URL: https://www.dialai.ca/integrations/paymentus Taking the payment on the call, live at the City of Kingsport #### The platform Paymentus handles electronic bill payment for utilities: cards, bank payments and the channels a customer might pay through outside a phone call. For a utility running it, Paymentus is where a payment is actually processed and recorded. #### The integration Grace takes the payment on the line through Paymentus rather than reading out a number for the customer to call back. The card details are captured through the telephony provider's PCI flow, so the number is never handled by the agent, and the result is written back against the account. #### What the agent can do A billing call that ends in a payment ends on the same call. The customer does not get told what they owe and then transferred, or given a different number to ring. This is live at the City of Kingsport, where the system of record is Hansen CIS and payments run through Paymentus. City of Kingsport: Grace Goes Live With a Twenty-Year System Replacement #### Common questions ##### Can Grace take a payment through Paymentus? Yes, and it is live at the City of Kingsport, where the account is read from Hansen CIS and payments run through Paymentus. The payment happens on the call rather than the customer being transferred or given another number. ##### Does the AI agent handle the card number? No. Card capture runs through the telephony provider's PCI flow, so the card number does not pass through the agent. What comes back is the outcome, which is written against the account. ### Grace and SpryPoint. URL: https://www.dialai.ca/integrations/sprypoint The engine under SpryAssist #### The platform SpryPoint builds cloud-hosted utility software. That turns out to matter for an AI agent: a cloud platform can be reached and written to in the moment, which is the difference between an agent that can look something up and one that can finish the job. #### The integration Dial AI and SpryPoint partnered to bring AI customer service to utilities running SpryPoint. The product is SpryAssist, SpryPoint's AI agent for utility customer service, running on Dial AI's technology underneath. It lives inside the platform rather than beside it. #### What the agent can do An agent that can only read a help page deflects a question. One wired into the system of record finishes it: confirms the balance, explains the charge, takes the payment, starts the move-in, and writes the result back. Dial AI Powers SpryAssist, SpryPoint's New AI Agent #### Common questions ##### What is SpryAssist? SpryAssist is SpryPoint's AI agent for utility customer service. It runs on Dial AI's technology and lives inside the SpryPoint platform rather than alongside it, so it can complete a request against the system of record rather than only answering a question about it. ##### Why does cloud-hosted utility software matter for an AI agent? Because the agent has to act, not just read. A platform that can be queried and written to while the customer is still on the line lets the agent confirm a balance, explain a charge, take a payment or start a move-in and record the result. An agent that can only read a help page can deflect the question but cannot finish it. --- ## Industries ### Electric Utilities URL: https://www.dialai.ca/industries/electric-utilities Grace helps electric utilities deliver clear, human service at scale. When a customer asks why their electricity rates are so complex, she explains how usage tiers, time-of-day pricing, and local generation costs affect their bill, using the same knowledge their best agents rely on. Grace assists staff during calls and chats, ensuring every explanation is accurate, consistent, and easy to understand. Customers get clarity, agents get support, and utilities build trust through better conversations. Book a demo Explore the platform #### The calls that fill your queue. Electric utilities live and die by the storm day. Volume that sits flat for weeks can go up tenfold in an hour, and the calls that arrive are the ones customers are most anxious about. Each of these as a workflow 01 ##### Outage reporting and restoration times Grace confirms the caller's service address, checks the outage management system for an active event, logs a new report if there isn't one, and gives the current estimated restoration time, the single highest-volume call on a storm day. 02 ##### High-bill and rate explanations Time-of-use windows, tiered usage and seasonal rates are the hardest thing a CSR has to explain. Grace walks through the actual bill, compares it to prior periods, and explains what drove the change. 03 ##### Start, stop and transfer of service Move-ins and move-outs are routine, form-shaped work. Grace collects the service address, dates and identity verification, then writes the order into the CIS. 04 ##### Payment arrangements and disconnect notices Customers calling about a past-due balance need options, not hold music. Grace explains the balance, what a notice means, and the arrangements available under your policy. #### What makes electric support hard. - Storm and outage surges that no staffing model can absorb - Time-of-use and tiered rate structures that generate repeat explanation calls - Regulated response and reporting obligations - After-hours coverage for outages that don't respect business hours Integration ##### Answers from your system of record Grace reads from the CIS and OMS you already run, including Cayenta, Harris NorthStar and Advanced, so restoration times and balances come from the system of record rather than a static FAQ. How the platform connects #### Other industries Water Utilities Gas Utilities Municipal Telecommunications Financial Services ### Financial Services URL: https://www.dialai.ca/industries/financial-services Grace helps financial institutions deliver accurate, trustworthy support at scale. When a customer asks why a transaction was flagged or how interest was calculated, Grace explains policies and account details using verified data and staff-approved guidance. Grace assists advisors and service teams by suggesting clear, compliant responses and handling routine inquiries. Customers get fast, reliable answers, while staff focus on building deeper relationships and solving complex financial needs. Book a demo Explore the platform #### The calls that fill your queue. In financial services the constraint is not capability, it is provability. Every answer has to be traceable to a policy and a record, which is exactly how Grace is built. Each of these as a workflow 01 ##### Balance, transaction and statement questions Grace verifies identity to your standard, then answers from the account record, the highest-volume and most easily automated call type. 02 ##### Flagged and disputed transactions Grace explains why a transaction was held, what the review process is, and what the customer needs to do, then opens the dispute with the details captured. 03 ##### Interest, fees and policy explanations Grace explains how a charge was calculated using staff-approved wording, so the answer is the same on every call. 04 ##### Card replacement and account maintenance Grace handles address changes, card reissues and contact updates behind proper verification. #### What makes financial support hard. - Regulatory obligations for accurate, consistent disclosure - Identity verification on every account-specific interaction - Auditability: every answer traceable to a policy and a record - Fraud and dispute volume that arrives unpredictably Integration ##### Answers from your system of record Grace works against your core banking, card and CRM platforms, and every conversation is logged and reviewable, so what was said, and on what basis, is always answerable. How the platform connects #### Other industries Electric Utilities Water Utilities Gas Utilities Municipal Telecommunications ### Gas Utilities URL: https://www.dialai.ca/industries/gas-utilities Grace helps gas utilities deliver faster, clearer communication with customers. When someone asks why their bill spiked during a cold month, Grace explains how weather, usage, and rate structures contribute, using real data and staff-approved information. Grace assists agents during high-demand seasons, ensuring customers get accurate, consistent answers across every channel. The result is fewer escalations, shorter calls, and stronger customer trust. Book a demo Explore the platform #### The calls that fill your queue. Gas is the sector where knowing what NOT to handle matters most. Safety calls are carved out explicitly. Everything else is the seasonal billing load that buries a call centre every winter. Each of these as a workflow 01 ##### Suspected leak or gas odor: routed immediately Grace does not attempt to handle safety calls. Odor and suspected-leak intents are detected in the first seconds and transferred straight to your emergency line under your own escalation rules. This is a hard rule, not a confidence threshold. 02 ##### Cold-snap bill spikes A hard freeze produces a wave of identical calls. Grace explains how degree days, usage and rate structure combined to move the bill, comparing against the customer's own history. 03 ##### Budget billing and payment arrangements Grace explains how levelised billing works, what the customer's current plan is, and what arrangements are available when a balance is past due. 04 ##### Service and appointment scheduling Grace books, confirms and reschedules service line and appliance appointments, and confirms the window back to the customer. #### What makes gas support hard. - Weather-driven demand spikes concentrated into a few winter weeks - Safety-critical calls that must never be automated - Seasonal billing complexity that generates repeat explanation calls - Appointment coordination across field crews Integration ##### Answers from your system of record Grace reads from your CIS and billing platform, and hands off to your existing emergency routing untouched. Safety escalation paths stay exactly as your regulator approved them. How the platform connects #### Other industries Electric Utilities Water Utilities Municipal Telecommunications Financial Services ### Municipal URL: https://www.dialai.ca/industries/municipal Grace helps municipalities provide faster, more transparent service to residents. When someone asks why their property tax changed or how to report a streetlight outage, Grace delivers clear, accurate answers using verified information from city staff and systems. Grace supports frontline teams by guiding responses and handling routine inquiries, ensuring residents get consistent information across phone, web, and in-person channels. The result is better communication, reduced workload, and stronger public trust. Book a demo Explore the platform #### The calls that fill your queue. A municipal line is really a dozen departments behind one phone number. Most of the work is routing, getting the resident to the right desk with the right information, first time. Each of these as a workflow 01 ##### Property tax and utility billing Grace explains assessments, due dates, instalment plans and what changed since last year, pulling from the tax and billing systems rather than a script. 02 ##### Service requests and 311-style reports Streetlight outages, potholes, missed collections and noise complaints. Grace captures the location and details, files the request, and gives the resident a reference number. 03 ##### Waste, recycling and collection schedules The highest-volume, lowest-complexity call a city receives. Grace answers it instantly by address, including holiday schedule changes. 04 ##### Permits, licensing and department routing Grace explains what a resident needs, what it costs, and what to bring, then routes to the right department with the context already gathered. #### What makes municipal support hard. - One phone number fronting many departments with different systems - Seasonal spikes around tax deadlines and collection changes - Public accountability for consistent, accurate answers - Council and FOI expectations for a documented record Integration ##### Answers from your system of record Grace connects to your tax, utility billing and work-order systems so a resident's answer and their service request land in the same places your staff already work. How the platform connects #### Other industries Electric Utilities Water Utilities Gas Utilities Telecommunications Financial Services ### Telecommunications URL: https://www.dialai.ca/industries/telecommunications Grace helps telecom providers deliver consistent, high-quality support at scale. When a customer asks why their bill increased or their data speeds dropped, Grace explains plan details, usage patterns, and network conditions using verified information from internal systems. Grace assists agents during peak periods, keeping responses clear and consistent across chat, voice, and email. Customers get fast, accurate answers, and staff stay focused on solving complex issues that drive loyalty. Book a demo Explore the platform #### The calls that fill your queue. Telecom support is high-volume, high-churn and mostly repetitive. The same handful of questions arrive thousands of times a week across four different channels. Each of these as a workflow 01 ##### Bill increases and plan changes Promotional pricing rolling off is the single most common billing complaint. Grace explains exactly what changed and what plans are available now. 02 ##### Speed, connectivity and troubleshooting Grace runs the first-line diagnostic script consistently every time (line checks, equipment restarts, known-issue lookup) and escalates with the results attached rather than making the customer start over. 03 ##### Outage and service status Grace checks for known incidents affecting the customer's address and gives the current status, instead of queueing them behind a hold message. 04 ##### Moves, upgrades and installs Grace captures the details, checks serviceability and books the appointment window. #### What makes telecom support hard. - High call volume with a narrow band of repeat questions - Churn risk concentrated in billing and speed complaints - Multi-channel expectations: chat, voice, email and SMS - First-line troubleshooting that must be run consistently Integration ##### Answers from your system of record Grace connects to your billing, CRM and provisioning systems so account state, plan detail and incident status come from your platforms. How the platform connects #### Other industries Electric Utilities Water Utilities Gas Utilities Municipal Financial Services ### Water Utilities URL: https://www.dialai.ca/industries/water-utilities Grace helps water utilities deliver clear, confident service at scale. When a customer asks why their bill is higher this month, Grace explains how usage, seasonal rates, or leaks may be involved, using verified data and guidance from utility staff. Grace supports agents during high-demand periods, keeping every response accurate and consistent. Customers get answers they can trust, and utilities strengthen relationships through better conversations. Book a demo Explore the platform #### The calls that fill your queue. For water utilities, almost every hard conversation starts with a bill that doesn't look right, and the cause is usually invisible to the customer until someone explains it. Each of these as a workflow 01 ##### High-bill investigations and leak adjustments A silent toilet leak can triple a bill. Grace compares the current read to history, explains what consumption pattern suggests a leak, and walks the customer through the adjustment request your policy allows. 02 ##### Meter reads, re-reads and estimated bills Grace explains why a bill was estimated, what the current read is, and schedules a re-read when the customer disputes it. 03 ##### Seasonal and irrigation rate questions Summer irrigation moves customers into higher tiers. Grace explains the tier they landed in and what drove it, before it becomes a complaint. 04 ##### Service start, stop and transfer Grace collects addresses, dates and verification for move-ins and move-outs, then writes the order back into the CIS. #### What makes water support hard. - Seasonal consumption swings that drive predictable bill-shock waves - Leak adjustments that require policy judgement and clear explanation - Conservation and restriction messaging during drought - Small teams covering a full service territory Integration ##### Answers from your system of record Grace connects to your CIS and billing platform (Cayenta, Harris NorthStar, Advanced and others), so reads, balances and adjustment history come from the system of record. How the platform connects #### Other industries Electric Utilities Gas Utilities Municipal Telecommunications Financial Services --- ## Customer evidence ### How Kingsport answered 20,379 calls with zero hold time URL: https://www.dialai.ca/customers/kingsport The phones were flooded. Zero hold time. How Grace handled a month of heavy call volume on the City of Kingsport's utility line, including a water outage, with zero hold time and a customer service team that never fell behind. - Period: Measured over the first month in production - Scale: ~47,000 utility accounts - System: Deployed with Hansen Technologies CIS - 20,379: Calls handled. Including 2,611 outbound - 60.3%: Resolved end to end. ~80% excluding caller-requested transfers - 1:53: Average call duration. Most finished inside three minutes - 0:00: Average time on hold. Answered on the first ring, day or night - 67%: Callers auto-verified. Authenticated from the number they called on Measured over the first month in production · ~47,000 utility accounts · Deployed with Hansen Technologies CIS. Grace has been answering the line ever since. “This is a gamechanger for both the City of Kingsport and its citizens.” Floyd Bailey · Chief Information Officer, City of Kingsport Featuring Floyd Bailey, CIO, City of Kingsport Industry Municipal water, storm and wastewater Runtime 22 min Filmed by Hansen Technologies Context #### The phones are full, and the team is stretched thin. A municipal utility's phone line is the front door for tens of thousands of accounts. Billing questions, payments, autopay, outages: it all comes through the phone, and in a busy month the calls do not let up. Over the month measured here, the line took more than 20,000 calls, and a water outage drove a wave of them on one of the days. The legacy phone system was a touch-tone menu. It could route a call, but not answer a billing question or take a payment. Without something that could hold a real conversation, that volume lands straight in the hold queue. The question is not whether to automate the phone line. It is whether the automation can carry real volume on a busy day, not just the quiet ones. This is not unique to Kingsport. Across the sector, call volume is rising while experienced staff retire faster than utilities can replace them. Deployment #### Grace answers first, and hands off when it counts. Dial AI deployed Grace on the City's main utility line, integrated with Hansen CIS and replacing the legacy IVR. Grace is the first point of contact: she verifies the caller, pulls their account from Hansen in real time, and handles the request end to end. When a call needs a person, she hands it to a CSR with the full story already written up. Conversational voice and SMS layer. System of record in Hansen CIS, payments via Paymentus, live-agent pickup through Cisco Webex Contact Center. What Grace handles - Bill inquiries - Payments - Auto Pay enrollment - Payment arrangements - Identity verification - Contact info updates - Paperless / bill delivery - Payment history - Outage reporting - Live-agent handoff #### Why it carried the volume. ##### Instant, bilingual, 24/7 Answers on the first ring, day or night, in fluent English or Spanish. No phone tree and no hold music, ever. ##### Real-time CIS integration Pulls live account, balance and billing history from Hansen during the call. Grace works from the customer's real record, not a static script. ##### Caller ID verification Authenticates most callers automatically from the number they are calling on. In month one, 67% were verified with no manual steps. ##### Mid-call SMS Texts payment links, forms and resources while the caller is still on the line, so they leave the call with what they need in hand. ##### Context-preserving handoff On transfer, the CSR receives a written summary and full transcript, posted to the customer's account. The caller never repeats their account number or reason for calling. ##### Tuned live, no downtime Outage reporting was added during the water outage and refinements shipped through the week, all pushed to production without taking Grace off the line for a moment. ##### Answers many at once Grace can handle more than 30 calls at the same time, so a spike in volume never turns into a queue. ##### Empathetic by design Grace reads caller sentiment as the conversation unfolds and adjusts its tone, so responses feel aware and human, not scripted. Results · month one #### Where every inbound call ended up. 83% of the 17,705 inbound calls were either resolved by Grace or routed to a person exactly because the caller asked for one. 62% of all escalations were operator-requested, not cases Grace could not handle. Counting only calls where the caller did not ask for an operator, Grace resolved 79.5% end to end. Billing and payments together were about 58% of all calls. Nearly a third of all customer payments in the month were completed through Grace. The customer journey #### A faster, simpler call for every resident. For the person calling about a bill, the change was immediate. No queue to sit through, no portal to learn, no callback to wait on. Here is the same call, before and after Grace. Before With Grace Hold music, a callback, or a wait until morning. Answered on the first ring, day or night, in English or Spanish. Wait for a rep to look up and read the balance. Balance read back in seconds, payment link texted to the caller. Get transferred to pay, or call back to do it later. Pay by phone in the same call, most finished in under two minutes. Answer a string of security questions first. Verified automatically from the caller's number two times out of three. Repeat the account number and story to each person. If a specialist is needed, they already know who is calling and why. The average call ran 1:53, roughly half the time a live agent needs, and most finished in under three minutes. The result is a shorter call, fewer call-backs, and a line that never makes anyone wait. When the water outage hit, Grace took outage reports as the calls came in, so residents got answers in the moment instead of a busy signal. The takeaway #### Proof on a busy month. Kingsport did not test Grace on a quiet week. The phones were full, a water outage drove a wave of calls, and Grace carried it: more than 20,000 calls answered on the first ring, the routine volume handled end to end, and a team free to spend the month with the callers who needed them. For any utility weighing automation against a stretched call center and a retiring workforce, that is the test that matters. A reference conversation with the City of Kingsport team can be arranged on request. Let's talk about your utility Keep reading #### Read the full Kingsport case study Two details and the rest opens right here. ### How PSE&G handled a two-million-meter rollout with zero queue time URL: https://www.dialai.ca/news/pseg-ami-deployment By Dial AI Grace handled more than 10,000 calls during PSE&G's advanced metering rollout, booking an appointment for 65% of engaged callers, with zero queue time and support in Spanish. PSE&G set out to replace more than two million meters as part of their Advanced Metering Infrastructure programme. Most customers moved across without difficulty. The remainder, households that had gone more than two years without scheduling an exchange, are the hard part of any rollout, and the part that decides when it finishes. Grace was brought in to work that queue. #### From off-hours to peak Grace started on off-hours calls only. Queue times during business hours stayed high enough that PSE&G widened her remit to take between 20% and 80% of calls at peak, depending on load. That flexibility is the point: the volume in a meter exchange programme is not steady, and staffing to the peak is what makes it expensive. #### What the numbers looked like Over the first stretch of the deployment: - More than 10,000 calls handled, over 5,000 of them in a two-week period. - 900 calls in a single day at peak. - About 60% of calls fully automated: greeting to booked appointment, no human involved. - The other 40% went to a person or a scheduled callback. #### Booking rate, which is the metric that matters An exchange programme is not measured in calls answered. It is measured in appointments on the calendar. Grace booked an appointment on 35% of all calls she took. Adjusting for accounts that could not be booked at all, that is 65% of engaged callers, comfortably ahead of the benchmark the programme was measured against. She also books faster, in roughly half the handling time, and that comparison excludes queue time entirely. Grace’s queue is always zero, because there is no queue. #### Spanish Support for Spanish was added during the deployment and handled over 400 calls in its first week, at success rates comparable to English. A caller who asks for Spanish is switched immediately, mid-call, rather than transferred to a separate line and made to wait. #### The wider point A meter exchange campaign is an unusually clear test. The task is narrow, the outcome is binary, and there is a human benchmark sitting right next to it. It is a fair place to ask whether an AI agent is actually doing the work or simply absorbing calls, and on the measure that decides whether the programme finishes, it did better than the benchmark. #### More reading October 16, 2026 ##### Product updates: phone payments, callbacks, rollout tracking and machine apps October 15, 2026 ##### Lakeland Electric goes live with Cayla, Cayenta's AI agent August 5, 2026 ##### How TransAlta put AI on the queue nobody can hear ### Consolidated Utility District puts Grace on every inbound call URL: https://www.dialai.ca/news/cud-go-live By Dial AI · June 29, 2026 Grace is now answering the phones for Consolidated Utility District of Rutherford County, the water utility serving Murfreesboro and the county around it. Water is the service people only ring about once something has already gone wrong. Grace has gone live on voice for Consolidated Utility District of Rutherford County, in Murfreesboro, Tennessee. She now takes the district’s inbound calls. CUD supplies drinking water across Rutherford County to roughly 70,000 accounts, through more than 1,400 miles of distribution main. It also runs decentralised wastewater treatment for the parts of the county that city sewer does not reach. It is a water district rather than a combined utility, and that shapes the call mix more than people expect. #### A water utility’s calls arrive all at once An electric utility has a storm season. A water utility has a main break, and a main break does not announce itself a week out on a forecast. When pressure drops on a street, everyone on that street rings within the same twenty minutes. The same thing happens on a boil notice, and again on the bill that lands after a customer-side leak has run unnoticed for a month. Volume in water is not a steady load with a seasonal curve on it. It is flat, and then it is vertical. That is an awkward shape to staff against. Hiring for the vertical means paying for idle capacity on every ordinary day, and staffing for the ordinary day means the queue collapses precisely when the thing people are calling about is whether their water is safe to drink. #### What she handles Grace answers questions about balances, due dates and consumption, takes payments, and starts and stops service, reading from the district’s own system of record rather than a copy of it. A high-bill call gets the usage history that explains it. Anything outside the flow’s scope goes to staff with the context already gathered, rather than as a cold transfer. The point is not that every call is automated. It is that the ordinary ones stop competing for the same people as the urgent ones. #### Why the phone Water skews heavily to voice, and it stays that way. A customer with no water is not opening a chat window, and a customer who has just been told to boil it wants an answer in the next minute, not the next business day. Delivered with NorthStar, in the Harris channel. #### More reading October 16, 2026 ##### Product updates: phone payments, callbacks, rollout tracking and machine apps October 15, 2026 ##### Lakeland Electric goes live with Cayla, Cayenta's AI agent August 5, 2026 ##### How TransAlta put AI on the queue nobody can hear ### How TransAlta put AI on the queue nobody can hear URL: https://www.dialai.ca/news/ai-email-goes-live By Dial AI · August 5, 2026 Grace is now handling inbound email for TransAlta, running the same agent and the same guardrails we put on voice. Email is the channel where a backlog is invisible, which is exactly why it needed the help. Grace has gone live on email for TransAlta, a Canadian energy company. Email joins voice, chat and SMS as a channel she handles, and it runs on the same agent we put on the phones elsewhere: same integrations, same guardrails, same escalation rules. #### Why email, and why it is harder than it looks Voice gets the attention because the pain is audible. A queue has hold music, and everyone in the building knows when it is long. Email has no hold music. A backlog of 400 unanswered messages makes no sound at all. Nobody hears it, the customer just doesn’t hear back, and the only signal that anything is wrong is the second email, usually a more frustrated one, arriving three days later. The channel where service quietly decays is the one nobody is watching. It is also the channel with the least slack in it. A contact centre can flex staff onto phones during a storm or a billing cycle. Nobody flexes onto the inbox, so the inbox absorbs the delay. That is why email backlogs form after exactly the events that generate them, and why they persist for weeks after the calls have stopped. And email is unforgiving in a way a call is not. A phone answer is spoken once and gone. A written answer is a durable artefact. It gets forwarded, quoted back, and attached to complaints. Getting it wrong in writing is worse than getting it wrong out loud, which is the real reason utilities have been slow to automate here. #### What we are seeing The deployment is deliberately conservative, and the interesting part is what Grace declines to touch. A routing agent reads first. Every inbound message is classified before anything is drafted. If the request falls outside the flow’s scope, it goes to staff untouched, no partial answer, no attempt to be helpful at the edges. Restraint is the feature. Threads that pre-date go-live stay with humans. An email conversation already in progress when Grace switched on is left alone. Picking up someone else’s half-finished thread is exactly where an agent produces a confident answer to a question it has not actually read. Staff can turn her off per conversation. Not as a global kill switch buried in an admin panel, but per thread, in the moment, by the person handling it. Escalation-risk detection shipped alongside it. The system watches for the signals that a conversation is deteriorating and surfaces it before the customer escalates it for you. #### Where this goes The point of running one agent across voice, chat, email and SMS is not channel coverage for its own sake. It is that the customer who called on Tuesday and emails on Thursday is the same person with the same problem, and should not have to explain it twice. Shared context across channels is the part that is hard to build and the part that customers actually notice. Email is where that matters most, because email is where the history already lives. #### More reading October 16, 2026 ##### Product updates: phone payments, callbacks, rollout tracking and machine apps October 15, 2026 ##### Lakeland Electric goes live with Cayla, Cayenta's AI agent June 29, 2026 ##### Consolidated Utility District puts Grace on every inbound call --- ## Comparisons ### AI agent or IVR? URL: https://www.dialai.ca/compare/ai-agent-vs-ivr Almost every contact centre has an IVR in front of it, and almost every conversation starts with the same question: does an AI agent replace it, or sit alongside it? #### What an IVR does well IVR has been the backbone of telephony self-service for decades because it is genuinely good at a specific set of things. It routes reliably. It is predictable, cheap to run, and completely deterministic: the same input produces the same result every time, which matters more in a regulated sector than it does elsewhere. For a caller who knows exactly what they want and can find it in the menu, a well-built IVR is faster than a conversation. It is designed around selection: the caller picks from a set of known options. That is its strength, and it is also the boundary of what it can do. #### What an AI agent does differently An AI agent is built for the calls that do not fit a menu option. It holds a multi-turn conversation, asking clarifying questions, keeping track of what has already been said, and adapting as the caller's actual problem becomes clear. That matters because callers rarely open with the real question. "I want to talk to someone about my bill" turns out to be a payment arrangement, a move-out final bill, or a charge from a rate change, and which one it is only emerges a few exchanges in. Underneath that, it is connected to the systems that hold the answers, so it can finish the job rather than describe it: retrieve the balance, explain the charge, take the payment, book the appointment, write the result back. #### Side by side. IVR compared with an AI agent - Criteria | IVR | AI agent - What it is built around | Selection. The caller picks from a set of known options. | Conversation. The caller says what they want in their own words. - Turn handling | One input, one branch. No memory of what came before. | Multi-turn, holding context across the call and adapting as the real problem emerges. - Determinism | Total. The same input produces the same result every time. | Bounded by guardrails and approved sources rather than by a fixed tree. - System of record | Usually read-only lookups, if connected at all. | Reads and writes: balances, payments, service orders, case notes. - Concurrency | Limited by the lines and the agents behind it. | No queue. The hundredth simultaneous caller is answered like the first. - Language | A separate menu tree per language. | Switched mid-call when the caller asks. - Where it is strongest | Emergency routing and simple, well-understood self-service. | Anything needing context, lookup, or more than one exchange. - Where it struggles | Anything that does not fit a menu option. | Paths where determinism is the requirement rather than a constraint. #### Running both. Most deployments are not a replacement decision. The common shape is: The IVR keeps the front door. Emergency routing especially. A gas odour or a downed line should hit a deterministic path to the right team, and that is a job for a system that does exactly the same thing every time. Straightforward self-service that already works well can stay too. The agent takes the conversations. Anything that needs context, lookup, or more than one exchange. In practice this is most of what would otherwise have ended at "please hold for the next available representative". Handoffs carry context. The point of integrating the two is that a caller does not repeat themselves at the boundary. Whatever the IVR collected goes with them, and whatever the agent has gathered goes to the representative when a call escalates. Where exactly the split falls is a per-deployment decision, and it usually moves after go-live. The pattern is that the IVR keeps the paths where determinism is the point, and gives up the paths where it was only ever a menu standing between a customer and an answer. What it looked like in practice The City of Kingsport replaced its legacy IVR and put Grace on the utility line. In one month she answered20,379 calls on the first ring, resolved 60.3% end to end, and averaged 1:53 against a 4:00 agent baseline, with zero hold time. Read the Kingsport case study → How the migration itself runs What voice specifically demands #### Common questions ##### Does an AI agent replace an IVR? Usually not entirely. Most deployments keep the IVR for emergency routing and simple self-service, where doing exactly the same thing every time is the requirement, and give the AI agent the conversations that need context, a lookup, or more than one exchange. ##### What is the difference between an IVR and an AI agent? An IVR is built around selection: the caller picks from known options and each input leads to a fixed branch. An AI agent is built around conversation: it holds context across turns, asks clarifying questions, and can complete work in the system of record rather than only routing the call. ##### Can an AI agent and an IVR run together? Yes, and that is the common shape. The IVR keeps the front door and the deterministic paths, the AI agent takes the conversations, and context passes between them so a caller does not repeat themselves at the boundary. ##### Is an IVR still worth keeping? For some paths, yes. Emergency routing is the clearest case: a gas odour or a downed line should follow a deterministic path to the right team, which is exactly what an IVR is good at. The paths worth moving are the ones where the menu was only ever standing between a customer and an answer. ##### What happens to callers who do not fit a menu option? In an IVR-only contact centre they end up holding for an agent, because the menu has nothing for them. That group is usually most of what an AI agent takes on, since callers rarely open with their real question. ##### Should we put an AI agent in front of our IVR or replace it outright? Most utilities start in front of it and end up replacing it. In front, the existing platform stays as it is: the agent takes the calls it can finish end to end and the rest pass through as they do today, so you learn what it can carry on your own call mix. Deciding up front is not required, and starting with the safer option costs nothing later. ##### How do we decide which calls stay on the IVR? Start from your call mix rather than the technology. Take the top ten reasons people ring and mark each one: does this caller know what they want, or are they trying to describe a problem? The first group suits the IVR, the second is where an AI agent earns its place, and most utilities are surprised by how much of their volume sits in the second. ##### Which handles an outage surge better, an IVR or an AI agent? An AI agent, for the reporting and status calls that make up most outage volume. An IVR's capacity is limited by the lines and the agents behind it, so a surge becomes a queue, while an agent answers the hundredth simultaneous caller like the first. Emergency calls about a gas odour or a downed line should still follow the deterministic path. ##### How do IVRs and AI agents differ on language support? An IVR usually needs a separate menu tree for each language, while an AI agent switches language mid-call when the caller asks. Dial AI supports 45+ languages. During PSE&G's advanced metering rollout, Spanish was added part way through and handled over 400 calls in its first week, at success rates comparable to English. ### Choosing an AI agent for a utility. URL: https://www.dialai.ca/compare/best-ai-agent-for-utilities Utility contact centres differ from general customer service in ways that narrow the field quickly. Here is what to compare on, and where the general-purpose platforms sit. Most customer service AI is built for a support inbox: a customer asks a question, the agent answers it from published material, and the contact ends. A utility contact centre is not shaped like that. The work is account-specific and transactional. A high-bill investigation needs the actual usage history. A payment arrangement needs the real balance and has to be recorded. A move-out has to end in a service order. Most of it arrives by phone, from people who are not browsing. And when a storm takes out a feeder, the volume does not ramp, it arrives all at once. Those differences are what the criteria below are for. What to compare on #### Eight criteria that matter in a utility. ##### System of record integration Reads balances, usage history, service orders and outage state from the billing and customer systems you already run, and writes results back, rather than answering from a static FAQ. ##### Outage and storm surge Absorbs the volume spike when something breaks. Storm traffic is not a gradual ramp, and capacity that has to be planned for is capacity that is not there when it is needed. ##### Billing conversations High-bill investigations, payment arrangements, leak adjustments and final bills are multi-step, emotional and account-specific, which is most of what fills a utility queue. ##### Service orders Move-ins, move-outs and transfers are form-shaped work that has to end in a record, not a promise to call back. ##### Emergency routing untouched A gas odour or a downed line must follow the deterministic path a regulator approved. An agent that cannot leave that path alone is not deployable. ##### Bilingual service English and Spanish at equal quality, switched mid-call, rather than a separate menu tree that halves the service for half the callers. ##### Auditability Every conversation reviewable, every answer traceable to an approved source, every action attributable. Regulated service is judged after the fact, not during. ##### Vendor ecosystem fit Whether the agent is already integrated with the platforms utilities actually run, or whether the integration is a project you fund. Side by side #### Against the general-purpose platforms. The tools below are the ones most often shortlisted alongside us. They are horizontal customer service platforms rather than utility products, which is the point of the comparison. AI agents compared on utility-specific criteria - Criteria | Dial AI | Zendesk AI | Ada | Salesforce Einstein - Built for | Utilities and regulated industries, voice first | Teams standardized on Zendesk | Chat-first digital support | Organizations embedded in Salesforce - Billing and customer system integration | Direct to the billing and customer systems utilities run, reads and writes | Via marketplace integrations | Common CRM and helpdesk tools | Native within Salesforce, other systems via integration - Utility vendor partnerships | Hansen, Harris and SpryPoint, with Grace available inside their suites | General marketplace | General integrations | Salesforce partner ecosystem - Voice and IVR depth | Native IVR and telephony orchestration | Limited, typically via third-party tools | Limited IVR capabilities | Available through Salesforce voice products - Outage and surge handling | No queue; concurrency is not capacity-planned | Bounded by the agent pool behind it | Chat concurrency, voice via third parties | Bounded by licensed capacity - Regulated readiness | Designed for regulated environments, SOC 2 Type II | Varies by configuration | Limited regulated-industry focus | Enterprise-grade, compliance dependent on setup - Published utility deployments | City of Kingsport, PSE&G, Philadelphia Water and others | Not utility-specific | Not utility-specific | Not utility-specific Competitor capabilities change. This table reflects our understanding at the time of writing and is not a statement about any vendor's current product. In production On the City of Kingsport's utility line, Grace answered20,379 calls in one month on the first ring, resolved 60.3% end to end and averaged 1:53 against a 4:00 agent baseline, through a customer system cutover and a water outage. Read the Kingsport case study → See the sector pages → #### Common questions ##### What is the best AI agent for utilities? The right answer depends on how much of your volume is voice, which billing and customer systems you run, and what your regulator expects to be able to audit. Utilities differ from general customer service in three ways that narrow the field: most contacts arrive by phone, the answers live in a system of record rather than a knowledge base, and volume spikes when something breaks. ##### Why do utilities need a different AI agent from other industries? Because the work is account-specific and transactional rather than informational. A high-bill investigation, a payment arrangement or a move-out needs a real balance, a real usage history and a record written back at the end. A chat agent answering from published content cannot finish any of those. ##### Can an AI agent handle outage calls? It can handle the reporting and status side, which is what most outage volume is. Emergency routing should stay on the deterministic path a regulator approved: a gas odour or a downed line goes to the right team the same way every time, and an AI agent should leave that alone rather than sit in front of it. ##### Does an AI agent replace a utility's billing or customer system? No. The system of record stays where it is. The agent sits in front of it, reading and writing through integrations, which is why a deployment does not involve migrating customer data. ##### How long does a utility deployment take? It depends on how many systems the agent has to talk to and how much of the call flow is in scope on day one. The City of Kingsport went live on a new customer system and a new AI agent in the same week, which is at the fast end because both projects were run together. ##### What should a utility ask an AI agent vendor to demonstrate? Ask them to decide and execute on your own data. Understanding the caller, authenticating, retrieving the record, documenting and escalating are table stakes that most vendors do. Deciding what this caller is permitted under your policy, then taking the payment or starting the service with write access to systems that bill real people, is where the category divides. ##### Is Dial AI SOC 2 compliant, and where is the data hosted? Yes. Dial AI is SOC 2 Type II and is hosted on Amazon Web Services in US and Canadian regions, with Canadian data residency available; it does not use AWS GovCloud. The agent reads and writes through the utility's own systems rather than keeping a shadow copy of customer data, and each organisation's data is isolated at the tenant level. ##### Which utility billing and customer systems does Dial AI work with? Hansen, Harris and SpryPoint are partners, with Grace available inside their suites. The City of Kingsport runs Grace on Hansen CIS, reading the live account during each call. Where a platform is one Dial AI already integrates with, much of the integration work is already built; otherwise connecting the system of record is the main piece of deployment work. ##### How is an AI agent for utilities priced? Pricing scales with volume. What you run, how much of it you want handled, and the volume behind it are what shape a quote, and the pricing page has a calculator for working out the saving on your own call volume. ### Choosing a customer service AI agent. URL: https://www.dialai.ca/compare/best-customer-service-ai-agent What separates an AI agent from a chatbot, the ten criteria worth comparing on, and where Dial AI fits against Zendesk AI, Ada and Salesforce Einstein. A customer service AI agent is an AI-powered system designed to handle customer inquiries across channels like voice, chat and email, while handing off to human agents when needed. Unlike basic chatbots, modern AI agents can understand intent, follow complex workflows, integrate with CRM and contact centre systems, and operate in real-world, high-volume support environments. They are used by support teams to reduce wait times, automate repetitive requests, support agents with real-time assistance and keep service quality consistent, especially in enterprise and regulated industries where accuracy, security and reliability matter. What to compare on #### What makes the best customer service AI agent? The best one depends on your channels, your compliance needs and how deeply the system integrates with your contact centre and CRM. These are the criteria enterprise and contact centre teams typically evaluate. ##### Channel coverage Works across voice/IVR, chat, email and SMS, not just web chat, so customers can get help in the channel they actually use. ##### Real workflow automation Can complete multi-step tasks (authenticate a customer, update an account, create a ticket, schedule a callback) rather than only answering FAQs. ##### Reliable human handoff Routes to the right team with context, transcript and customer details, without making the customer repeat themselves. ##### Integrations that matter Connects to your CRM, ticketing and helpdesk, knowledge base, telephony and CCaaS, and internal tools. ##### Knowledge quality and controls Uses approved knowledge sources, supports content governance, and reduces hallucinations with grounding and guardrails. ##### Performance in real environments Handles high volume with strong uptime, low latency, and monitoring and alerting. ##### Security and compliance Supports enterprise security requirements (data handling, access controls, audit logs) and is suitable for regulated industries where needed. ##### Customization and orchestration Supports custom intents, routing rules and business logic, without requiring a full rebuild for every change. ##### Analytics and reporting Provides containment and deflection, CSAT impact, top intents, failure reasons, and agent performance insights. ##### Cost model and ROI clarity Pricing aligns with your usage and makes the return easy to understand. Side by side #### Top customer service AI agents compared. The tools below are among the most commonly evaluated for enterprise and contact centre support. Competitor capabilities change. This table reflects our understanding at the time of writing and is not a statement about any vendor's current product. #### When Dial AI is the best fit. A strong fit for organizations with complex support environments where voice, compliance and workflow orchestration are critical. - Voice-first and IVR-heavy support. Ideal for organizations where a large share of customer interactions happen over the phone rather than chat alone. - Regulated industries. Well suited to utilities, public sector and other regulated environments that require strong data handling, auditability and control. - Complex workflows. Best when customer requests involve multi-step processes such as authentication, account updates, service requests or case creation. - Human and AI collaboration. Works well when AI needs to assist agents and hand off conversations with full context, rather than fully replacing humans. - Custom orchestration needs. A good fit for teams that need flexibility in routing, intents and business logic beyond out-of-the-box chatbot flows. In production On the City of Kingsport's utility line, Grace answered20,379 calls in a month on the first ring, resolved 60.3% end to end and averaged a 1:53 call against a 4:00 agent baseline, with zero hold time. Read the Kingsport case study → #### When it may not be. Dial AI may be more than some teams need when support requirements are simple, limited to chat, or tightly constrained by a single platform ecosystem. - Very small teams with simple needs. Teams looking for a lightweight, plug-and-play chatbot for basic FAQs may prefer simpler tools. - Chat-only use cases. Organizations that do not support voice or IVR and only need web chat automation may not need Dial AI's full capabilities. - Highly locked-in platforms. Teams that require all AI functionality to live entirely within a single CRM ecosystem with minimal external orchestration may prefer native options. ##### What "complex workflows" actually look like Many requests need more than a single intent. A production voice workflow might involve: - Authenticating the caller using account information or phone verification - Retrieving account details from a CRM or billing system - Identifying the reason for the call, such as a service change, outage or payment issue - Executing a multi-step action such as scheduling a service date or updating account preferences - Creating or updating a case in a ticketing system - Confirming the outcome by voice, and optionally sending an SMS or email confirmation Platforms vary significantly in their ability to orchestrate these steps end to end, which matters most in voice-first environments and regulated industries. #### Questions about customer service AI agents. ##### What is the best customer service AI agent? The best customer service AI agent depends on an organization's support channels, operational complexity and compliance requirements. Some tools focus on chat automation, while others support voice, IVR, workflow orchestration and human handoff for more complex environments. ##### How is a customer service AI agent different from a chatbot? A chatbot typically answers predefined questions or automates simple conversations. A customer service AI agent can handle multi-step workflows, integrate with backend systems, support voice and IVR, and transfer conversations to human agents with full context. ##### Is Dial AI better than Zendesk AI or Ada? Dial AI, Zendesk AI and Ada are designed for different use cases. Dial AI is often chosen for voice-first, regulated or workflow-heavy environments, while Zendesk AI and Ada are commonly used for chat-focused automation within their respective platforms. ##### Can customer service AI agents handle voice and IVR support? Some customer service AI agents support voice and IVR natively, while others focus primarily on chat and messaging. Organizations with call-heavy support operations typically evaluate AI agents with built-in telephony and IVR capabilities. ##### Are customer service AI agents suitable for regulated industries? Yes, some customer service AI agents are designed for regulated environments and include features such as auditability, controlled knowledge sources and secure integrations. Suitability depends on the platform's architecture and deployment configuration. ##### How do AI agents hand off conversations to human agents? Modern AI agents can transfer conversations to human agents along with relevant context, such as conversation history, customer data and intent. This reduces repetition and helps agents resolve issues more efficiently. ##### What channels can customer service AI agents support? Depending on the platform, customer service AI agents may support voice, IVR, web chat, SMS, email and messaging apps. Channel support varies and is an important factor when comparing tools. ##### How are customer service AI agents typically priced? Pricing models vary by platform and may be based on usage, number of conversations, resolutions, seats or deployment complexity. Many enterprise-focused AI agents use custom pricing aligned with operational needs. See it on your own call flow --- ## Reference ### Contact centre metrics, defined. URL: https://www.dialai.ca/learn/contact-centre-metrics Nine measures that come up in every contact centre conversation, what each actually counts, and the ways each is commonly reported to look better than it is. #### Containment Self-service rate The share of contacts resolved without a person. A call is contained when the customer's request is finished end to end by the system they reached, rather than transferred, escalated or abandoned. How it is measured. Contained contacts ÷ total contacts. Watch what the denominator includes: counting only calls that entered the automated flow flatters the number against counting everything that reached the line. #### Deflection The share of contacts kept out of the queue at all, usually by answering the question in another channel before the customer calls. Often used loosely as a synonym for containment, which it is not. How it is measured. Contacts avoided ÷ contacts that would otherwise have arrived. Genuinely hard to measure honestly, because the counterfactual is an estimate. #### Average handle time AHT How long an agent spends on a contact from answer to finished, including talk time, hold time and the after-call work of writing it up. How it is measured. (Talk + hold + after-call work) ÷ contacts handled. A falling AHT is only good news if resolution holds: the fastest way to cut it is to transfer more. #### First contact resolution FCR The share of issues settled on the first attempt, with no callback, no second contact and no follow-up ticket. How it is measured. Issues resolved first time ÷ total issues, usually over a window long enough to catch a repeat call, often seven days. #### Average speed of answer ASA How long a customer waits in the queue before a person or a system picks up. Distinct from AHT, which starts once the contact is answered. How it is measured. Total wait time ÷ contacts answered. It excludes abandoned contacts, which is why a bad day can show a healthy ASA. #### Abandonment rate The share of customers who hang up before being answered. The metric that most directly reflects what waiting feels like, and the one most likely to be excluded from a queue report. How it is measured. Abandoned contacts ÷ total contacts offered. A short-abandon threshold, discounting hang-ups in the first few seconds, is common and should be stated. #### Transfer rate The share of contacts passed to another person or team. High transfer is a routing problem, a training problem, or a sign that the first line cannot reach the system holding the answer. How it is measured. Transferred contacts ÷ contacts handled. Worth splitting into customer-requested and system-driven transfers, which have very different causes. #### Occupancy The share of an agent's logged-in time spent on contacts rather than waiting. A capacity measure rather than a service one. How it is measured. Handle time ÷ logged-in time. Sustained occupancy above roughly 85% is associated with burnout and attrition, so it is a cost signal in both directions. #### Customer satisfaction CSAT A direct post-contact rating from the customer, usually a single question on a short scale. How it is measured. Satisfied responses ÷ responses received. Response rates are low and skew to the extremes, so it reads better as a trend than as a number. What these look like in production On the City of Kingsport's utility line, Grace resolved60.3% of 20,379 calls end to end, averaged 1:53 against a 4:00 agent baseline, and held speed of answer at zero: every call answered on the first ring, so nothing was abandoned in a queue. Read the Kingsport case study → #### Common questions ##### What is call containment? Containment is the share of contacts resolved without a person: the customer's request is finished end to end by the system they reached, rather than transferred, escalated or abandoned. It is calculated as contained contacts divided by total contacts, and the definition of the denominator matters, since counting only calls that entered the automated flow flatters the result. ##### What is the difference between containment and deflection? Containment is about contacts that arrive and are resolved without a person. Deflection is about contacts that never arrive, because the question was answered somewhere else first. They are often used interchangeably, but deflection depends on estimating a counterfactual and containment does not. ##### What is a good average handle time? There is no universal figure, because handle time depends on the mix of work. It is also easy to improve for the wrong reason: the fastest way to cut handle time is to transfer more calls, so it should always be read alongside first contact resolution and transfer rate. ##### Is average speed of answer the same as average handle time? No. Average speed of answer measures how long a customer waits before being answered. Average handle time measures how long the contact takes once it has been answered. A contact centre can have a healthy speed of answer and a poor handle time, or the reverse. ##### Why can abandonment rate look better than the experience feels? Because it is often reported with a short-abandon threshold that discounts hang-ups in the first few seconds, and because average speed of answer excludes abandoned contacts entirely. A queue that many customers give up on can report a healthy speed of answer for the ones who stayed. ### What does a call actually cost? URL: https://www.dialai.ca/learn/what-a-contact-centre-call-costs It is the first number anyone needs and the hardest to get a straight answer on. Here is what goes into it, how to work out your own, and why the published benchmarks disagree so wildly. Search for a cost per call and you will find figures anywhere from about a dollar to well over twenty. They are not contradicting each other so much as counting different things. A number built from wages alone and a number that includes supervision, systems, attrition and the time agents are paid to be available describe the same operation and differ several times over. Which is why the useful exercise is not finding a benchmark. It is working out your own, from figures you already have. #### The short method. Build it from the top down rather than the bottom up. Take everything the contact centre costs over a period, then divide by the contacts it handled in that period. cost per call = total cost of the agent pool ÷ contacts handled Top-down is more reliable than adding up per-call components, because it captures idle time without you having to model it. Agents are paid for the shift, not for the minutes they are talking, and any figure that misses that is too low. What goes in #### Seven things the total has to include. ##### Wages Base pay for the time the contact actually occupies, which is talk time plus hold plus the after-call work of writing it up. ##### Benefits and payroll costs Employer contributions, insurance, paid leave. Typically adds a quarter to a third on top of base pay, and it is the component most often left out. ##### Supervision and support Team leads, quality assurance, workforce management and trainers, divided across the agents they support. ##### Recruitment and training Hiring, onboarding and the ramp before a new agent is productive. In a contact centre with high attrition this is a recurring cost, not a one-off. ##### Systems and telephony Per-seat licences for the desktop, the CCaaS platform, the knowledge base, plus carriage. ##### Facilities and overhead Space, equipment and IT support, or the remote-work equivalent. ##### Idle time Agents are paid for the whole shift, not the handled minutes. Dividing by occupancy is what turns a cost per talking-minute into a cost per call. #### Cost per call is the wrong target. The fastest way to reduce it is to transfer more calls, and that makes the operation worse. A transferred call is handled twice. A repeat call is handled again next week. Both look like a shorter average handle time and neither is a saving. The number worth tracking is cost per resolved issue: the same total, divided by the issues that actually ended. It moves in the right direction only when work genuinely gets finished, which is also the only thing a customer experiences. The figure we use Our savings calculator defaults to $6per agent-handled call. That is a starting point for a conversation, not a benchmark we are asserting, and it is the first thing to replace with your own number. For reference on the other side of the sum: on the City of Kingsport's line, an agent call averaged 4:00 and Grace averaged 1:53. Try it on your own volume → The metrics, defined → #### Common questions ##### How do you calculate the cost per call in a contact centre? Take the fully loaded cost of the agent pool over a period, including wages, employer costs, supervision, recruitment and training, systems, telephony and facilities, then divide by the number of contacts handled in that period. Dividing total cost by total contacts is more reliable than building the figure up per call, because it captures idle time automatically. ##### What should be included in the cost of a call? Everything it takes to have someone able to answer it: wages and employer costs, supervision and quality, recruitment and training, desktop and telephony licences, and facilities. The components most often left out are employer costs on top of base pay, the recurring cost of attrition, and the idle time agents are paid for between contacts. ##### Why do published cost-per-call figures vary so much? Because they are counting different things. A figure built from wages alone and a figure that includes supervision, systems, attrition and idle time can differ by several times over for the same operation. Any benchmark is only comparable if the components behind it are stated. ##### Does a shorter call always cost less? Only if the work still gets finished. Handle time falls quickly when more calls are transferred, but a transferred call is handled twice and a repeat call is handled again later, so cost per resolved issue can rise while cost per call falls. ##### How does automation change the cost per call? It changes the mix rather than the unit cost. Contacts finished without a person carry the automation's cost instead of an agent's, while the ones that still reach a person cost what they did before. The saving is in the share moved, not in the calls that still need someone. --- ## Safety ### What Sets Grace Apart as a Safe AI Solution URL: https://www.dialai.ca/news/what-sets-grace-apart-as-a-safe-ai-solution By Grady Palfrey · December 8, 2024 Safety in a utility contact centre is specific: don't invent answers, don't act without verifying who is calling, don't quietly get worse over time, and make every conversation reviewable afterwards. Here is how Grace is built to do that. “Safe AI” is a vague phrase, so it is worth being concrete about what it has to mean when an agent is answering a utility’s phone. It means the agent does not invent an answer when it does not know one. It means nothing consequential happens to an account before the caller is verified. It means one organisation’s data is unreachable from another’s. And it means that if someone asks six months later why a particular answer was given, there is a record. Here is how each of those is built. #### Fine-tuned for function calling and instruction following Grace runs on models fine-tuned specifically for calling functions correctly and following instructions precisely, rather than for open-ended conversation. That focus is what reduces hallucination in practice. An agent asked for a balance should call the balance function and read out what comes back, not produce a plausible-sounding number. Training against that specific behaviour is more effective than instructing a general-purpose model to be careful. #### Guardrails and adversarial testing Grace stays inside the scope she was configured for, and we test that deliberately. Prompt-injection and jailbreak attempts are part of our own testing, so the failure modes get found by us rather than discovered live. Off-scope requests route to a person rather than being attempted. Declining to answer is the correct outcome far more often than it is treated as one. #### A review agent over every conversation Conversations are assessed automatically after the fact and flagged when something looks wrong, a bad answer, a missed escalation, a caller getting frustrated. This matters because the failure mode of a deployed agent is not a dramatic collapse; it is a slow accumulation of slightly wrong answers nobody is watching. Flagged conversations go to a human, corrections are recorded, and those corrections feed back into the configuration. #### Working inside your systems, not beside them Grace reads from and writes to your CIS and billing platform directly, and holds no shadow copy of your customer data. That is a safety property as much as a functional one: there is one source of truth, the answer a caller gets matches what a CSR sees on screen, and there is no second database to secure, synchronise or explain to an auditor. #### Tenant isolation Data is filtered at the tenant level, so one organisation’s information is not reachable from another’s. That isolation is enforced by the platform rather than left to configuration, which is what allows organisations to meet their security requirements without giving anything up functionally. Deployment is cloud-agnostic, so this fits the infrastructure and data-residency constraints a utility already operates under. #### Verification is enforced, not requested Sensitive actions, a payment arrangement, a change to an account, anything that alters a customer’s record, are locked behind verification, and the lock is held in state that the conversation cannot talk its way past. A caller cannot persuade Grace that they were verified earlier, because whether they were is not something the conversation decides. This is the mechanism we would point a security reviewer at first, because account takeover through a customer service channel is a real attack rather than a theoretical one, and “the agent was instructed to verify people” is not a control. #### Continuous review Automated review, manual review and success metrics all feed the same loop: find where the agent got it wrong, correct it explicitly, verify the correction held. The intent is an agent that measurably improves against your own calls over time, with each change traceable to the conversation that prompted it. #### The short version Safety here is not one feature. It is the fine-tuning, the guardrails, the review agent, working inside your systems, tenant isolation, enforced verification, and a review loop that closes. If your security team wants to go through any of these properly, that is a conversation we are always happy to have. It is usually the most productive meeting in a deployment. #### More reading October 16, 2026 ##### Product updates: phone payments, callbacks, rollout tracking and machine apps October 15, 2026 ##### Lakeland Electric goes live with Cayla, Cayenta's AI agent August 5, 2026 ##### How TransAlta put AI on the queue nobody can hear --- ## Company ### Customer service as a strategic asset. URL: https://www.dialai.ca/company Our mission is to turn customer service into something that delights customers and fuels business growth, not a cost centre to be minimised. #### What we do Dial AI, based in Vancouver, BC, brings over 30 years of collective experience in the utility software space. Since our inception we've specialised in building, deploying and maintaining advanced AI systems. Our expertise is in crafting language models tailored to your business and connecting them to the systems that hold your customers' data. By blending the precision of AI with natural language understanding, Dial AI is changing how customers interact with your business. #### Why utilities Regulated industries were the hardest place to make this work, which is exactly why we started there. The systems are old, the obligations are real, and the answers have to be right. Building for that constraint produced a product that holds up anywhere. - 30+ years: Collective experience in utility software - Vancouver: Built and supported from North Vancouver, BC Milestones #### How we got here. From automated meter reading to utilities on three continents. Before Dial AI ##### Teaching machines to read meters Our core team built some of the first AI solutions used for automated meter reading, at ServiceLink, later acquired by Harris Computer. Applying AI to utility operations is not a pivot for us; it is where we started. 2023 ##### Dial AI founded Formed to take that same approach to the other end of the utility, the contact centre, where the systems are just as old and the volume is far less forgiving. 2024 ##### Key projects at PSE&G and Philadelphia Water Large-scale deployments at two of the most demanding utilities in North America, programmes where the call volume was too high, and the window too short, to solve by hiring. 2024–25 ##### Partnerships with the platforms utilities already run Agreements with the utility software vendors, Hansen, Harris, SpryPoint and others, so an AI agent could be added in front of an established system of record rather than requiring a new one underneath it. 2026 ##### Across Canada and the United States Utilities, municipalities and co-ops in production across voice, chat, email and SMS. Now ##### Going international Early work underway in Europe and Australia, where the regulatory obligations differ and the underlying problem does not. #### Leadership ##### Peter Dunn Executive Director ##### Grady Palfrey Chief Operating Officer ##### Harper Grieve Chief Artificial Intelligence Officer ##### Jaden Howard Chief Product Officer ##### Marcus Dunn Chief Technology Officer ##### Adib Vahedi Head of Growth ##### Luke Duncan Head of Customer Engineering Contact us #### Drop us a message. We read everything that comes in and we'll get back to you. ### The answers live in your systems. URL: https://www.dialai.ca/partners We believe partnerships are essential to transforming customer service. By working with organisations that bring deep industry expertise and run the critical software, we build solutions that address real-world problems rather than demo-friendly ones. #### Vendors of essential software Our ideal partners provide the foundational systems for their industries, customer management, billing, and the other workflows the business runs on. Those systems are where the answers live, which is what makes an intelligent customer experience possible in the first place. #### Domain experts Organisations that genuinely understand their industry's needs and constraints. Those partnerships let us co-develop solutions that hold up against how the work is really done, not how it looks from outside. Our approach is to integrate AI capability into these systems. Increasing their value and letting our partners expand what they offer without building it themselves. How Grace connects to each platform. #### Why partner with Dial AI? ##### Tailored solutions Adapt our AI to the specific needs of your industry, rather than bending your customers to a generic product. ##### Integrated innovation Add AI capability to your platform without building an AI team, your customers get more from the system they already bought. ##### Mutual growth Shared success through customer satisfaction, operational efficiency, and revenue-sharing arrangements. #### Get in touch Tell us about your platform and the customers you serve, and we'll come back with what an integration would actually look like. --- ## Where to go next - Index of the site for language models: https://www.dialai.ca/llms.txt - Verified company facts: https://www.dialai.ca/ai-information - Customer stories: https://www.dialai.ca/customers - Integrations: https://www.dialai.ca/integrations - Pricing and savings calculator: https://www.dialai.ca/pricing - Book a demo: https://www.dialai.ca/demo ## Contact for corrections If anything here, or anything an AI assistant says about Dial AI, is wrong or out of date, email support@dialai.ca.