Shaped by Your Best Calls.
Not a Generic Script.
Most AI voice agents answer from a prompt somebody wrote in an afternoon. This one opens the call already briefed on how your highest-scoring calls handled the same question, already knowing whether this caller has reached you before, and — once they verify — already looking at whether their line is actually down.
What the agent draws on, before it says hello
01
Shaped by Your Top-Scoring Calls
Every week the calls that clear your Best Calls Library bar (a score of 85 by default) are distilled into a playbook draft: how your people open, acknowledge, explain and close. Nothing reaches a caller until someone you have authorised publishes it, and any earlier version is one click to roll back to. Mid-call the agent can pull how your best reps handled the same situation, and it answers factual questions from your Knowledge Base. Its own calls are never fed back in, so it keeps learning from your people rather than from itself.
02
Recognises the Caller Without Trusting Caller ID
A phone number is trivially spoofable, so matching one earns the caller very little. The agent knows whether this number has reached you before and how recently — and nothing more — until identity is actually verified. What it is allowed to say expands as trust does, and the filtering happens before the data is handed to the model rather than by asking the model to be discreet.
03
Talks to the Network, Not Just the Customer
Connected to Calix Cloud, the agent fetches the caller’s network snapshot the moment they verify: whether their equipment is online, whether their phone line is affected, whether neighbours report the same fault, and whether the line can carry their plan. It can lead with what it sees instead of making the customer describe it. Before verification it sees nothing about the line, because a caller ID alone proves nothing.
On the call today
What the Agent Can Do
Each capability is switched on by the integration or setting it depends on, so what a caller gets matches what you have connected.
Answers and Acts
- Answers questions from your Knowledge Base
- Files a NISC iVUE trouble ticket on a verified caller’s own service, using only the report codes your admins approved — or comments on the ticket already open
- Answers “can I get service at this address?” from your records, when you switch it on
Reads the Caller’s Line
- Once the caller verifies: whether the line is up, whether the phone is affected, and whether the line can carry the plan
- Wi-Fi that is hidden, switched off or bridged, and a gateway that rebooted minutes ago
- The latest speed test against the plan, where your Calix licence includes it
- Fiber or copper and the device in the home, from NISC — and the caller’s own IP address if they ask
Speaks First When It Should
- Once the caller verifies, raises what is outstanding — an open ticket, a commitment, an earlier escalation
- Says up front what this call cannot do, such as read out a balance, before asking the caller for anything
- Each kind of offer is made once at most, under a cap per call, so it does not keep repeating itself
Gets to the Right Person
- Rings a person or a department, each with its own hours and a line on what that desk handles
- Where your phone system reports live agent status, it decides between a warm whisper and a straight transfer into the queue — never which desk
- The whisper is spoken in the agent’s own voice; callers hold to your own music
- When nobody is free it takes a callback into a queue your team claims in one click, and can email a fallback address
Stays Inside Its Limits
- A daily minute budget reserved before each call is answered, and refunded to the real length afterwards
- Limits per caller number on verification attempts and on how often it can ring a person
- Attempts to talk it out of its instructions are flagged for a person to read
- A “What your agent can do” card shows which capabilities are on, and what turns the rest on
Watch It Work
- Live IQ shows every call in flight beside the lookups behind each answer, and the Knowledge Base entries it used
- Pick up any call from the AI without waiting to be rung
- AI Calls lists every call it answered, with whether the disclosure played first
A Generic AI Voice Agent vs. This One
| Dimension | Generic AI voice agent | Call Coach IQ |
|---|---|---|
| Knowledge source | A static prompt, or a knowledge base someone maintains by hand | A playbook distilled from your own highest-scoring human calls, plus your Knowledge Base |
| Caller context | A CRM lookup keyed on caller ID, with the match taken as proof of identity | The same CRM, read in tiers — a number match earns recognition, not account detail |
| Identity verification | None, or a security question the model itself decides the caller passed | CPNI authorisation runs in NISC iVUE — protected fields never reach the model until it passes |
| Network diagnostics | Not available; the caller has to describe the fault | A live Calix Cloud snapshot after verification, and a NISC trouble ticket when it needs a technician |
| Changes over time | No — the same prompt on every call until someone rewrites it | A new playbook draft each week from your latest top calls — live only once a person publishes it |
What Has to Be True Before It Answers
Five independent switches have to agree before a caller ever hears the AI: your plan tier, a super-admin toggle on your company, your own opt-in, enablement on that specific phone number, and a platform-wide kill switch. Any one of them off and the call goes to a person instead — the fallback is a human, never a dropped call.
A company that has not written its AI disclosure text is refused outright, and the line cannot be armed until your own admin has signed off the exact disclosure and greeting wording. If the recorded disclosure ever fails to play, the agent speaks your approved wording itself, and the call records why — an alert fires on any call where the recording did not play. None of the agent's tools accept a customer identifier as an argument, which means there is no phrasing a caller can use to talk the agent into looking up somebody else's account.
Customer outcomes
What the Caller Actually Experiences
The plumbing underneath is the point for you. For the person on the phone none of it matters — only what changes about the call itself.
Answered, and told it is an AI
No hold queue. The agent picks up and identifies itself as an AI before anything else — a company that has not written that disclosure does not get to run the agent at all.
Does not start from zero
It knows whether this number has reached you before and how recently — and, once the caller verifies, whether something is still open — so the call opens somewhere better than "how can I help you today?"
Gets a real answer
Handled the way your best-scoring calls handle it — the approach that actually works on your customers, not a scripted deflection or a vague "I can’t help with that."
The line gets checked, live
A connection problem is read from network telemetry during the call rather than logged for somebody to look at tomorrow.
When it can't solve it
The agent transfers to your team, and it is built to do that readily rather than talk its way around a question it should not answer. On a warm transfer, the person who picks up hears a short spoken summary before they are connected, and when the call is routed to a named teammate they see the caller's account and history on screen, so the caller does not start again. Every AI call is recorded, transcribed and summarised, and checked for promises it didn't keep and for what it misheard — never graded against a rubric written for people.
Who Benefits from the Live AI Voice Agent
For the Customer
The phone gets answered, by something that tells you what it is, already knows you have called before, and can see that your connection is down without asking you to unplug the router first.
For the Agent
The calls that reach you are the ones that needed a person. The repetitive tier-one traffic that used to fill your queue does not disappear into a void either — every AI call is recorded, transcribed and summarised, so you can see exactly what the agent handled on your behalf.
For the Business
Coverage outside staffed hours without writing a script, using the approaches your own best calls already proved. Five separate switches, mandatory AI disclosure, and trust-tiered data exposure mean turning it on is a decision you can reverse in one click.
How It Fits with the Rest of the Platform
The voice agent is not a bolt-on — it runs on the data the platform already collects:
The scores that decide which calls are worth learning from. No separate training step — the same rubric that grades your agents shapes what the voice agent is briefed with.
The voice agent answers callers from the same Knowledge Base your team searches, and Live IQ lists exactly which entries it used on each call.
The full interaction history for a customer, for the people entitled to see it. The voice agent deliberately reads far less than this until the caller is verified.
AI-handled calls are deliberately kept out of a rulebook written for people. Instead each one records when the approved disclosure played and which trust tier the caller reached, and is checked for promises the agent made that its tools did not keep.
See What It Does with One of Your Calls
Book a walkthrough and we'll show you what the agent is briefed with on a real inbound call, where the trust tiers cut it off, and what a transfer looks like from your team's side.
Request a DemoEnterprise plan. Live today as a single-tenant pilot, armed per company and per phone number after a provisioning step and your own sign-off on the disclosure wording. Twilio for telephony, NISC iVUE for accounts and tickets, Calix Cloud for network diagnostics.