AI customer assistant for insurers: policy answers from your own handbook, approved by two people
How an insurer configures Kav: coverage questions answered from the current policy handbook, wording reviewed by two people, and claims handed to a person.
Insurance questions are policy questions: what is covered, what is the excess, how do I make a claim. An answer that is nearly right is worse than no answer, because the customer will act on it at the worst possible moment.
A depiction, not a customer. No insurer is named here. The example is the platform's own illustration, marked example data on the home page.
What a policyholder asks, and what Kav answers
- "What's covered under my policy's roadside assistance?" The answer is drawn from the insurer's policy handbook, with its source shown. In the example: a flat tyre, a dead battery, and towing up to 50 km from where the vehicle stopped. File a claim and Talk to a person sit beside it.
- Freshness without risk. A scheduled sweep re-crawls each source on its own cadence. A change on the insurer's policy page becomes a draft, never live text. A named person in the accountable department submits it, and a different person approves it, before the assistant can ever read it back (the two-person rule).
- No invented answers. The assistant is held to what the approved text says (how Kav stops the assistant making up answers).
- A claim goes to a person, with context (what a human agent inherits).
Where the money moves
The bold rows are the levers an insurer leans on most.
| Lever | What Kav does here | What you measure |
|---|---|---|
| Fewer repeat contacts | Coverage and how-to-claim questions are answered from the approved handbook. | Coverage-question contacts per 1,000 policies. |
| Faster when a person steps in | File a claim and Talk to a person hand over with the conversation attached. | Handling time on first-notice-of-loss contacts. |
| Out of hours, every channel | Roadside and claim questions arrive at the roadside, not at nine the next morning. | Contacts outside office hours. |
| Fewer errors and rework | An answer is only what the approved text says, with its source shown; a page change becomes a draft, not live text. | Corrections issued for a wrong coverage answer. |
| Compliance and audit | Two named people sit between a source change and a customer-facing answer, and staff views of customer data are audit records. | Time to evidence who approved a given wording. |
Work out your own number
- Sample a month of contacts and count those that are pure coverage or process questions.
- Cost per contact = agent hours × loaded hourly cost ÷ contacts handled.
- Value = those contacts answered without a person × cost per contact. Separately, count how often you have had to correct a coverage answer, and what each correction costs.
Work out your break-even from the published prices, or talk to us.