Insurance Chatbot: Quote Intake to Handoff

The bot does not quote

This is the first and most important decision in an insurance chatbot: the bot does not give a price. The reason is not technical but legal — a figure stated is treated as an offer, and revising it afterwards is both a trust problem and frequently a contractual one.

That does not make the bot useless. It clarifies its job: collect the fields a quote requires, explain published terms, and hand over to an underwriter or sales consultant the moment the information is complete.

The practical difference is that a customer can give their details at eleven at night and receive a complete quote in the morning. Today they fill in a form, receive 'someone will be in touch', and answer the same questions again the next day.

Quote intake: fields by product line

The fields depend on the product, and a generic form fits none of them. Here is the minimum set for three common lines.

1. 1MotorMake, model, year, engine size, use (personal or commercial), driver age and licence history, previous incidents and the coverage level wanted. All factual, and none requires an underwriting judgement.
2. 2PropertyProperty type, size, year built, construction, location, approximate value, owned or rented, and what they want covered.
3. 3TravelDestination, dates, number of travellers and their ages, purpose of travel and whether any additional activities are involved. Age is decisive here and is the most commonly forgotten field.
4. 4Common to all threeContact details, whether they are an existing customer, and by when they need it. That last field sorts the queue and is almost never asked.

The sensitive data boundary

Some eligibility information in insurance is sensitive and must not be collected in chat. Write this boundary down in advance, because deciding mid-conversation is too late.

• Health conditions, chronic illness, medication and medical history — these belong in an authenticated channel or with an underwriter.
• Document numbers, identity document scans and bank details — never requested in chat.
• Detailed previous claims history — a general question is fine, but the detail belongs with an operator.
• Information about third parties — family members, employees, additional drivers — should not be collected without their consent.
• The published notice explaining why personal data is collected and how long it is kept is a question the bot can and should answer.

Questions that can safely be answered

• Which products exist and what each covers, quoted from the published terms.
• Standard exclusions — what is not covered. This is the most asked and least written question in the sector.
• How a policy comes into force, payment options and the payment schedule.
• Documents required for a claim and the process steps in order.
• Renewal, amendment and cancellation rules.
• Branch addresses, opening hours, and which transactions are online versus in person.

Claims deserve a separate path

A claim is not a sales conversation and must not land in the sales flow. This is a common and expensive misroute — offering insurance to someone who has just had an incident is the worst possible impression.

• Detect the intent early: words like 'accident', 'damage' and 'incident' should route to the claims path immediately.
• For emergencies — a road accident, a fire, a medical event — give a pre-written reply: the official emergency numbers and the company's claims line, then transfer to a person immediately.
• The bot can state the process steps and the document list. That is factual and genuinely useful.
• The bot must never say that a claim will be paid, how much will be paid, or whether an incident is covered. Those are underwriting and loss-adjustment decisions.
• The status of an existing claim requires an authenticated channel, because identity cannot be verified in chat.

What to measure

1. 1Complete-intake rateEnquiries reaching the underwriter with every required field present. This determines how many rounds a quote takes.
2. 2Time from enquiry to quoteMeasured separately for out-of-hours enquiries. In insurance the customer typically approaches three companies the same evening.
3. 3Rounds per quoteHow often you go back for missing information. A falling number is the clearest evidence the intake is working.
4. 4Claims misroute rateClaims conversations landing in the sales flow. This should be near zero and it shows whether intent detection is working.
5. 5Out-of-hours enquiry shareUsually large, and it justifies the project on its own.

How Vexvon works in insurance

The bot answers from the knowledge base — your site crawled to as many as 4,500 pages, PDFs up to 25 MB for policy wordings and product documents, and bulk entries at up to 200 per request. Entries are typed into thirteen categories including policy, pricing, branch and campaign, which is what lets published terms be held separately and replaced with a single edit when they change.

Answers are tied to retrieved material rather than improvised: six results by default, with a defined fallback when no context is found. Answer transparency records which entry produced a reply — important in insurance, because you need to be able to show where a coverage statement came from.

Extraction runs on the conversation: phone numbers are detected through a five-stage process, free-form fields such as vehicle details, dates and desired coverage are captured alongside, and a one-sentence summary is written. A buyer-intent filter removes conversations that are not genuine enquiries before a lead is created.

The escalation paths are explicit: a request for an operator raises a notification, an agent's stop character pauses the AI on a conversation for thirty minutes, and the AI can be switched off for a conversation entirely — which is the mechanism the claims path requires. The lead reaches the sales group as a Telegram notification with a Start button and moves through a nine-stage pipeline.

Frequently asked questions

1. 1Can an insurance chatbot give a price?No. A figure stated is treated by the customer as an offer, and price depends on underwriting rules. The bot's job is to collect the fields a quote requires.
2. 2What should quote intake ask?It depends on the product: for motor, make, model, year, engine, use, driver age and licence history; for property, type, size, year built and value; for travel, destination, dates and traveller ages. All three add contact details and when it is needed by.
3. 3What must not be collected in chat?Health conditions and medical history, document numbers and scans, bank details, and information about third parties without their consent.
4. 4How should claims be handled?On a separate path. Detect the intent early, give a pre-written reply in emergencies, let the bot state only the process and document list, and never let it decide payment or coverage.
5. 5Can the bot answer coverage questions?Yes, quoted from the published terms, including standard exclusions. It must not say whether a specific incident is covered — that is a loss-adjustment decision.
6. 6How should success be measured?Complete-intake rate, time from enquiry to quote, rounds per quote, and the share of claims conversations that landed in the sales flow.

Review your last fifty enquiries

Take the last fifty quote enquiries and mark, for each, which fields the underwriter had to go back and ask for. In most companies the same two or three fields account for most of the list, and asking for them at intake is an afternoon's change.