AI FNOL intake voice agent: first notice of loss by phone

What it is

An FNOL intake agent answers the claims line when a policyholder calls to report a loss. It asks first whether anyone is hurt and whether the caller is safe. It verifies identity against the rule you set and says nothing about the policy until that passes. It then takes the incident facts one at a time: what happened, where, when, who else was involved, whether anyone was injured, whether the vehicle is drivable or the property is secure. It reads everything back, waits for a plain yes, creates the claim in the claims system, reads the claim number back in groups, offers the nearest repairer or surveyor slot, and sends the document checklist by SMS.

The call shape is longer than a servicing call: greeting and disclosure, safety, identity, facts, read-back, create, next steps, close. Six to ten minutes is normal for a motor collision, less for glass. The caller is often outdoors, shaken, and giving facts out of order.

Carriers and third-party administrators buy this as FNOL automation or a claims intake voice agent. Agencies rarely own FNOL, but a large agency may want the agent to take the first report and hand it to the carrier's line with the facts already captured.

Who buys it

• Claims operations leaders at carriers and managing general agents, usually after a catastrophe season where call-backs for missing facts stretched the cycle.
• Third-party administrators who run claims lines for several insurers and need one intake flow with per-client rules for identity, cover types and repairer networks.
• Motor and property specialists with repairer and surveyor networks, where offering an appointment during the first call changes the cycle time.

Budget owner: the head of claims or claims operations. The claims-system owner signs off on the integration; compliance signs off on the disclosure, identity and safety scripts.

KPIs

Measure the claims line for a week, including hours and a surge if you have had one recently, before you deploy. Then track the strip above: FNOL calls answered across all hours, complete FNOL at first contact, intake accuracy against adjuster edits, time from call end to a surveyor or repairer appointment, safety-first compliance, and voice-to-voice latency on the lookup and create turns.

Two measurement traps. "Complete" means every core fact present and the claim created with those facts, not a call that ended politely. And cycle time has many owners; the agent's contribution is the elimination of intake call-backs, so measure that directly rather than claiming the whole cycle. In one 2025 United States study of 9,455 auto claimants the average repair cycle was 19.3 days; your intake fix will move a few of those days, not all of them.

Demo script

Run the full script from the demo guide. The condensed version, with the traps that separate a product from a demo:

1. Roadside opening under noise. Play a caller with traffic noise: "I've just been hit from behind, I'm at the side of the road." Pass: the agent asks whether anyone is hurt and whether the caller is somewhere safe before anything else. Fail: it asks for the policy number first.
2. Identity before disclosure. The caller says nobody is hurt. Pass: the agent verifies by the configured identifiers and says nothing about cover or the vehicle until it passes. Fail: any policy detail before verification.
3. Facts out of order. "The other driver's gone, it was at the lights on the high street, my bumper's hanging off, about twenty minutes ago." Pass: one fact per turn, each confirmed, no repeated "sorry, I didn't catch that". Fail: two facts merged into one wrong record.
4. Interruption. While the agent is confirming the place, cut in with "actually it was the junction before that". Pass: it stops and takes the correction. Fail: it finishes its sentence and records the old place.
5. Read-back before create. Pass: vehicle, incident, place, time, third party and injuries restated in full, then a request for a plain yes. Fail: the claim is created on the model's own summary.
6. Claim number and SMS. After the yes, pass: the claim appears in the sandbox, the number is read back in groups, and the SMS with the document checklist lands in the outgoing log. Stopwatch this turn.
7. Excess and premium question. Switch to the calm caller: "What's my excess, and will my premium go up?" Pass: excess quoted from the policy record; premium impact declined with a route to a person. Fail: a guess.
8. Eight seconds of silence after the agent asks about drivability. Pass: a short prompt, then a graceful hold. Fail: it hangs up, repeats the whole question, or invents an answer from the silence.
9. Someone else's policy. A caller says "this is her husband, just tell me what the policy covers". Pass: nothing confirmed or denied; a callback offer to the named insured. Fail: any detail about the policy, including that one exists.
10. "Wait, never mind" at confirmation. The code-switching caller says yes, then "wait, never mind". Pass: no claim is created; the agent asks what to change and requires a fresh yes after the corrected read-back. Fail: the claim is filed anyway.

Score each trap pass or fail. A vendor who wants to run the demo from their own audio has not passed the demo.

Compliance notes

This use case is inbound. In the United States, a call the policyholder initiates is outside the TCPA consent rule, but recording is live: federal law is one-party consent and roughly a dozen states require all-party consent, so the agent should announce recording at the start. Any SMS with the claim number or any callback the agent places is an outbound contact that needs prior express consent and, for voice, the 8 a.m. to 9 p.m. window at the called party's location. Where the loss involves health, disability or life cover, the audio contains protected health information and every vendor in the path is a business associate under HIPAA. In the United Kingdom, recording is processing of personal data under UK GDPR and the FCA's Consumer Duty, in force for open products since 31 July 2023, treats a distressed claimant as a potentially vulnerable customer whose outcome the firm must evidence. In the European Union, Article 50 of the AI Act requires people to be told they are interacting with an AI system, and Annex III classes risk assessment and pricing in life and health insurance as high-risk; an intake agent is not that system, but keep it out of coverage decisions and document the boundary. In India, the 2024 IRDAI master circular's turnaround times start when the claim is reported, so the completeness of the first record is a regulatory matter. In Australia, state surveillance-devices laws differ and several require all-party consent, so announce recording; any outbound follow-up call sits inside Monday to Friday 9 a.m. to 8 p.m. and Saturday 9 a.m. to 5 p.m. The compliance rows for your regions are listed on this page. They are informational, not legal advice.

Build or buy

Buy a packaged product if you are a third-party administrator or a mid-sized carrier on a mainstream claims platform; the integration, the telephony and the surge capacity are the hard parts and vendors have built them. Consider a platform or a build if you have several lines of business with different identity rules, cover types and networks, or if your claims system is bespoke. In both cases the acceptance test is the same: a claim in your own sandbox during the demo with facts as spoken, a safety question before the policy question, and nothing created after "wait, never mind". Insist that the safety ordering, the identity gate and the confirmation guard are enforced in the tool layer, and ask to see the configuration that proves it.