What problem does it solve?
The first notice of loss is the moment the insurer's promise is tested. A customer who has just had a car accident, a burst pipe or a stolen bag calls or logs in to report it, often upset and often at night or at the weekend. Today much of that intake is a scripted conversation with a human handler who types answers into the claims system: policy number, date and place, what happened, who was involved, photos and documents to send later.
The work is repetitive but it drives everything downstream. Missing or inconsistent facts at intake cause call backs, wrong triage, slower settlement and missed fraud or recovery signals. After a storm, many customers report losses at the same time, and queues grow exactly when customers need the insurer most. Intake by web form alone does not solve it: forms can be long, and many customers still prefer to call (Travelers launched its voice agent for exactly those customers).
How does it work?
- Identify the customer and the policy. The agent recognises the caller or logged in user and matches the policy, for example by reading back a spoken policy number or registration.
- Take the story in the customer's words. It asks what happened and extracts the structured facts the claim type needs (date, place, cause, parties, damage, injuries), asking follow up questions only for what is missing.
- Collect evidence while the customer is there. It asks for photos, receipts or reports through a link or upload and checks that they are readable and relevant.
- Open the claim and set expectations. It creates the claim in the claims system through an API, returns the claim number and explains the next steps and timelines for this claim type.
- Hand over when it should. Injuries, vulnerability signals, disputes about cover, complex commercial losses and anything the customer asks a person for go to a handler with the transcript and extracted facts attached.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Phone and voice, Web chat, Mobile app, WhatsApp
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Accuracy | Too few to pool | 90% | 1 | 1 vendor |
| Automation rate | Too few to pool | about 55% | 1 | 1 organization |
| Containment rate | Too few to pool | 96% | 1 | 1 organization |
Value drivers: Customer experience, Lower cost to serve, Speed and cycle time, Inclusion and access.
Indicative value
A motor and home insurer that receives 300,000 claims a year
USD 315,000 to USD 2.5 million
Handler time avoided at first notice of loss per year
How this is calculated
Formula: claims * assistedShare * containment * minutesPerIntake / 60 * costPerHour. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Claims reported per year claims, claims per year | 300,000 | 300,000 | The reference insurer. |
| Share of claims reported through a human handler today assistedShare, fraction of claims | 0.4 | 0.6 | Editorial assumption; replace with your own channel mix. |
| Share of those intakes the agent completes without a handler containment, fraction of assisted intakes | 0.3 | 0.6 | Conservative against the evidence on this page (Lemonade reports that its claims bot takes the first notice of loss without human intervention 96% of the time), because that figure comes from an insurer whose customers already file claims by chatting with its bot in the app. |
| Handler minutes per first notice of loss, including wrap up minutesPerIntake, minutes per claim | 15 | 25 | Editorial assumption, replace with your own handling time. |
| Fully loaded cost per handler hour costPerHour, USD per hour | 35 | 55 | Editorial assumption, replace with your own. |
What it leaves out: Intake effort only. It leaves out the effect of better captured facts on triage, leakage and fraud detection, surge capacity after catastrophes, the cost of running the agent and the integration work with the claims and policy systems.
Who already uses it?
5 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Hippo
United States · Insurance · 2026
Hippo, a US insurance platform that describes itself as a multi line carrier, reported in its first quarter 2026 investor update that it launched Clara in that quarter, an AI assistant for first notice of loss and end to end claims processing, as part of a wider move to agentic AI in claims, including capacity for catastrophe events. The presentation gives expectations, such as the share of homeowners claims it expects to be filed digitally, but no measured results yet.
No outcome disclosed.
Travelers
United States · Insurance · 2026
In its 2025 annual report to shareholders, Travelers says it launched a natural language generative AI voice agent that takes first notice of loss by phone, used first for auto damage claims and planned to expand to more lines of business and claim interactions. The same letter reports that more than half of all claims are eligible for straight through digital processing, which customers choose about two thirds of the time, and that another 15% of claims are handled with advanced digital tools. No outcome figures for the voice agent itself were published; the company describes early adoption and feedback as positive.
No outcome disclosed.
Lemonade
United States · Insurance · 2025
Lemonade's claims bot AI Jim takes the first notice of loss in a chat with the customer, pays or declines simple claims within seconds and assigns the claims it may not settle, or has concerns about, to human claims experts based on their specialty, workload and schedule. A separate system, Forensic Graph, uses machine learning to predict, detect and block fraud across the customer engagement. The annual report states that AI Jim took the first notice of loss without human intervention 96% of the time and that roughly 55% of claims were automated from start to finish, both as of December 31, 2025.
- Containment rate: 96%, First notice of loss taken without human intervention, as of December 31, 2025
"AI Jim is our claims bot, and, as of December 31, 2025, 96% of the time, it is AI Jim that will take the first notice of loss from a Lemonade customer without human intervention"
Claimed by: organization - Automation rate: about 55%, Share of claims automated end to end, as of December 31, 2025
"As of December 31, 2025, roughly 55% of our claims were automated, resulting in instant or near-instant processing from start to finish."
Claimed by: organization
Progressive
United States · Insurance · 2025
In its 2025 letter to shareholders, Progressive says it implemented digital claims capabilities in 2025 that let customers interact from the first notice of loss through investigation, damage assessment and repair. The same program gave claims employees a new text and email communication platform that includes a customer facing generative AI assistant for automated tasks, information retrieval and tailored follow up actions. No outcome figures were published.
No outcome disclosed.
DOMCURA
Germany · Insurance · 2023
DOMCURA, a German underwriting agent, turned its claims chatbot Claimens into a phone based AI agent with Parloa that guides callers through the recurring steps of reporting a claim, matches the caller to the policy by recognising the policy number, and covers more than 20 types of damage claims that the DOMCURA team configured itself. It went from kickoff to live launch in three months. Parloa reports a 90% recognition rate for caller requests.
- Accuracy: 90%, Recognition of caller requests
"90% recognition rate for requests"
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- The question set and required evidence per claim type and line of business
- Policy and customer data reachable through APIs
- Recorded or transcribed historical intake calls to design and test against
- Approved wording for next steps, timelines and cover explanations
Systems to integrate
- Policy administration system for policy and cover lookup
- Claims management system to create and update the claim
- Telephony or contact centre platform for voice intake and handover
- Document and photo upload with storage linked to the claim
- Identity verification and customer authentication
Complexity: Medium
The conversation is the easy part. The work is in policy lookup and identity checks, writing a complete claim into the claims system through APIs, handling many claim types with different questions, and a clean handover for injured or vulnerable claimants.
- 1
Start with one line and simple claim types
Pick high volume, low complexity intakes such as glass, minor motor damage or personal possessions, where the facts are predictable and injuries are rare.
- 2
Write the intake schema per claim type
List the mandatory facts, the optional ones and the evidence per claim type, and let the agent fill that schema instead of following a fixed script.
- 3
Connect the systems before going live
The agent must create a real claim with a real claim number. An agent that only emails a transcript to a queue moves work around without removing it.
- 4
Design the handover rules
Decide which signals in what the caller says send them to a person (injury, vulnerability, anger, disputes about cover, commercial losses) and pass the extracted facts so nobody asks twice.
- 5
Test with real recordings and surge scenarios
Replay historical calls, noisy lines, accents, partial policy numbers and catastrophe volumes, and track field level accuracy against what a handler recorded.
- 6
Measure claim quality, not only containment
Follow each agent intake downstream: call backs for missing facts, triage corrections and customer complaints tell you more than the share of calls without a handler.
Guardrails
- The agent never tells a customer that a loss is covered or declined; cover decisions stay with the claims process
- Automatic handover on injury, vulnerability, distress, disputes about cover and any request for a person
- Every extracted field is confirmed back to the customer before the claim is created
- Identity and policy match checks before any claim is opened or personal data is disclosed
- Personal and health data masked in logs, minimised in model prompts, with retention aligned to claims files
KPIs to instrument
- Share of intakes completed without a handler, per claim type
- Share of agent opened claims that need a call back for missing or wrong facts
- Field level accuracy of extracted facts on a weekly sample
- Time from first contact to claim number, and customer satisfaction after intake
- Handover rate and handover reasons
Human in the loop
Handlers own every claim the agent opens and review a daily sample of agent intakes against the recording. They take over complex, injured and vulnerable claimants, and claims leadership approves each new claim type before the agent handles it.
Common failure modes
- Fast intake, poor claim file
- The agent completes the call but misses facts handlers need, so the work moves downstream. Measure call backs and triage corrections per claim type.
- Implied promises about cover
- A friendly answer such as "that will be covered" creates an expectation the insurer may not meet. Block cover statements in the prompt and in output guardrails.
- Vulnerable claimants kept in automation
- Bereaved, injured or distressed customers are pushed through a script. Detect the signals in what the customer says and hand over early.
- Collapse under catastrophe volume
- The agent is sized for normal days and fails during a storm. Load test for catastrophe peaks and keep a simple fallback that still captures the essentials.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A customer facing intake agent must be designed so that people know they are interacting with AI (Article 50(1)). Claims intake and claims handling are not listed in Annex III: point 5(c) covers risk assessment and pricing in life and health insurance, not claims. One design choice changes this: detecting distress by inferring emotions from the caller's voice is emotion recognition based on biometric data, which is high risk under Annex III point 1(c) and needs disclosure under Article 50(3). Detecting vulnerability from what the caller says does not. The limited tier assumes that design: every handover signal on this page (injury, distress, anger, vulnerability) is detected from the words of the conversation, and inferring emotions from the voice itself is out of scope.
Guidance
- Opinion on Artificial Intelligence governance and risk management (European Insurance and Occupational Pensions Authority, Europe). Published in August 2025 and addressed to national supervisors, it clarifies how the principles and requirements of existing insurance legislation apply to AI systems, following a risk based and proportionate approach.
- Regulation (EU) 2024/1689 (AI Act), Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). People must be informed that they are interacting with an AI system unless this is obvious from the context.
Controls to put in place
- AI disclosure at the start of every intake conversation
- Inventory entry with an accountable claims owner and the list of claim types in scope
- Transcript and extracted fields stored with the claim for audit and dispute handling
- Regression tests on recorded calls for every change in prompts, models or claim types
- Monitoring of outcomes for vulnerable customers and complaints that mention the agent
Frequently asked questions
- How much of first notice of loss can an AI agent take without a human?
- For simple, digital first claim types it can be most of it: Lemonade's annual report states that, as of December 31, 2025, its claims bot took the first notice of loss without human intervention 96% of the time. Incumbent insurers with phone heavy, complex lines should expect lower shares and start with simple claim types such as glass or minor motor damage.
- Does a voice agent make sense when most insurers push digital claims?
- Yes, because many customers still call. Travelers says it launched a generative AI voice agent for first notice of loss by phone to serve customers who prefer to call, starting with auto damage claims, next to its straight through digital journey.
- Should the intake agent tell customers whether they are covered?
- No. The agent records the loss and explains the process; cover decisions belong to the claims process with its rules and handlers. Statements that imply cover create expectations and complaints, and should be blocked by guardrails.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for first notice of loss claims intake", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/claims-first-notice-of-loss-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published