What problem does it solve?
For motor and many property claims, the size of the loss is decided by an expert looking at the damage: a field appraiser drives to the car or house, or a desk assessor reviews a repairer's estimate and photos. Appraiser capacity is limited, the visit can add days to the claim, and after a hail storm, flood or other large event the number of claims surges and the backlog grows. Customers wait without a car or with a damaged home, and repairers wait for approval before they can start.
Desk review has its own problem: assessors check large numbers of estimates line by line, and the consistency of the decision depends on who looks at it. Inflated or duplicated items can slip through, while honest estimates wait in the same queue.
How does it work?
- Capture the images. The policyholder receives a link at first notice of loss and is guided to take the right photos, or the repairer uploads them with the estimate.
- Check the images. The model confirms the images show the insured vehicle or property, are usable and have not been reused from another claim.
- Assess the damage. Computer vision identifies the damaged parts, the severity and the repair or replace decision per part, and estimates labour and parts cost from repair data.
- Decide the path. A clear, low value case gets an estimate or cash settlement offer within minutes; a likely total loss goes to the total loss process; everything else goes to an assessor with the AI findings.
- Review estimates from repairers. For repairer estimates, the AI compares each line with the photos and flags items that are not supported, so assessors review exceptions only.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Mobile app, Web chat, API and system to system
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 |
|---|---|---|---|---|
| Interactions handled | Not pooled | 12,000 to 160,000 | 2 | 1 organization, 1 vendor |
Value drivers: Speed and cycle time, Lower cost to serve, Customer experience, Risk and loss reduction.
Indicative value
A motor insurer handling 100,000 repairable vehicle damage claims a year
USD 1.8 million to USD 9 million
Appraisal cost avoided per year
How this is calculated
Formula: claims * eligibleShare * inspectionCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Repairable vehicle damage claims per year claims, claims per year | 100,000 | 100,000 | The reference insurer. |
| Share of claims assessed from photos instead of a physical inspection eligibleShare, fraction of claims | 0.3 | 0.6 | Conservative against the evidence on this page (Tractable reports that 90% of Admiral Seguros claim estimates were processed without human appraisers in 2021), because eligibility depends on the claim mix and customer uptake. |
| Cost of a physical or detailed desk assessment avoided inspectionCost, USD per claim | 60 | 150 | Editorial assumption covering appraiser time and travel; replace with your own appraisal costs. |
What it leaves out: Appraisal cost only. It leaves out shorter rental car periods, better estimate accuracy and leakage control, customer satisfaction, catastrophe surge capacity and the cost of the service and the integration.
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.
Covéa
France · Insurance · 2026
Covéa, the French mutual group behind the MAAF, MMA and GMF brands, has used Tractable's AI since 2018 to analyse photos of vehicle damage in real time, so that its partner repairers receive instant assessments instead of waiting on administrative steps after an accident. In April 2026 the two renewed the partnership for three years. Covéa's head of the auto networks division says more than 160,000 claims have been finalized since the collaboration began.
- Interactions handled: at least 160,000, Claims finalized since the partnership began in 2018
"Since the beginning of our collaboration, more than 160,000 claims have been finalized."
Claimed by: organization
Foyer
Luxembourg · Insurance · 2026
Foyer, a Luxembourg insurer, announced in May 2026 that it has formalised a partnership with Tractable to identify damage automatically from photographs for minor motor incidents. The aim is immediate analysis of the damage the policyholder reports, faster handling of the simplest claims and fewer visits to the garage. No outcome figures were published.
No outcome disclosed.
Tokio Marine & Nichido Fire Insurance
Japan · Insurance · 2025
Tokio Marine & Nichido Fire Insurance uses Shift Technology's claims intake and claims fraud detection solutions, extended with generative AI that extracts data from structured and unstructured sources such as images and documents. The system highlights the points handlers should check for consistency across estimates, damage photos and claim statements, which makes reviews more efficient and more standardized, including during the surge of claims after large disasters, and it helps detect suspicious claims, which tend to rise after such events. No figures were published.
No outcome disclosed.
PZU
Poland · Insurance · 2022
In March 2022 Tractable and PZU, Poland's largest insurer, announced that policyholders can submit smartphone photos of car damage when they report an accident. Tractable's AI assesses the damage and calculates the repair cost, the claim handler can share the result within minutes, and the policyholder can accept a cash settlement instead of waiting days. PZU had worked with Tractable since 2017 and already used its AI to check how body shops carry out repairs, processing several hundred thousand claims with its AI based tools. No outcome figures for the photo journey were published.
No outcome disclosed.
Admiral Seguros
Spain · Insurance · 2021
Admiral Seguros, the Spanish business of Admiral Group, sends motor claimants a link to a web app in which they photograph the damage, and Tractable's AI produces the repair estimate within minutes. Tractable reports that Admiral Seguros processed 12,000 touchless claims this way in 2021, that 90% of claim estimates were processed without human appraisers and that 98% of claims were completed in less than 15 minutes.
- Interactions handled: 12,000, Touchless claims in 2021
"In 2021, Admiral Seguros processed 12,000 touchless claims using Tractable AI."
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
- Local repair cost, labour rate and parts data, or a vendor that holds it
- Historical claims with photos and final repair costs to calibrate and test
- Rules for photo settlement per claim type, value and customer situation
Systems to integrate
- Claims management system for the claim, reserve and payment
- Customer photo capture link or app, sent at first notice of loss
- Estimating platform and repairer network systems
- Total loss valuation process
- Fraud detection for image reuse and manipulation checks
Complexity: Medium
Specialised vendor models exist for vehicle damage, and the motor deployments on this page use one (Tractable) rather than a model built in house; Tokio Marine & Nichido Fire uses Shift Technology's claims platform to review damage photos and estimates. The work is in the customer photo journey, integration with estimating and claims systems, repair cost data for the local market, and the rules that decide when a claim may settle on the AI estimate.
- 1
Choose the journey
Decide whether the AI starts with the policyholder's photos at first notice of loss, with repairer estimates, or both. Repairer estimate review can be the faster first step, because repairers already send photos with every estimate.
- 2
Calibrate on your own claims
Run the model over a few thousand closed claims with known repair costs and compare estimates, repair or replace decisions and total loss calls with what happened.
- 3
Design the photo capture for customers
Guided capture with examples and live checks for blur and angle decides how many claims can be assessed at all. Test it with real customers, not staff.
- 4
Set settlement rules
Define the value limits, claim types and confidence thresholds under which an AI estimate may be offered or approved without an assessor.
- 5
Close the loop with repair outcomes
Feed final invoices, supplements and reinspections back into monitoring so estimate accuracy is measured on real repairs.
Guardrails
- Cash settlement offers only inside value limits and confidence thresholds per claim type
- The customer can always ask for a human assessment
- Image checks for reuse, manipulation and the wrong vehicle or property before any offer
- Supplements and reinspection results monitored against AI estimates per repairer
- Total loss and injury indications always go to a person
KPIs to instrument
- Share of claims assessed from photos, per claim type
- Days from first notice of loss to estimate and to repair authorisation
- Difference between AI estimate and final repair cost, including supplements
- Share of customers who complete the photo journey
- Complaints and disputes about AI based estimates
Human in the loop
Assessors review every claim outside the settlement rules and every flagged estimate line, and sample AI settled claims each week. Customers who disagree with an estimate get a human assessment.
Common failure modes
- Customers do not finish the photo journey
- Poor guidance leads to unusable photos and a fallback inspection, which is slower than before. Invest in guided capture and measure completion.
- Estimates that miss hidden damage
- Photos show the outside, not the structural damage behind it. Allow supplements and track them against AI estimates.
- Low offers that damage trust
- A fast but low settlement offer creates complaints and regulatory risk. Monitor disputes and offer a human assessment by default.
- Reused or manipulated images
- Fraudsters submit old or edited photos. Check image metadata, similarity across claims and signs of manipulation before any payment.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
Assessing damage to vehicles or property for property and casualty claims is not listed in Annex III, which covers insurance only for risk assessment and pricing of natural persons in life and health insurance (point 5(c)). Article 50(1) transparency duties apply when the customer interacts directly with the AI, for example a guided photo journey that returns an AI estimate or offer, or a chat agent. A purely internal repairer estimate review with no customer interaction is minimal. A settlement or refusal decided solely by automated processing can fall under GDPR Article 22.
Rules that apply
Guidance
- Opinion on Artificial Intelligence governance and risk management (European Insurance and Occupational Pensions Authority, Europe). Addressed to national supervisors (August 2025). Sets out how insurers should govern AI systems across the value chain, including claims, with measures proportionate to their risk and impact on customers, such as data governance, explainability and human oversight.
Controls to put in place
- Documented settlement rules and thresholds under change control
- Estimate accuracy monitored against final repair costs per model version
- Customer disclosure that AI assesses the photos, with a right to a human assessment
- Image integrity checks logged per claim
Frequently asked questions
- How many claims can be assessed from photos without an appraiser?
- It depends on the claim mix and on how many customers finish the photo journey. Tractable reports that 90% of Admiral Seguros claim estimates in 2021 were processed without human appraisers, with 98% of claims completed in less than 15 minutes, and that 70 to 75% of customers who receive the link complete the claim. Complex damage, injuries and likely total losses are best kept with a person.
- Is this only for customers' own photos?
- No. Photo AI can also serve repairers and assessors who review estimates: Covéa's partner repairers receive instant assessments based on photos of the damage, and at Tokio Marine & Nichido Fire the AI highlights points to check for consistency across estimates, damage photos and claim statements.
- Does photo AI increase fraud risk?
- It changes it. When the estimate rests on photos alone, reused, edited or AI generated images become a risk to manage, so image integrity checks and similarity search across claims should run before any payment.
How to cite this page
Blits.ai AI Use Case Library, "AI for photo based damage assessment in insurance claims", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/photo-based-damage-assessment. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published