Computer vision claims

Category: AI in insurance · Reviewed by Taylor Watts, Broker · New Business · Last reviewed 2026-06-10

Computer vision in claims is the use of convolutional neural networks and related image-processing models to detect, classify and assess physical damage from photographs, video and remote-sensing imagery. In United Kingdom insurance the technology has matured into a production layer behind motor damage assessment and property roof analytics, supplied chiefly by specialist vendors and integrated into insurer and managing-general-agent workflows.

Category: AI in insurance · Aliases: AI image claims, automated damage assessment, photo-FNOL · Established: UK motor production use from c.2018 (Tractable et al.); aerial property analytics c.2020 onwards · Related: AI in claims processing, Neural network underwriting, Insurtech

Definition

Computer vision claims workflows typically use convolutional neural networks (CNNs) and transformer-based vision models to:

• detect and classify damaged vehicle parts from customer-supplied photographs;
• estimate repair cost using an associated parts and labour database;
• detect roof condition, swimming pools, debris and tree-canopy from aerial or satellite imagery;
• detect property damage post-event from drone or satellite imagery; and
• verify identity documents during onboarding.

Legal / Regulatory basis

• ICOBS Chapter 8 — automated damage assessment outputs feed claims decisions that must still be handled promptly and fairly.
• FCA Consumer Duty (PS22/9) — consumer-understanding and consumer-support outcomes; the customer must be able to understand the decision.
• UK GDPR — photographs may contain personal data (vehicle registration, faces) and biometric special-category data; lawful basis and DPIA are required.
• Equality Act 2010 — image models must be tested for performance across user demographics and across vehicle / property segments to avoid disparate impact.
• FCA & PRA DP5/22 / FS2/23 — CNNs are explicitly within scope of supervisory governance expectations.
• Civil Aviation Authority (CAA) rules for drone operation under the UK Air Navigation Order 2016 and the post-Brexit UAS regulations (CAP 722) where insurers commission drone imagery.
• PRA SS1/23 — directional model-risk discipline for the underlying CNNs.

How it works in practice

A typical UK motor photo-FNOL flow:

1. Customer reports a claim and is invited to upload images through the insurer’s app, with on-device guidance to capture all required angles.
2. The vendor’s CNN (e.g. Tractable, Solera/Audatex Qapter, Mitchell Intelligent Open Shop, CCC Intelligent Solutions) detects damaged panels and predicts severity.
3. The model proposes an estimate, leveraging the vendor’s parts and labour database for the UK market.
4. The output is fed back to the handler — repair vs. total loss recommendation, with confidence scores.
5. For low-severity, well-defined damage, the estimate may be auto-approved within delegated authority.
6. For higher-severity or contested cases, a human assessor or engineer reviews.

Property workflows often substitute aerial imagery from CAPE Analytics, EagleView, Nearmap and emerging satellite analytics players.

Common variations / Subsequent developments

• Auto-FNOL apps that guide image capture in near-real time at the roadside.
• Drone-based surveys for large commercial losses (industrial fire, storm damage) integrated with vision models that produce volumetric damage estimates.
• Salvage and total-loss valuation integrated with auction data.
• ABI guidance on photo-based claims handling and customer communications.
• Cross-vendor data sharing under contractual safeguards to improve damage models for rare vehicles.
• Generative AI overlay producing customer-facing summaries of the damage assessment.

Example

A UK home insurer integrates aerial roof analytics into its underwriting and claims platforms. At new business, the vendor returns a roof-age, material and condition score, which feeds the GBM pricing model. Following a windstorm, the same vendor flies revised imagery; the insurer pre-empts claim notifications by identifying damaged roofs and contacting affected policyholders, consistent with the Consumer Duty consumer-support outcome.

See also

• AI in claims processing
• Neural network underwriting
• Insurtech
• Parametric insurance

References

• FCA Handbook, ICOBS Chapter 8. https://www.handbook.fca.org.uk
• FCA, PS22/9 — Consumer Duty, July 2022.
• FCA & PRA, DP5/22 / FS2/23 — AI and Machine Learning, October 2022 / 2023.
• PRA, SS1/23 — Model risk management principles for banks, May 2023.
• Civil Aviation Authority, CAP 722 — Unmanned Aircraft System Operations in UK Airspace.
• Lloyd’s, Emerging Technology in Claims, market bulletins.
• ABI, Consumer Duty and Claims, member guidance.
• UK GDPR; Data Protection Act 2018, https://www.legislation.gov.uk

This entry is part of the Apex Insurance Wiki. Last reviewed by Matt Bartlett on 2026-06-10. Next review: 2026-12-10.

Apex Insurance Brokers Limited. Authorised and regulated by the Financial Conduct Authority, FRN 724952. Registered in England and Wales, Companies House 07014570. This entry provides general information about UK insurance concepts and is not regulated advice. Consult your insurance broker on your specific position.