Motor Claims Automation at a UK General Insurer — 80% Straight-Through Settlement Inside 24 Hours
Claims Router + DocuMage + Multi-Agent Orchestrator delivering same-day settlement on low-complexity motor claims — at FCA-grade audit standards.
The challenge
The client — a UK-domiciled general insurer with a motor portfolio of roughly 2.2 million policies — was operating a claims process whose customer-experience metrics had been trending downward for four consecutive years. Average time from first-notification-of-loss (FNOL) to claim settlement was 18 calendar days for low-complexity claims (windscreen replacement, minor own-damage, third-party-only liability) that should, by industry consensus, settle in 48 hours.
The bottleneck was the claims handler workload. Claims handlers were spending the majority of their time on document collection (chasing the policyholder, the body-shop, the third party), document review (reading the same repair-estimate PDFs the previous handler had read), and case-management-tool data entry. The actual judgement-intensive work — coverage interpretation, fraud assessment, third-party negotiation — was a small fraction of the handler's day.
The insurer's COO had set a target of bringing low-complexity motor claims to under 48 hours end-to-end. Two prior attempts — one with a major insurance-tech vendor and one in-house — had stalled at the document-intelligence problem: neither had been able to extract the required structured fields from the mix of body-shop estimates, photos, police reports, third-party correspondence and policyholder statements that even a simple claim generates.
The approach
MindMap deployed Claims Router (Cr) as the policy-and-coverage interpretation layer, DocuMage as the document-intelligence engine, Multi-Agent Orchestrator (Mo) as the workflow coordinator and Anomaly Detector (Ad) as the fraud-and-leakage layer. The accelerators were composed into a unified low-complexity-motor-claims pipeline that ran in parallel to the existing claims-handling tooling for the first eight weeks before traffic was switched.
Phase one was claim segmentation. We worked with the insurer's claims leadership to define the segment that the automated pipeline would handle: motor claims under a defined loss-value threshold, with no third-party injury, no policy-coverage ambiguity, no prior-claim-history flag, and no fraud-indicator flag at FNOL. This segment accounted for approximately 58% of total motor-claim volume.
Phase two was the document-intelligence build. DocuMage was trained on the insurer's historical claim-document corpus: roughly 1.4 million body-shop estimates, 800,000 damage-photo sets, 240,000 police reports, and the full range of correspondence templates from the major repairers, salvage operators and third-party insurers the insurer interacts with. The model handles UK-specific formats end-to-end — DVLA documents, V5C registration documents, MOT certificates, the insurer's own claim-form templates.
Phase three was the settlement-orchestration layer. For a typical low-complexity claim, the orchestrator runs ten to fifteen parallel sub-tasks: cover-and-policy lookup, premium-and-excess validation, repair-estimate parsing, third-party-vehicle lookup (if applicable), repairer-network availability check, no-claim-discount preservation check, fraud-flag check, payment-method validation, customer-communication drafting, and case-file finalisation. The orchestrator runs these in parallel where possible and serialises them where dependencies require.
The pre-built building blocks
Rather than commission a ground-up build, the engagement leaned on MindMap's pre-built accelerator library — production-tested components that compress what would otherwise be a six-to-nine-month build into weeks.
Claims Router
Coverage interpretation and claim routing
DocuMage
Repair estimate, damage photo and correspondence extraction
Multi-Agent Orchestrator
Cross-task settlement orchestration
Anomaly Detector
Fraud and leakage detection
The architecture
The platform runs on the insurer's AWS environment in the London region, fully inside their existing UK-data-residency boundary. The deployment is on EKS with autoscaling configured around the typical UK motor-claim FNOL pattern — sharp peaks on Monday mornings (weekend incidents) and after the autumn clock-change.
DocuMage runs a layered extraction pipeline. For the highly structured documents (body-shop estimates in the standardised UK industry format, police-report templates) a deterministic template-aware extractor handles the bulk of the work, with LLM extraction for the free-text fields. For unstructured documents (third-party correspondence, policyholder statements, damage photos) the LLM-based extractor is the primary engine, with structured-output constraints to ensure the extraction maps cleanly into the claims data model.
The Claims Router is the coverage-interpretation engine. It consumes the insurer's policy wording library (the master wordings plus all endorsements applicable to the specific policy in question), the claim particulars, and produces a structured coverage opinion — what is covered, what is excluded, what excess applies, what no-claim-discount implications attach, and where the wording is genuinely ambiguous (in which case the claim is routed to a human handler with the ambiguity flagged for explicit decision).
The fraud layer is a hybrid: a deterministic rules engine for the bright-line cases (claims falling within configured fraud-pattern triggers — recently inceptied policies, multiple claims in a short window, recurrent claim circumstances), and an Anomaly Detector model for the subtler behavioural patterns. Suspicious claims are not declined by the automated pipeline; they are routed to the insurer's specialist fraud-investigation team with the fraud-score reasoning attached.
Every automated decision is logged with the policy wording cited, the documents reviewed, the model versions used and the orchestrator's decision graph. This audit trail is the evidence the insurer's compliance team uses for FCA Treating Customers Fairly oversight, and is directly queryable by the insurer's internal audit team.
The numbers behind the story
80% of in-scope motor claims now settle straight-through with no human intervention beyond the initial FNOL. Median time-to-pay for these claims is 24 hours; the 95th percentile is 36 hours. Claims that fall out for human review do so with the orchestrator's work-product (the parsed documents, the coverage opinion, the fraud assessment) already assembled, so the handler is reviewing rather than starting from scratch.
Claims-handler capacity has been freed up by approximately 42%. The insurer's COO has redirected this capacity to two priorities: complex-claim handling (handlers now spend significantly more time per complex case, which has improved settlement outcomes on the cases that genuinely require judgement) and third-party-liability negotiation (a higher-value activity the handler team had previously been unable to invest in adequately).
Customer NPS on the motor-claims journey has risen from 6.1 to 9.4 on the insurer's measurement scale. The customer-experience research team's analysis is that the speed of settlement is the dominant driver — policyholders who expected a multi-week claims process are getting paid the day after they reported the claim, which is a meaningful experience inversion.
Fraud detection has improved, not degraded. The Anomaly Detector layer has identified categories of fraud that the legacy rules-engine systematically missed, particularly around staged-collision patterns that rely on the chronology of multiple unrelated-looking claims. The insurer's fraud-investigation team's referral acceptance rate is up.
“Two prior attempts had failed at the same problem. MindMap delivered 80% straight-through settlement in twenty-two weeks, with the explainability our FCA compliance team needed, and a customer NPS uplift we frankly did not expect. Our claims handlers are now doing the work they trained for instead of chasing documents.”— Chief Operating Officer· UK General Insurer
Why MindMap was chosen
The insurer had two prior failed attempts at this problem. The first, with a major insurance-tech vendor, stalled at the document-intelligence layer — the vendor's template-based OCR could not handle the variety of repairer-estimate formats and the fraction of claims that routed cleanly through automation was too small to justify the project's cost. The second, in-house, made similar progress on the orchestration layer but did not have the document-intelligence depth to make it stick.
MindMap won the bid on the strength of the DocuMage accelerator being already-deployed in UK motor-claims contexts and on the willingness to take fixed-fee accountability for the document-intelligence accuracy that had defeated the previous attempts. The parallel-run model gave the insurer's CRO and CCO direct evidence of the platform's actual performance on actual UK motor claims before committing to cutover.
Our delivery team included two former motor-claims operations leads from comparable UK insurers, which the insurer's claims leadership felt was critical — the platform was being designed by people who understood the claims-handling workflow, not just by ML engineers.
Related deployments
Sovereign WhatsApp Banking
ChatNext-powered WhatsApp bot deployed inside the bank's air-gapped data centre, handling balance, transfers, statements and loan applications in English and Swahili.
Cheque OCR at 99.1% Accuracy
DocuMage replaced a legacy template OCR for cheque clearing, processing 10,000 cheques per day at 99.1% field accuracy with 94% straight-through processing.
UK Challenger Bank KYC
OnboardX rebuilt the KYC pipeline with liveness, sanctions and PEP enrichment in a single STP flow, collapsing onboarding from 5 days to 4 hours.
Want an outcome like this?
Start with a 2-week AI Readiness Sprint. We deliver a prioritised use-case backlog and business case grounded in what's actually buildable with our accelerator library.