Finovox deploys AI fraud detection for Admiral's French arm

Admiral Group's French subsidiary L'Olivier Assurance has deployed document-verification software from Paris-based Finovox to counter a rise in falsified insurance claims, in a partnership that illustrates how generative AI is simultaneously powering fraud and the defences against it.

L'Olivier, which launched its fraud-detection initiative in 2023, serves more than 600,000 customers entirely through digital channels. That fully digital model, a source of competitive advantage in pricing and customer journey, also creates a wider attack surface for document fraud than a branch-heavy insurer would face. The company selected Finovox, founded in 2019, whose proprietary AI analyses submitted documents for signs of retouching, alteration, and internal inconsistency, routing high-risk files to human reviewers rather than attempting fully automated decisions.

The scale of the fraud problem

The backdrop is not confined to France. The Association of British Insurers estimates that £1.16 billion in fraudulent general insurance claims were detected across the UK in 2024, a 12% year-on-year rise in case volume. Motor lines accounted for 53% of that total, and the ABI separately estimates that a comparable volume of fraud goes undetected annually, implying a true industry cost roughly twice the reported figure. Exaggerated-loss claims alone totalled £466 million. Those losses are not absorbed by insurers: they are redistributed to honest policyholders through higher premiums, making fraud a systemic pricing problem as much as a financial one.

Julien Bouverot, Director General of L'Olivier Assurance, pointed to exactly that mechanism when describing the results achieved in 2024: "Fraud benefits a minority while costing everyone: it drives up the overall level of premiums, and ultimately it's policyholders who end up unfairly footing the bill. These positive results have also allowed us to shield our customers from a price increase in 2025 and we intend to do the same in 2026."

The release does not disclose specific fraud-detection rates or financial savings figures, which limits independent verification of the claimed impact.

Convergence: the generative-AI arms race in financial services

The broader strategic context is one that every insurer, bank, and lender with a digital-first model is navigating. Generative AI has dramatically lowered the cost of creating convincing falsified documents, payslips, vehicle registration certificates, repair invoices, meaning that fraud attempts which once required specialist skills can now be industrialised at scale. That same generative-AI capability is being absorbed into the detection stack, creating an iterative arms race between fraudsters and verification vendors.

For the insurance sector specifically, this dynamic is accelerating a structural shift in underwriting and claims infrastructure spend. Document-verification and identity-proofing tools, once considered back-office compliance costs, are increasingly being positioned as pricing levers: the L'Olivier case is notable precisely because management is linking fraud containment directly to premium stability rather than treating it as a siloed risk function.

The capital landscape for fraud-prevention AI is correspondingly active. Finovox's client roster, which includes Allianz Direct, BpiFrance, Orange Bank, and PwC alongside L'Olivier, suggests the technology is crossing the traditional boundary between insurance and banking, a convergence point worth watching as open-finance regulation in Europe continues to blur those sector lines.

For macro investors assessing insurtech and regtech valuations, the sharper question is whether AI-native document verification becomes a commodity utility embedded in cloud-based policy-administration platforms, or whether it sustains a specialist-vendor margin as fraud techniques continue to evolve. L'Olivier's decision to partner rather than build in-house hints at the latter view, at least for now. The next inflection will come when generative-AI fraud tools are sophisticated enough to mimic metadata and cryptographic document signatures, at which point the verification layer will need to move from visual-pattern analysis toward deeper cryptographic provenance checks.