Aviva Put 80+ AI Models Into Claims and Reported £90 Million in Savings to Investors

Aviva plc rebuilt its claims operation around more than 80 artificial intelligence models and published the outcome in the most scrutinized document a public company produces: its annual results. The 2025 full-year announcement, released on 5 March 2026, sits alongside an annual report recording over £90 million in claims cost savings, a 23-day reduction in liability assessment time for complex cases and a 65% drop in customer complaints. CEO Amanda Blanc cites artificial intelligence among the drivers of a Group operating profit that rose 25% to £2.2 billion.

Complex claims moved at paper speed

Aviva is the UK's largest general insurer, a FTSE 100 group serving customers across the United Kingdom, Ireland and Canada. Before the transformation, the motor claims journey ran on manual judgment at every step. Assessing liability in complex cases — multi-vehicle collisions, disputed fault, personal injury — required weeks of file review. Routing decisions determined which team handled each claim, and misassignments sent complex cases to generalist handlers and simple cases to specialists, wasting capacity on both ends. Customer complaints tracked that friction. In life underwriting, a parallel bottleneck: underwriters read GP medical reports that sometimes exceed 90 pages to extract a handful of decisive facts. The case for AI rested on a structural insight: a claim generates data at every stage, from the first notice of loss to the final settlement, and each stage offers a measurable decision to improve. McKinsey's documentation states the ambition plainly: Aviva set out to build the UK's leading claims operation, with AI improving outcomes at every step of the process.

The decision that made it possible: saturate one journey, then earn adoption

Many enterprise AI programs start with a single hero use case. Aviva chose breadth. Working with QuantumBlack, McKinsey's AI arm, the insurer built more than 80 AI models reflecting the specific needs of its claims teams and embedded them across the entire motor claims journey. The human investment matched the scale of the model portfolio: a team of more than 50 data scientists, engineers, business leaders, change professionals and translators, working across six dimensions — strategy, talent, agile operating model, technology, data, and adoption and scaling. The most revealing line item sits far from the models themselves: Aviva invested more than 40,000 hours of training to build data and AI skills across the claims organization, treating adoption as an engineering discipline rather than a communications exercise. The same philosophy governs the underwriting tool launched in late 2025: it condenses long medical reports into concise summaries, and Aviva's underwriters review each summary and make the final decision. The company put the tool through eighteen months of testing and 1,000 cases in an active pilot phase before rollout. CEO Amanda Blanc framed the strategic position in the results announcement: “We have clear strengths in artificial intelligence which are creating major opportunities to transform claims, underwriting and customer experience.”

The result — with full context

The operational numbers are documented in McKinsey's published case study: the average time needed to assess liability for complex cases fell by 23 days, claims routing accuracy improved by 30%, customer complaints dropped by 65%, and customer satisfaction measured by total net promoter score climbed more than seven-fold. The financial layer sits in Aviva's own investor reporting. The Annual Report and Accounts 2025 records over £90 million of claims cost savings delivered through the transformation, alongside materially improved customer experience. The full-year 2025 results announcement reports Group operating profit of £2.2 billion, up 25%, with Blanc citing “significant claims indemnity benefits and pricing sophistication through AI models” and “early success with claims summarisation and medical underwriting tools”. The underwriting tool, live since 28 November 2025, reduces review time per case by around 50%. Full context requires two clarifications. First, McKinsey served as Aviva's transformation partner, so the operational metrics originate from a party invested in the story — readers should weigh them accordingly. Second, the results announcement itself keeps its AI language qualitative; the quantification lives in the annual report and the case documentation. The strength of this story flows precisely from that layering: a consultant's case study makes claims, and an audited annual report — signed by the CEO and scrutinized by regulators, auditors and markets — confirms the financial substance behind them.

What other organizations can learn

Three lessons transfer. First, a portfolio beats a pilot. Aviva compounded 80+ models inside one domain — motor claims — before expanding, and each model improved the data and decisions available to the next. That compounding arrives when models share a single domain and a single accountable owner. Second, adoption is a budget line, sized like one. Forty thousand hours of training is a commitment measured in millions; it converted a claims workforce into daily users of model output and produced the complaint reduction that pure model accuracy leaves on the table. Third, the gold standard for AI ROI verification is investor-grade reporting. Vendor press releases announce intentions; case studies describe projects; annual reports carry legal weight. When a CEO attributes part of a 25% profit increase to AI in an audited document, boards evaluating their own programs gain a benchmark they can trust. The replication conditions are clear: a data-rich, end-to-end process owned by one accountable team; human decision rights preserved at the final step; and a multi-year commitment that outlasts budget cycles. Insurers sit closest to Aviva's playbook, and any organization running high-volume, judgment-heavy processes — banks, utilities, healthcare payers, logistics — can apply the same sequence: pick one journey, saturate it with models, train the people, and report the results where auditors look.