In September 2026, AXA’s official presentation of its 2027–2029 strategic plan, Growing Forward, made artificial intelligence one of the most explicit value-creation levers in the group’s next phase of development.
At first sight, another major insurer intends to use more AI. In fact, this goes further. AXA has attached a recurring pre-tax value target of €500 million to €700 million by 2029 to its AI programme, while aiming for 80% of employees to use AI regularly. Those figures sit inside a wider financial framework that includes 7% to 9% annual growth in underlying earnings per share, a 15% to 17% underlying return on equity and around €25 billion of cumulative cash remittances over the plan period.
This matters because it connects AI adoption to group performance rather than treating it as a separate technology agenda. The plan therefore gives boards and executive teams a useful benchmark: an AI strategy can be assessed through economic contribution, workforce adoption and changes to core insurance processes, not merely through the number of tools deployed.
From experimentation to a quantified portfolio
AXA’s use cases cover several points in the insurance value chain. The group highlights AI-assisted distribution and sales coaching, support for customer calls and email responses, pricing and underwriting, claims handling, and broader process automation. It also refers to agentic AI, signalling an ambition to move beyond tools that generate content towards systems able to execute parts of a workflow under defined controls.
The €500 million to €700 million target is especially important. It implies that use cases will need to be managed as an investment portfolio, with common methods for establishing a baseline, attributing benefits and avoiding double counting. Faster handling time, improved loss ratios, higher conversion and lower operating costs are economically different outcomes. They require different measures, owners and validation methods.
For boards, the central question is therefore not whether AI creates activity but whether management can demonstrate repeatable value. A credible dashboard would distinguish realised benefits from capacity released, show the cost of models and controls, and track whether productivity gains are retained, reinvested or passed on to customers.
Adoption becomes an organisational measure
The target of 80% regular employee usage is equally revealing. It recognises that value depends on behaviour at scale. Yet usage alone is not evidence of impact. An employee may open an assistant frequently without changing a decision, improving a customer outcome or reducing processing time.
The more demanding management task is to combine adoption data with workflow-level indicators. In underwriting, that could mean measuring submission triage time, referral quality and portfolio performance. In claims, it could include cycle time, leakage, customer satisfaction and the frequency with which human handlers override AI recommendations. In distribution, it could mean conversion, retention and the quality of advice rather than the volume of generated messages.
This is also where training, job design and governance converge. Employees need to understand when a model may assist, when a decision must remain human and how to challenge an output. Managers need authority to redesign processes, not simply add an AI layer to existing steps. Risk, compliance, data and technology teams must be involved early enough to make deployment both safe and usable.
A governance test for the insurance sector
AXA’s plan sets a visible test for the industry. If the group reports progress against its value and adoption targets with sufficient clarity, peers will face pressure to explain their own AI economics. If benefits prove difficult to isolate, the experience will be just as instructive: it will expose the gap between promising demonstrations and enterprise-wide transformation.
For global benefits organisations, the implications of AXA’s approach extend beyond property and casualty insurance. Many of the same patterns apply to multinational benefits governance: extracting information from local contracts, comparing terms, triaging exceptions, monitoring providers and preparing decision material for headquarters. The opportunity is substantial, but only when outputs are traceable, sensitive data remain protected, and expert review is built into the workflow.
The takeaway: AI becomes material when it is linked to explicit economic outcomes, embedded in operating processes and governed as a business capability. AXA has now put numbers against that proposition. The next question is whether execution and disclosure will make those numbers credible.
Share this:
- More [2]
