AI, behaviour, and trusted data are rewriting how insurance works
At a recent event in London, Vitality AI and Google outlined a simple idea with big consequences: pair artificial intelligence with behavioural science and high-integrity data, and you can change health outcomes at scale. The goal is early disease detection and personalised nudges that people actually act on-then fold those results back into how products, pricing, and engagement get built.
As Emile Stipp, Managing Director of Vitality AI and Global Chief Actuary at Discovery, put it, AI should live inside the business model, not sit beside it. The intent isn't just to cut costs; it's to help people get healthier-and to align incentives so insurers, employers, and members all win.
Beyond efficiency: make AI the product, not a side project
Most teams deploy AI to automate back-office work. Useful, but limited. The bigger opportunity is to point models at target variables that matter-health outcomes, screening uptake, sustained activity-then let those insights guide product rules, pricing signals, and engagement flows.
That shift enables actions that were hard a few years ago. Think personalised health prompts, timely risk alerts, and context-aware coaching built straight into the policy experience. Relevance scales, cost of engagement falls, and the risk pool improves.
Build on trusted data or don't build at all
Behind the scenes, the advantage comes from data integrity. Vitality has spent more than two decades building architecture and standards that determine which data is fit for modelling. A medallion structure rates trust from bronze to silver to gold, and only gold-level data enters production models.
Lineage, cross-referencing, and explainability are non-negotiable. Every recommendation is traceable: click to see why a model advised a screening, flagged a drug-food interaction, or adjusted a goal. That transparency improves decisions and builds trust with clients, regulators, and partners.
Personalisation that moves the needle
Personalisation works when it's practical. Example: warning a member that grapefruit conflicts with statins, or suggesting sleep improvements based on real patterns-not generic tips. Engagement has exceeded expectations within months because the advice is timely and specific.
Channel fit matters too. In some markets, WhatsApp is the best route; in others, iMessage. Goals differ by region, culture, and baseline activity. Activation first, then adapt. One current push: lift cancer screening participation from roughly a third to two-thirds in a year-a change backed by clear, personal prompts and simple next steps. For context on national screening, see the NHS overview of programmes here.
Responsive systems: from person to employer to population
Lifestyle signals proved their value during COVID-19; people who exercised tended to have better outcomes. With continuous feedback loops, models learn from behaviour, behaviour shifts from better prompts, and the cycle compounds.
Employers benefit from anonymised insights: mental health trends, engagement patterns, and cohort-level risk indicators. Reports adapt to group size with strict privacy thresholds, giving HR leaders actionable guidance without exposing individuals.
Health economics meets habit formation
GLP-1 weight-loss drugs are changing care pathways, but behaviour still decides durability. Data shows better outcomes when weight loss is paired with exercise. That's why incentive designs are moving to "medication plus activity" rather than medication alone.
The insurance lens is straightforward: pay for what moves long-term risk. Reinforce habits until they become self-sustaining. Once a habit sticks, it's easier to repeat tomorrow.
What insurers can do next
- Make AI a product decision engine. Choose target variables tied to health and risk (e.g., screenings completed, weeks of sustained activity).
- Enforce data tiers and lineage. Restrict production models to data that meets "gold" criteria with clear provenance.
- Embed explainability. Require click-through reasons, audit trails, and human-readable outputs by default.
- Fit the channel to the member. Test WhatsApp, iMessage, email, and app notifications; measure response by segment.
- Set contextual goals for screening, exercise, and sleep. Start with activation, then iterate market by market.
- Close the loop with employers. Share anonymised trend reports and recommended actions with strict privacy guards.
- Link incentives to sustained behaviour. Tie medication support, premium adjustments, or rewards to proven activity and adherence.
Partnerships that accelerate the mission
Collaborations like the one between Vitality AI and Google Health help move from intent to implementation-especially for early detection and personalised guidance at scale. For insurers, the practical playbook stays the same: reliable data, explainable models, and incentives that make healthy choices the default.
If you're upskilling insurance teams to apply AI across product, pricing, and engagement, explore curated AI learning by role here.
The takeaway
AI shouldn't just predict risk; it should help people change it. With trustworthy data, clear feedback loops, and behaviour-first design, insurers can ease pressure on healthcare systems and deliver better client outcomes at the same time.
Your membership also unlocks:
