What an Agentic Simulator can tell an insurer

An insurer considering a health programme has several decisions to make. Who should receive the offer? What support should it include? What reward would make participation worthwhile? How will the insurer know whether the programme made a difference?

Our Agentic Simulator is being developed to make those decisions easier to examine before committing budget.

Start with people who behave differently

The simulator’s intended role is to represent a population of synthetic members with different circumstances and responses. Some may accept an offer. Others may ignore it, participate briefly or change their behaviour without an incentive.

Those differences are central to the question an insurer is trying to answer. A programme can attract participation while spending heavily on activity that would have happened anyway.

Compare a concrete decision

Imagine an insurer exploring a programme that rewards members for completing a blood-pressure check.

One option is to offer it broadly. Another is to make a more focused offer to a relevant group, using information members have permitted the insurer to use. A third option is to continue without introducing the programme.

The model can be structured to compare invitations, participation, reward costs and follow-through under explicit assumptions. Further outcomes depend on what happens after the check and on the evidence supporting each link in that pathway.

This makes the decision more specific: what would have to happen for each option to be worthwhile?

Make uncertainty visible

An agentic simulation still depends on assumptions about the world. Giving synthetic members more detailed behaviours does not, by itself, establish that those behaviours match real people.

Our approach is to make assumptions inspectable, compare alternative scenarios and identify which uncertainties most affect the decision. If the recommendation changes sharply when participation falls, that is something the pilot should investigate.

HealthID’s modelling work uses synthetic populations. Modelled outcomes are scenarios, not observed customer results. Real participation and behaviour must be measured in a live programme, with an appropriate comparison to understand what the programme changed.

The useful output is a clearer decision and a better-designed test: who to approach, what to offer, what to measure and what would justify the next investment.