AI is making insurance pricing models faster to build, but the time between validating a model and applying its output to customers can still stretch into months.
An insurance pricing model can be recalculated in minutes, but that does not necessarily create value if its output takes months to reach the market. Earnix director Mathieu Edmond argues that the gap between model development and deployment remains a significant challenge for insurers as AI and automation accelerate technical work.
In one example highlighted by Edmond, a pricing model was validated by a technical committee in February, but the resulting price was not applied to policyholders until October.
The model itself may not have changed during that period. The market around it, however, may have.
Customer behaviour, portfolio composition and other market conditions can evolve between the point at which a model is validated and the point at which its output is deployed.
The delay is not necessarily caused by the modelling process. Validation, IT integration, controls, testing and the trade-offs involved in pricing decisions can all contribute to the time required to move from a model to a live price.
AI and automation can reduce the time required for modelling, documentation and analysis. They do not automatically remove the constraints surrounding data, validation, deployment and governance.
This creates a different challenge for insurers. The risk is no longer simply that a model is not good enough. There is also a risk of assuming that a strong model will automatically result in a good business decision.
A model can estimate the probability of a loss, predict lapse or recommend a price. It cannot determine how that recommendation should be balanced against risk appetite, profitability, competitiveness, fairness and compliance.
Those decisions require organisational judgement. For some insurers, pricing decisions involve actuarial teams, business lines, distribution, compliance, IT and governance bodies. Each may have different requirements or priorities before a pricing change can be implemented.
The complexity can increase for mutual insurers, bancassurers and groups operating across multiple distribution channels, where pricing decisions can involve a wider range of stakeholders.
Governance is therefore not simply an obstacle to faster pricing. It is part of the decision-making process. The question for insurers is whether those processes can keep pace with the speed at which models and recommendations can now be produced.
Data can create another limitation.
Earnix’s 2026 Industry Trends Report, The Race to Reinvent surveyed 400 insurance executives globally. It found that just 30% said their organisations can quickly obtain the information needed to make business decisions.
Meanwhile, 46% said their technology provides the speed required for effective decision-making, while two-thirds identified poor data quality as a factor that slows decisions and limits AI effectiveness.
The findings highlight the importance of looking beyond the model itself.
Improving the speed of modelling does not necessarily solve problems with the data needed to support a decision or the processes required to act on the result.
AI governance is another consideration. The Earnix research found that 92% of respondents conduct formal AI governance reviews regularly. However, fewer than one in three said they were fully confident that these reviews keep pace with evolving regulatory requirements.
Regulatory and legal exposure was also the main ethical concern around AI deployment for 38% of respondents. At the same time, insurers do not appear ready to remove people from the process entirely.
Some 56% of executives said they favour a gradual approach that retains human intervention for at least the next three years. As more technical pricing tasks become automated, this could change the role of actuaries.
Instead of focusing primarily on producing models, actuaries may increasingly be involved in designing and governing the decision systems around them. That includes determining where automation should be used, where human oversight is required, how decisions should be explained and how conflicting objectives should be managed.
The shift does not make the underlying model less important. Instead, it places greater emphasis on what happens after the model has produced its output.
For insurers, the ability to deploy models quickly may become as important as the ability to build them quickly. That requires more than modelling technology. It depends on data, integration, validation, governance and the processes used to turn model outputs into decisions.
As AI becomes more embedded in insurance pricing, the distinction between model development and decision-making is likely to become increasingly important. A model can produce an answer quickly, but insurers still need to determine whether that answer is appropriate, how it should be used and who is responsible for the decision.
For Earnix director Mathieu Edmond, the challenge is therefore not simply making pricing models faster. It is ensuring that the wider decision-making process can make effective use of what those models produce.
Copyright © 2026 InsurTech Analyst
