For decades, a penalized driver paid the price for his statistical category. Data and artificial intelligence are shaking up this system, without canceling it.
Get out of group judgment
A motorist with a high penalty has long had only one option: contact a specialized insurer like SOS Malusfor lack of an accessible alternative. His file was based on statistical affiliation: bonus malus, age, history with the former insurer. Nothing distinguished his actual conduct from that of his tariff category.
The models built by French insurtechs are reversing this trend. Rather than a fixed classification, they analyze dated and verifiable behavior. Braking, regularity of journeys, areas traveled, traffic schedules. The calculation does not pronounce any definitive sentence. It brings together factual elements which the insurer then chooses to retain or reject according to its own decision grid.
French insurtechs in action
This development remains concrete among several French companies, far from generic marketing discourse on AI. Shift Technology is the most visible example. The company became a unicorn after a fundraising of 220 million dollars. It helps insurers automate fraud detection through the analysis of textual and visual data.
Zelros follows a different path. The start-up equips advisors to guide underwriting decisions, without ever replacing them entirely. These tools process considerable volumes. Claims history, customer profiles, mobility data captured in real time. The stated objective remains the precision of the calculation, not the total automation of the commercial verdict.
Other French start-ups are banking on a similar approach. Assurly or Seyna are developing their infrastructure around native data processing. They thus avoid grafting it onto legacy systems. This architecture theoretically facilitates a more detailed re-evaluation of the profile, step by step of the customer journey.
The essential legal and regulatory framework
This technological shift does not exempt anyone from the already existing legal net. The Central Pricing Bureau (BCT) continues to guarantee a right to insurance, regardless of the score calculated upstream. No model, no matter how precise, can circumvent this obligation.
The General Data Protection Regulation (GDPR) also strictly regulates the use of this behavioral information. The Prudential Supervision and Resolution Authority (ACPR) requires insurers to document their models and actively monitor bias. A company can no longer simply point to a score to justify refusing coverage.
Governance and trust: the real selection criterion
For an investor or a manager in the sector, the stakes go far beyond the raw performance of the models. The real question concerns the governance displayed by each market player. Explainability of decisions, traceability of data used, ability to demonstrate the absence of hidden discrimination in the calculation.
The commercial discourse here deserves careful reading. The French Assurtech association confirms this in its own way. Of the applications received by its accelerator this year, nine out of ten present themselves as based on artificial intelligence. A manager of the association considers this figure indicative of a fashion effect more than a technical reality. The term has become so commonplace that it no longer guarantees much about the solidity of a model, nor its regulatory compliance.
The insurtechs that gain the trust of historic groups will not necessarily be the quickest to deploy a model. They will be the ones able to document each step of their process, rather than promising a fairer price without proving it. It is this criterion, more than the raw power of the model, which will determine the sector’s next partnerships.
