How will home insurers using AI replace those without AI?

The real debate is therefore not one of segmentation versus risk pooling, but rather one of risk selection (pro-selection versus anti-selection).

Nor is the debate a struggle between “virtuous” insurers who practice risk pooling without AI and “greedy” insurers who segment using AI. We can paraphrase this by betting that data and AI will not replace insurance, but assert with near certainty that insurers using AI will replace those without it.

The impact of granular risk knowledge is indeed significant for players in the home insurance sector. Until this granular knowledge is acquired, the multitude of weather events will alternately result in pleasant or unpleasant surprises, depending on the randomness of the underwritten risk mix in the affected areas.
Conversely, a granular knowledge deficit inevitably raises the question of how to measure the pooling of risks within a portfolio across a given territory. Each player must address this question in light of its own characteristics and underwriting policy.

If this knowledge is held only by a few market players, a phenomenon of pro-selection seems inevitable, as these players will inevitably select the most resilient properties, particularly in high-risk areas, if only to comply with any new risk-pooling rules imposed by public authorities. Other players lacking this knowledge will, without realizing it, face adverse selection regarding the most vulnerable risks in these sensitive areas. This scenario, leading to significant asymmetry among insurers, is of course undesirable for market equilibrium, as it calls into question the very principle of insurance and the insurability of a risk in a given sector or for specific buildings.

New entrants also pose a significant threat because, thanks to AI and risk selection, they can refine their models by attracting the right risks through more tailored rates or terms, thereby capturing market share that is vital to the balance of existing portfolios. These new entrants can focus on a segment of low-risk customers using only external data (without historical data) and thus threaten the balance of established insurers’ risk pooling arrangements. Those who were paying more than the cost of their risk are thus encouraged to be pooled with others who are less risky in order to pay less, while maintaining higher profitability than the new entrant. This danger is exacerbated by the financialization of the sector and the pursuit of easy, immediate profits at the expense of insurance risk pooling and the principles of solidarity.

New players for whom insurance is merely a gateway to a broader range of services can also leverage AI and external data to offer insurance products (whereas a lack of historical data, limited understanding of risks, and the sector’s low profitability previously prevented them from doing so). Past experience and the market share gained by new entrants, namely bancassurance companies, in the 1970s and 1980s demonstrate that the danger is very real.

Finally, the highly restrictive insurance regulations that previously protected existing players could, on the contrary, with new data and AI provisions, penalize them and favor new entrants (insurers, comparison sites, brokers, and distributors). The FIDA regulation and the requirement for established insurers to share premiums and coverage data perfectly illustrate the loss of value of established insurers’ experience data and could “unduly” favor disruptors.

The government’s desire to protect consumers is certainly commendable, but it can unfortunately lead to unintended consequences due to a lack of understanding of insurance techniques.

While this failure to utilize data and AI severely undermines competitive positioning, it also weakens the customer management value chain.

First, a more detailed and granular understanding of risk allows coverage to be tailored to the policyholder’s needs. In this case, AI enables the insurer to meet the exact coverage needs of each policyholder by taking into account the actual and recent characteristics of the insured building. The insurer can simultaneously adjust the premium, protect itself from a similar competing offer by increasing the policyholder’s satisfaction, and reduce the risk of significant dissatisfaction or litigation in the event of a claim.

External data also helps streamline the underwriting process, particularly by reducing the number of questions asked by the insurer while minimizing errors. In an online market where competition is fierce and the time it takes to purchase an insurance policy is measured in minutes, external data becomes essential for reducing the amount of information that needs to be collected.

The more automated an underwriting process is, the more crucial external information about the insured property becomes for identifying potential non-standard risks, organizing potential underwriter intervention for a more precise identification of the property, and considering appropriate pricing.

Finally, the problem of a premium that does not match the risk of a building cannot be solved solely by increasing the premium, terminating the policy, or refusing to issue the contract. Reducing the cost of coverage to make the premium acceptable may require modifying the contract to make it affordable, and there is a range of well-known solutions (deductibles, coverage limits, etc.).

Finally, one of the operational risks to watch for is the ‘blind’ use of AI. Indeed, without a strategic framework, AI tools can:

• Produce ‘black box’ segmentations disconnected from the risk-pooling strategy;
• Underwrite risks blindly based on “hallucinations” that pose a major risk to the insurer;
• Promote a purely technical and financial approach;
• Strip insurers of their control by entrusting their data and decisions to third parties or software tools incorporating AI.

The adoption of AI and future uses of Agentic AI require a deeper understanding of the models used (in underwriting, pricing, and adapting the product) and the risks covered (with the finest possible granularity). To put it simply, if there is a multitude of coverage options and rates, it must be possible to determine why and how that coverage and rate were applied and, if necessary, to explain and defend the model before the regulator.

Building-level data enables more granular segmentation within socially and strategically acceptable limits. It can thus help limit uninsurability to only the buildings in question by assessing the risk of each individual home, rather than excluding or withdrawing coverage from an entire street, municipality, or region. Building-level data then makes structures insurable even when located in areas considered uninsurable, by allowing insurers to identify resilient homes in high-risk zones.

Having the most granular possible data, attributes, and risk scores (at the building level) remains the sole solution to truly manage portfolio balance and risk pooling with full knowledge and in accordance with the insurer’s strategy and principles (risk appetite, degree of solidarity, etc.).

Building data is no longer an optional advantage: it is the foundation on which the technical balance of home insurance portfolios rests. Without it, an insurer is exposed to invisible adverse selection, the loss of its best risks, and a vicious pricing cycle that is difficult to break. With it, an insurer can manage its risk pooling with full knowledge, adapt its underwriting, and maintain profitability without sacrificing its societal role.

To go further, discover our ebook“How can AI and data save the insurance industry? The example of home insurance in France”, written by Addactis experts.