Personalization, price and fairness

A line caught my eye in a shareholder letter from the US insurer Root: the number one reason a customer chooses an insurer, and the number one reason they leave, is price. I lingered on those lines, because everywhere in the world insurance comes back to the same place: price. But in insurance there are really not one but two separate pricing problems.

Two pricing problems

The first is matching price to risk. This is the domain of actuarial science, data and models; a closed system, clear mathematics. As a company grows, the data gets richer, the model improves, the price moves closer to the risk. This is where the industry has invested most over the past twenty years. Digital technologies have raised this precision like never before: risks separate more clearly, good risks are rewarded.

The second is talked about far less: matching price to the person. Two customers with the same risk profile may not be at the same place in life at the same time. One’s job ended last month; another is rebuilding their budget for a newborn; another overspent last month and is cutting back this month. All three look at the same policy and see the same price. Yet the meaning of price isn’t the same for everyone. Person-based pricing, unlike risk, is open-ended and fluid; it has no single right answer, it has context. This is most likely where AI could be truly disruptive in insurance: a pricing that reads not the customer’s profile, but their current state.

This second problem is AI’s biggest opportunity and its most sensitive limit at once. A pricing that can read the customer’s current state, set up right, offers people accessible protection at exactly the moment they need it. Set up wrong, it becomes a mechanism that uses the customer’s weak moment to raise the price. The difference lies in intent and transparency: the same technology can both protect and exploit.

The appeal and the risk of behavior-based, dynamic pricing are both here. Data-fed models can truly understand the customer and produce fairer prices; but the same models, when they become unexplainable and unauditable, quickly burn through trust. When the customer asks why their price is what it is, if there’s no defensible answer, personalization is not an advantage but a reputational risk.

A concrete example clarifies this second problem. Two drivers of the same age, same car, same area may technically carry the same risk; but for one the annual premium is a comfortable line item, for the other a burden that strains the end of the month. Traditional pricing sees both the same. Yet retaining or losing a customer is often hidden not in the risk, but in this context. As Root also points out, the customer’s reason for coming and going is ultimately price; but the meaning of price varies from person to person.

We mustn’t forget personalization’s positive face either. A well-designed model can reward good behavior: it can offer a fairer price to the safe driver, to the customer who lives healthily. Here price becomes not a punishment but a tool of incentive. The issue isn’t personalization itself, but whether it’s transparent and explainable. If the customer can understand why their price is what it is, personalization builds trust; if they can’t, the same tool quickly tears it down.

The limit of personalization: fairness and access

But personalization has a limit. Fully personalized risk-based pricing can make insurance expensive, even inaccessible, for some individuals. Especially in health, disaster and compulsory insurance this is no longer technical but a social matter. Measuring risk perfectly doesn’t mean everyone can be protected; on the contrary, it carries the risk of pushing those who most need protection out of the system.

At this point the balance between insurance’s mathematics and public benefit must be rediscussed. Which risks should stay inside the market, which should be supported by public mechanisms? The answer isn’t only insurers’ responsibility but also that of regulators and society. Because at the heart of insurance is the spreading of risk across a pool rather than a single person, that is, solidarity. Excessive personalization can quietly weaken this solidarity.

On the fairness side, history offers ready examples. Pool mechanisms in disaster insurance, community-based pricing in health, state support in compulsory lines; all are attempts to manage the tension between pricing every risk individually and leaving no one outside. In a country like Turkey with high earthquake and flood risk, this balance isn’t theoretical but extremely concrete.

At the heart of insurance is the spreading of risk across a pool rather than a single person, that is, solidarity. Excessive personalization quietly weakens this solidarity: if everyone pays only the price of their own risk, the most fragile fall out of the system. Yet insurance’s social value lies precisely in being able to protect these fragile ones.

In Turkey this balance is already written into the legislation. Compulsory lines like the mandatory earthquake pool (DASK) and motor third-party liability are solidarity mechanisms that gather risk in a pool rather than dumping it on each individual. In a country where earthquake and flood risk is this high, keeping those who most need protection from being pushed out of the system isn’t a technical but a social necessity. When excessive personalization weakens these pools, the loser isn’t only the individual, but the whole system.

There’s one more dimension: willingness to share data. Person-specific pricing depends on the customer sharing their data; and the customer does so only when they see fair value in return. So personalization must be not a one-sided data collection but a mutual agreement. If the customer can say I gave my data and in return got a better price and a real benefit, the system is sustainable; if they say they took my data and I’m still in the same place, trust collapses.

So AI’s real test in pricing is not technical but moral. Building the model is no longer hard; what’s hard is keeping it both sharp and fair, both personal and transparent. What will carry insurance into the future isn’t the institution that does the most aggressive personalization; it’s the one that can balance personalization with solidarity. Because setting the price right is mathematics, while making the price legitimate is a matter of trust.

Regulators, too, are drawing this line ever more clearly. Explainability, non-discrimination and the limits of data use have become the compliance-requiring, not technical, side of pricing models. So keeping personalization transparent and accountable isn’t just an ethical choice, it’s increasingly a necessity. The industry’s test in the coming years won’t be building the smartest model; it’ll be being able to explain that model to the customer, the regulator and society.

That’s why I don’t see pricing as only an engineering problem. It contains psychology, ethics and public benefit as much as mathematics. Measuring risk correctly is necessary; but a pricing that looks after the individual and society is the sustainable one. In the end insurance is a business of trust, not numbers; and once trust is lost, no model can bring it back.

Conclusion

Matching price to risk still matters, but on its own it’s no longer enough. Digitalization brings back onto the agenda not only price but the definition of fairness. My stance is this: personalization is a powerful tool, but applied blindly it erodes the solidarity at the heart of insurance. Good insurance is insurance that sharpens the mathematics while also looking after access and fairness. Whoever doesn’t think about the two pricing problems together will, sooner or later, lose either their profit or their legitimacy.