The promise was to do to insurance what others had done to taxis and hotels. Buy coverage in ninety seconds, be paid by an algorithm, pay a price cut by data. A decade later, the incumbents are still standing and many of the disruptors have vanished, been absorbed, or turned into something far more modest than their pitch.
A decade ago, a wave of start-ups raised dizzying sums on a simple intuition, insurance was a backward sector, slow, opaque, unloved, and all it took was an app and an algorithm to remake it. Underwriting would happen in an instant, claims would be settled in minutes by an artificial intelligence, price would stop being a crude average and become a faithful mirror of each person. The story was seductive, the capital poured in. A decade later, the landscape tells another tale, valuations collapsed after the euphoria, the players who promised to replace insurers often ended up working for them, and the factory of risk did not move. The question hanging over the sector is blunt, did insurtech lie.
The honest answer is no, it did not lie, it mistook the problem. It assumed insurance was a software problem, a bad interface waiting for a good app. But insurance is not a software problem, it is a capital and risk problem. Insurtech beautifully rebuilt the shop window and discovered the factory was untouched, because the factory, the capital that absorbs the loss, the actuarial pricing, the regulatory solvency, obeys constraints no interface dissolves. It rebuilt the visible, easy part, and ran into the invisible, hard one.
The error was understandable, because in most industries the software wave really had found the value in the interface. A taxi is a car plus a dispatcher, and the dispatcher was the weak, digitizable link. A hotel night is a room plus a middleman, and the middleman was the fat to be trimmed. The disrupters won because the thing being sold, the ride, the room, was simple and real, and only its distribution was broken. Insurance looked like the same setup, a simple product wrapped in a hateful interface. But insurance sells nothing simple or real, it sells a promise about an uncertain future, priced today, backed by capital that must still be there when the future arrives. There is no room, no ride, only a bet, and a bet cannot be growth-hacked.
One must distinguish two layers in insurance. The first is thin and visible, it is distribution, user experience, brand, the feeling one has while buying coverage and filing a claim. The second is thick and hidden, it is underwriting, the pricing of risk, reserves, regulatory capital, reinsurance. Insurtech excelled at the first, and there lies its real merit, it made buying fluid, the app elegant, the journey clear. But it discovered, at its own expense, that the first layer is neither where insurance is hard, nor where it makes money.
For one can acquire a million customers and lose money on every one of them. The unit that matters in insurance is not the download, it is the loss ratio, that quiet relationship between what you take in and what you pay out. A viral app says nothing about the quality of the risk it attracts, and a spectacular growth curve can be nothing but the accelerated accumulation of bad policies. The techniques born of the digital world, acquisition, growth, virality, know how to fill a book, they do not know how to say whether that book is good. The start-up's dashboard measured users, insurance is measured in losses, and the two numbers have almost nothing to do with each other.
Here the subject turns, and the analysis must go further than the note of a commercial failure. Insurtech's great pride, its cardinal virtue, was the removal of friction, buying in ninety seconds, with no heavy questionnaire, no agent, no delay. Yet in insurance, and in almost no other sector, the absence of friction is not a pure advantage, it is often a trap. For the friction of classic insurance did hidden work, it filtered. The tedious questions, the slowness, the intermediary, all of it discouraged, sorted, and quietly turned away part of the bad risks.
By removing the friction, insurtech removed the filter. The easier and faster a policy is to buy, the more it attracts precisely those an insurer should fear, the ones who know they are exposed and seek cover without being questioned. This is the mechanism of adverse selection, and it punishes smoothness. The ninety-second signup insurtech was so proud of was quietly loading its book with degraded risks, while its dashboards celebrated the conversion. Speed, in insurance, is never free, it can be adverse selection dressed as performance. What was a virtue everywhere else became, in the one sector insurtech chose to enter, a structural vice.
The same trap waited at the other end of the chain, in claims. Settling a claim in minutes by algorithm was the flagship promise, and paying fast is indeed easy. Paying correctly is the hard part. An algorithm that pays instantly without scrutiny is a beacon for fraud, and the friction insurtech scorned in claims handling, the questions, the checks, the delay, was again doing hidden work, deterring the opportunist and catching the inflated claim. Speed at the front door and speed at the back door remove the same filter, and a book that is easy to enter and easy to be paid from is a book built to be exploited.
An insurer is not only a brand and an app, it is a balance sheet, compelled by regulation to hold capital behind every risk it writes. Solvency rules demand that behind each promise stands enough capital to survive a very bad year. A start-up that carries the risk directly discovers that growth devours capital, because each new policy demands its own cushion, and the faster it grows, the more it must raise, at the mercy of a loss ratio it cannot yet predict. Growth, the start-up's oxygen, becomes in insurance a capital furnace.
Before this wall, most so-called full-stack insurtechs made the same retreat. Rather than carry the risk, they ceded it to reinsurers and kept only a fee, becoming underwriting agencies, fronting arrangements, distributors fitted with an algorithm. The move was rational and revealing, it amounted to giving up on being an insurer to become a supplier to insurers. The disrupter that claimed to replace the balance sheet ended up renting someone else's, and renting the balance sheet means letting the reinsurer, not the app, hold the real power, the power to withdraw its capacity at the first ratio that sours.
This is why the incumbents survived. Their moat was never their technology, often mediocre, but their balance sheet, their license, their decades of reserves, their reinsurance relationships, the regulator's trust. None of that yields to a better interface or an acquisition campaign. The insurtechs attacked the one part of the fortress that was undefended, the drawbridge, and left the keep untouched. They took the gate and thought they had taken the castle.
To this is added a more intangible barrier, trust. Insurtech wagered that a polished brand and a knowing tone would suffice to replace the trust incumbents had built over a century. But trust, in insurance, does not play out in the aesthetics of the app, it plays out at the moment of loss, in the certainty that the company will still be there, solvent, to keep its promise years after making it. A start-up with no track record and uncertain capital cannot promise that credibly, however beautiful its signup journey. One buys not only a policy, one buys the conviction that it will be honored, and that conviction is built over time, not in design.
Finally, one must reckon with the geography of law. Insurance is licensed country by country, each market demanding its own license, capital, regulator, and rules. The software dream, write once and deploy everywhere, shatters here, because one does not cross an insurance border with a click, one starts everything over, the license, the compliance, the reserves. The blitzscaling model that carried the digital platforms, win one market then the next at near-zero cost, meets here a regulatory wall that makes every expansion slow, costly and local. Insurance does not go global like an app, it is conquered territory by territory, like an insurer.
It would be wrong, though, to conclude it was pure illusion. Insurtech failed to conquer the heart of the trade, the bearing of risk, but it durably raised the bar at its edges. It forced settled, often sleepy insurers to modernize their distribution, to tend to their customer experience, to exploit at last the data they hoarded without reading. Many start-ups that dreamed of replacing insurers ended up selling them their tools, or becoming underwriting agencies backed by the capacity and reinsurance of others, producing risk without carrying it. The honest verdict is therefore neither that insurtech lied, nor that it won, but that it was right about the window and wrong about the factory, and that its lasting legacy is a better window bolted onto the same factory.
The wave's most solid success came, tellingly, from where it stopped trying to be an insurer. Embedded insurance, slipped inside another purchase, a rental, a ticket, a device, thrives precisely because it embraces being smart distribution backed by the capacity of an established risk-carrier. It no longer claims to rebuild the factory, it wires the shop window more tightly into the flow of commerce, and it was by giving up the original promise that it found a durable model. The lesson repeats, insurtech succeeds when it accepts its true place in the chain, the place of the link, and fails when it tries to occupy the place of capital.
There remains the question the next wave, that of artificial intelligence, puts back on the table. What if technology finally reached the factory. What if, this time, the models could select risk, price continuously, detect fraud, in short transform not the window but the workshop itself. Optimists see there the true revolution, the one insurtech announced too early and on the wrong floor. Skeptics recall that better pricing does not solve everything, because it meets a limit deeper than technology, and that limit is the very reason insurance exists.
There is besides a mundane obstacle the data dream keeps striking. The data that truly predicts risk is often not what an app can gather, and most of what would predict best, a person's health, their genes, their finances, their movements, is precisely what the law forbids an insurer to use, or what the customer refuses to surrender. The promise of pricing everyone by data therefore collides at once with a technical limit, the useful data is hard to obtain, and a political one, the most predictive data is exactly the data society has decided must not set your premium. The factory resists not only for want of capital, but because the law stands guard over what may enter it.
For an insurer that prices everyone perfectly has ceased to pool. Insurance works only because it does not bind the premium tightly to individual risk, because the lucky subsidize the unlucky, and that silent transfer is the protection itself. To push precision to its end would be to hand each person back exactly their own risk, and an insurance that pools no one is no longer insurance. The technology that truly penetrated the factory would therefore not save it, it would lead it toward a paradox, the more the insurer knows you, the less it needs to mix you with anyone, and the day it knows you perfectly, it has nothing left to sell you but your own reflection.
Insurtech did not lie, it taught a whole industry, at its own expense, the difference between the shop window and the factory. The lesson was costly, paid in billions of venture capital, and it carries an irony the AI wave will soon sharpen. The dream was not false, it was misplaced. Rebuilding the interface was useful but insufficient, reaching the workshop is necessary but dangerous, because at the end of the perfectly optimized workshop stands the dissolution of mutuality.
The arc is now legible. Through the second half of the 2010s, capital surged into the sector on the promise of disruption, valuations swelled, several champions reached the public markets at heights that assumed the victory already won. Then the loss ratios spoke. The listings that had soared came back down, some brutally, funding dried up, and the survivors reinvented themselves as something quieter, technology suppliers, agencies, partners of the very incumbents they had meant to bury. The cycle was not a scandal, it was an education, the market discovering in real time that acquiring insurance customers and profitably insuring them are two different trades.
The fate of the start-ups that promised to change everything thus became a parable about the nature of insurance. One can modernize access to risk almost without limit, one does not so easily modernize the act of carrying it, because carrying risk demands capital, regulation, and above all the acceptance of a share of ignorance, the chance one pools precisely because one cannot master it. Insurtech wanted to abolish that ignorance through data, and it discovered that abolishing it entirely would amount to abolishing insurance. The real question is therefore no longer whether it lied, but whether the industry has understood what it so clearly, and unintentionally, demonstrated.
There is a last reading, wider than insurtech itself. The decade taught that the digital playbook, remove friction, scale fast, personalize everything, does not merely fail in insurance, it inverts there, each of its virtues turning into a danger. Frictionlessness becomes adverse selection, scale becomes a capital furnace, total personalization becomes the death of the pool. Insurance is one of the few industries that punishes the very reflexes that conquered the others, because it is not a service laid over a real good, it is the management of the unknown, and the unknown does not reward those who pretend to have abolished it. Whoever wants to change insurance must first accept what it is, a shared wager against an uncertain world, and no interface, however elegant, changes the nature of a wager.
Further reading, the analyses of insurtech funding cycles, the pivot of a great many of these players toward the underwriting-agency model backed by third-party capacity, and reinsurers' commentary on the loss ratios of full-stack books illuminate the gap between the promise and the trade.
In echo, AlgoPolis foundational article 13, insurtech and the insurance value chain, details the separation between distribution, underwriting and risk-bearing.
Insurtech promised to disrupt insurance as fintech had shaken up banking. But one does not disrupt the bearing of risk as one disrupts an interface. The promise was neither kept nor betrayed, it was transformed, from conquest toward infrastructure.
Since 2018, the financial lines market has offered the purest textbook case of the underwriting cycle. A cleansing of supply collided with the fear of a wave of Covid-related claims to produce the most brutal hard market of the century.
Telematics promises to make each person pay the fair price of their own risk. But as pricing individualizes, risk pooling erodes. Behind actuarial fairness lies a choice of society, between knowledge of risk and the solidarity of the group.
The rain did not fall, a satellite saw it, and the money reaches the farmer's account before he has declared anything at all. No one came to measure his loss. The contract paid itself, and that autonomy changes the very nature of what insurance is.
One analysis with every new publication, on insurance, reinsurance, cyber and AI. No account, no noise.
No spam, one-click unsubscribe, no data ever sold.