Every proposal has an appendix nobody reads. This one had a single cell in it that has stayed with me longer than anything on the main deck, because I can still go and check whether it is true. It is.
The engagement was a competitive response to a European composite insurance group standing up a group-level analytics function. Most of the pack is method: how to build a value map of the insurance chain, how to run the exercise inside each operating company, how to harmonise what comes back so the results can be ranked against each other. Behind that sits an appendix working one market through in detail, the connected home, laying out three ways in, ordered by ambition. Each route carries a description, the technology it would need, the partnerships it would lean on, a row for market examples, and an indicative reading of what the move would demand of the insurer across five headings: the proposition, marketing, costs, resources and capability.
The first two routes have their market-example rows filled in with live insurer partnerships. The third route's row opens: no existing insurance proposition. The only comparator it can offer sits outside insurance altogether, a partnership deploying assistive technology for vulnerable people.
Three ways in, and one of them had no insurance examples
The first route is the one everybody took. Keep the household policy exactly as it is and let data off connected devices inform the price or reduce the risk: how often the property sits empty, whether it is locked, whether the smoke alarm still works, whether there is water where there should not be. Delivery is a partnership with somebody who already makes the sensor, and the commercial expression is a discount. The market examples collected at the time run 7.5 to 12.5 percent off contents cover for taking a partnered alarm, and a quarter off a home premium for accepting a bundle with a leak sensor, door and window sensors and a motion detector in it. The reading against that route was low, low, low, low, and medium against capability alone. A new field arrives at the pricing engine and the rest of the company carries on.
The second route bundles protection services with the policy rather than only pricing inputs: the same sensors, now wired to early detection, prevention and a response when something trips. Market examples existed here too, a telecommunications operator's home platform joined to an insurer's emergency assistance service, and a bank-owned insurer selling a monitoring device that alerts both the customer and an operations centre for fire, smoke and flood. The reading moved to medium on proposition and medium on marketing, and stayed low on costs and low on resources. The step up was in what the company sold and how it explained it, not in what it spent.
The third route is the one with no insurer against it. It covers the household and the family: detection, prevention and response for the property and for the people in it. The appendix is specific about where that leads, and it leads out of property insurance entirely. Information shared with care providers. Informal and formal carers made more productive. Patients monitored remotely so somebody can intervene when the signal says intervene. People supported in managing their own conditions so that reliance on care falls. The worked example is one cover that watches the house and also monitors and supports a vulnerable family member.
The two readings that sat in the top band
Read the third route's five readings in order and the shape of the problem falls out. Proposition: high. Marketing: medium. Costs: medium. Resources: medium. Capability: high.
Cost sat in the middle band. Resources sat in the middle band. They sat in the same band that the second route's marketing sat in, and nobody thought the second route was blocked. The only two readings in the top band were the proposition itself and the capability to run it.
That is not a budget refusal. It says the company would have had to sell a different thing and be able to do a different thing, and that the spend, the headcount and the go-to-market were ordinary by comparison. An insurer is very good at pricing an event it can define and settle after the fact. The third route asks it to operate a continuous service in somebody's home and somebody's health, where the value is created by an intervention that stops the loss happening. The claim never arrives, so the thing you were selling never appears in the ledger where the business is used to looking for it.
The technology list makes the same point from the other side. Hardware in the building, external body-worn sensors such as patches, watches, glasses and trackers, internal ones such as implants, pumps, chips and stimulators, and a further group covering temperature sensing, microfluidics, energy harvesting from the body and connections through a local phone. Every one is a continuous stream with a duty of care attached. The pricing route needs a field. This route needs an operating discipline.
The demand side was not the problem
It would be tidy to conclude that customers did not want this. The same appendix argues the opposite, with other people's numbers rather than its own.
Home insurance churn intent ran at 36 percent of customers expecting to switch within twelve months. Sector net promoter scores sat at minus 6. Installed home connected-device solutions had gone up fourfold across five years. The line that follows those figures in the pack is blunter than a consultant usually writes: the customer sees nothing in the product worth valuing and owes it no loyalty, so the sale falls to price and nothing else. A product nobody can tell apart from its neighbour is sold on price, and a product sold on price cannot be defended.
The hardest number on that page rates who was thought likely to create the connected-home ecosystem. Two consumer technology firms came out on top at 67 and 61 percent, telecommunications operators at 50, home appliance manufacturers at 44, utilities at 28, other technology firms at 22. Insurers came last, at 6 percent. The industry closest to the risk, holding the customer relationship and the loss history, was rated least likely to build the thing that would reduce the loss.
- No insurance proposition
- What the third route's market-example row records first
- High / Medium / Medium / Medium / High
- Its indicative demand on the insurer, across proposition, marketing, costs, resources and capability
- 36%
- Home insurance customers intending to switch within twelve months
- 6%
- Insurers rated likely to create the connected-home ecosystem, last of seven candidate industries
Why the gap is still there
Go and look for that third product now. In most European markets the first route is everywhere and the second exists in a dozen recognisable forms. The third is still largely unoccupied. Pieces of it live in care technology and in health services, but as an insurance product, underwritten and priced and sold as one cover over a home and the people in it, finding an example remains a research exercise.
None of the reasons are technical any more, which is what makes the persistence interesting. Ingesting device telemetry is ordinary work. Machine learning pipelines that watch a stream and raise something are assembly at this point, model deployment is close to solved in most stacks, and the language-model pilots crossing my desk this year stand up in days even when the governance around them takes months. Feature stores, lineage and process mining all exist to answer exactly the questions this product would raise: where did this reading come from, what did it mean, who acted on it, how long did the act take. Those four questions are the whole of what RealAI's Platform readiness pass asks of an estate before anyone writes a model.
What has not changed is the second top-band reading. Capability, in the sense the appendix meant it, is not a data platform. It is somebody accountable for what happens between the sensor firing and a person knocking on a door, a service-level commitment measured in minutes rather than claims cycle days, a safeguarding duty an underwriting function has never carried. It is a decision about what the business is for, taken by people whose incentives sit on the combined ratio and who can meet that target without ever attempting any of this.
Cost sat in the middle band and so did resources. The only two readings in the top band were the proposition itself and the capability to run it, which is another way of saying the company would have had to become a different kind of company.
That is how insurance products actually get born, and how they fail to. Not through a failed business case. Through a business case that was never the constraint, sitting next to a demand on the operating model that no committee is chartered to approve. The first two routes shipped because they could be approved by a pricing function and a marketing function acting alone. The third needed somebody to change what the company does.
The assessment did its job. In three bands across five headings it said the money was fine and the identity was not. What it could not say was whose signature would be required, and a ranked list of opportunities is only as good as that question. RealAI's Consult team asks it first for exactly this reason: rows that read medium against cost and high against capability are the ones that vanish between the roadmap and the budget cycle, because their sponsor is nobody in particular.
If a scan of your own portfolio is producing rows like that, modest spend against an impossible-looking capability demand, those rows are the portfolio. Everything else is what your competitors have already built.
Readings are as recorded in the connected-home appendix of a competitive proposal written for a European composite insurance group: three entry routes, their indicative demand on the insurer across five headings, and the market examples found against each. Those were estimates offered in a bid, not measured outcomes, and the market statistics on that page came from published third-party research. Reading that missing insurance example as a statement about operating models rather than about demand is ours.
“Cost sat in the middle band and so did resources. The only two readings in the top band were the proposition itself and the capability to run it, which is another way of saying the company would have had to become a different kind of company.”
Get in touch
Put RealAI’s applied-AI team on your hardest data problem.
We help enterprises move from pilots to production: sovereign models, governed data, and agents you can audit. Start with a value-first assessment.