Why I left insurance to build a construction data company.

Over almost two years of development underwriting I kept running into the same thing: the insured already held, or would hold, records that would have changed how their risk was perceived and managed; it had not occurred to anyone to share them.

Not withheld. Not disputed. Simply never offered, and never asked for.

It started with mass timber

The work that made it obvious was trying to improve the insurability of mass timber.

Mass timber is a good test case because the market's difficulty with it is not really about the material. Insurers understand the fire performance reasonably well. What they struggle with is water — moisture ingress during construction, and the damage and delay that follow it — combined with a thin claims history to price against. Aon put it plainly in their own guidance: limited historical data makes it hard to develop comprehensive underwriting practice, and insurers are looking for more evidence of proper project execution than is currently available.

Read that last part again. Not more capital, not better modelling. Evidence of execution.

And the same guidance says what would help: documenting the construction phase in much greater detail, involving risk engineers in site reviews as the build proceeds, proper moisture control and monitoring. Every one of those is something the project can already do, and on a well-run mass timber job, largely already does. The moisture readings exist. The sequence is photographed. Somebody knows when the panels were exposed and for how long.

So here was a type of risk where insurability could be improved with evidence of how the project was actually built — and that data existed but did not travel.

The pattern was not confined to timber

Once I had seen it there I saw it everywhere. Across those two years the shape repeated: insureds had, or could easily have had, data and records that fundamentally changed how their risk should be understood. Almost none of it was shared. Almost nobody suggested sharing it.

Water damage is the clearest case. It is routinely cited as the market's biggest cause of loss, the monitoring and automatic shut-off technology is proven and widely deployed, and there is now a joint code of practice for managing escape of water on construction sites — yet specifying that technology as a requirement of cover remains variable. Part of that is the cycle: a softening market rewards writing the premium over insisting on the controls. Part of it is simpler. A condition nobody can verify between site visits is a condition few underwriters will impose.

It is worth being clear that this is not anyone behaving badly. There was no route for the information to travel down. Nobody's job description included building one. The insured had no particular reason to imagine that a moisture log or a set of progress photographs was of interest to their insurer, and nobody had told them otherwise. A market that has always worked from proposal forms and periodic surveys does not develop the habit of asking for anything else. And driving product development across a syndicated market is close to impossible: the risk is shared, so none own enough of it to pay for something new — and everyone who waits gets it for nothing if someone else does.

The information sat on one side of a line and the people who needed it sat on the other, and both sides had perfectly reasonable grounds for staying where they were.

The obvious next move, and why it fails

If you want to fix that, the obvious move is to prove the point commercially. Show that projects with continuous site evidence produce better outcomes and the market will pay for the connection.

You cannot. Not now, and not for a long time.

Construction losses split into two kinds, and neither co-operates. Attritional losses — water above all — are frequent enough to measure, and the market has already responded in its own way: deductibles pushed up until much of that cost sits back with the contractor. Whatever monitoring saves there is mostly the insured's own money, which is why the trades that exist are deductible trades. The severe losses — the fires, the collapses, the defects — are what a book is really priced on, and they arrive too rarely and too unalike for any book to isolate what continuous evidence contributed. The frequent losses are retained; the rare ones can't be counted. Either way, nobody can hand you a study that proves the value of visibility on the cover itself.

This is a profound problem if you are trying to launch a product rather than write a think piece. The pitch would be: buy this, and something you cannot measure will improve by an amount I cannot state, on a timescale I cannot promise. That is not a product. It is a hypothesis with an invoice attached, and the market is right to refuse it.

Risk engineering is where the argument closes

Which is why Pikt starts where it does.

Risk engineering is the one place in this chain where the value of better site data is direct, immediate and measurable — and it is measurable without waiting for a single loss to occur or not occur.

The measurement is simply coverage. A risk engineering fee is finite — not because anyone worked out what level of oversight would create value across the life of the policy, but because the fee divided by the cost of a survey produces the number of surveys. Frequency is set by the budget rather than by the risk, and between surveys the site is unseen. Connect the technology already running on those sites and the same fee covers every project continuously, with changes flagged as they happen and physical surveys directed where the data says they matter most.

More risk engineering from the same fee. That is an efficiency you can state at the start of a contract and check at the end of it. No loss study required.

It is also the point in the chain where one piece of work pays several parties at once. The engineer arrives knowing what has changed rather than discovering it on the day. The carrier's engineering budget covers more of its portfolio. The project team stops writing progress reports by hand, because where those updates are already owed to insurers, the site's own data can assemble them — verified and time-stamped, instead of somebody's recollection compiled on a Friday afternoon.

Three benefits, one connection, an existing budget line. In a market where almost every proposition asks someone to spend new money on an unproven return, that is unusually good value, and it is available now.

The rest — what continuous evidence eventually does for the physical damage cover, and for the arguments that follow a loss — is a longer conversation and one the data will have to win on its own terms. It is not what anyone is being asked to buy today.

What we built

Pikt gives risk engineers continuous visibility of the projects they oversee. It connects to the technology already running on site — fixed cameras, 360° walk-throughs, environmental sensors for water, moisture and weather, HSE systems, drone capture and progress tracking — and turns the output into a time-stamped view of each site. No new hardware. Nothing extra for site teams to do.

The engineer still goes to site. That does not change, and it should not: there are things you can only learn standing on the floor. What changes is everything around the visit.

The honest version of why I left

I did not leave because insurance is broken. It works, and the people in it are careful and serious about risk.

I left because of the gap between what a project knows about itself and what the people carrying its risk are able to see — and because the reason for that gap was never difficulty. It was that connecting the two had never been anybody's job.

That is a strange thing to keep noticing. It is also, if you are going to build something, about the best position there is.