Technology for insurance can do fabulous things. It can deliver a wealth of incredibly detailed information about a specific risk to the underwriters’ desktop, including up-to-the-minute earth-observation images and property details gleaned from local building inspection reports. It can sift through submissions to find the ones that match appetite. It can parachute into gigantic troves of data and extract exactly the nuggets of information that an individual needs, and deliver them at exactly the right time.

But insurance technology cannot yet perform miracles. It cannot eliminate uncertainty.* Consider hurricane prediction and loss modelling. Technology can give us, with a fair degree of confidence, and idea whether a forthcoming hurricane season will be active or benign. But it cannot predict how many tropical cyclones will hit Florida in September.

Even after the fact, it cannot tell us what damage was done by those that did make landfall. Case in point: Hurricane Milton. Last week the major loss modellers, in quick succession, published their estimates of the storm’s expected cost to insurers. Despite enormous advances in data and modelling over recent years, their estimates diverge by nearly 200%.

The best experts in the business, with the most advanced tools that computing and data science have been able to deliver so far to assess the cost of a catastrophe, computed a loss estimate range from $17 billion to $50 billion. The $33 billion gap is greater than the GDP of Iceland. It’s the uncertainty gap.

The technology is good, but it cannot know things that are unknown. Nor can it deliver conclusions based on known information to which it has no access. Technology can predict things based on what it does know, but only with a degree of uncertainty. These knowledge gaps are filled by assumptions – in this case, $33 billion worth.

For insurers, the uncertainty is the sweet spot. Without it there is no risk, and therefore no insurance.

Meanwhile, on the process side, technology can reduce frictional expenses when trading with trusted partners. But it cannot do so if those partners aren’t digitally connected, or if they’re connected with systems that incompatible, or if the data isn’t standardised across the connected platforms. And even when all those things are in place, technology will not reduce costs if its use is simply overlayed onto existing processes. It must be embraced.

The maximum benefits of technology – even the very best technology – can be realised only when two conditions are met. The data supplied must include as much as is known, to reduce uncertainty as far as is possible. Second, the people using the technology must do so well. Finally, of course, for any technological feat to come as close to miraculous as possible, very good, robust, intuitive technology is needed. Leave that to us.

* Nor should it. In the absence of uncertainty, there’d be no need for insurance, and we’d all be out of a job.

Guillaume Bonnissent was a Lloyd’s underwriter for 15 years until he founded Quotech.