Little in France was ever more controversial than Superphénix. My dad couldn’t stop talking about it, because it was supposed to make its own fuel.

Super-Phoenix is a 1,242 megawatt fast-breeder nuclear reactor sited between Lyon and Geneva. It was a world first, a power generation facility on a huge scale which promised to create more fissile material than it consumed. Construction of the ground-breaking, science-leading plant began in 1976. Ten years later it was supplying a trickle of electricity to the French national grid. Twenty years later it was closed forever.

The thing cost 60 billion francs, about €19 billion in today’s money.

“The problem with Superfiasco,” my father told me before its off-switch had been permanently flipped, “is that the des ronds-de-cuir didn’t know when to stop.” He meant that the technocrats at the Ministry of Widgets kept pouring in the money when it was obvious they should have desisted.

Times had changed before the behemoth was even running at speed. Breeders were rendered unnecessary because uranium turned out to be abundant. In any case, the writing was on the wall for nuclear power (albeit in disappearing ink) decades before the under-performing plant was decommissioned.

The problem was this, may father said (and he was wise about such things). Shutting down Super-Phoenix would leave a great many civil servants, not to mention a large cadre of politicians, with big splodges of green-glowing egg on their faces. Once you’re, say, two billion into a project, it’s pretty hard to admit: “Err, sorry, got that one wrong.”

That’s the principal reason why business organisations in the US and UK have wasted an astonishing average of 2.4% of annual revenues on IT projects which have simply failed to deliver the value expected. A measure of failure is to be expected – there’s no universe where every IT project is a screaming success – but the shocker is how long it takes some companies to pull the plug.

A consulting firm called Emergn asked senior executives from 700 US and UK companies about their AI initiatives and IT transformation programmes. Fewer than a third said it’s normal for them to be shut down when they’re underperforming. In contrast, nearly half said that the typical practice in their organisation is to put the brakes on only after the wasted investment of “significant” quantities of time and money (on average, again: 2.4% of turnover).

“Transformation and AI investments keep leaking value,” Emergn wrote. Their survey found 70% of respondents do not stop underperforming AI cleanly. Instead, they end projects late, or not at all. “Projects survive purely because too much has already been invested, and that stopping decisions are driven more by politics than evidence”, Emergn concluded from the comments received.

Almost daily we see more headlines about the percentage of insurers that are actively incorporating AI into their business, so it’s worth reporting a few more findings from this survey. Of the 700 companies, each with 1,000 or more employees, 71% can’t produce a real-time view of all the initiatives they have under way. To do so would either take days or weeks, or is deemed impossible. And it’s no wonder they cannot keep track: 78% report simultaneously running between four and twenty transformation and AI initiatives across their organizations.

Shockingly, 40% of respondents said that up to a quarter of their transformation and AI initiatives are not subject to any formal tracking, reporting, or governance at all. They’re simply allowed to run, and spend money as they go. That gives failing work the green light to survive, and to continue to erode the bottom line as it continues unchecked.

When reporting is required, it’s often akin to a smoker’s assessment of their own lung health. A fifth of US respondents and a quarter of those in the UK said that status reports “present a more positive view than reality”. UK leaders are more likely to report improvisation (40%), on-the-job learning (42%), and key-person dependency (36%) than their US counterparts.

When it comes to AI specifically, investment in tools is greater than the amount spent on the people deploying them, and leaders know it. Only almost half have received “substantial ongoing training”. Meanwhile, 22% of leaders avoid speaking up about AI for “fear of getting it wrong”.

Getting it wrong is part of the process. At least some development projects need to follow Mark Zuckerberg’s “move fast and break things” approach. When things are changing so quickly in the world of technology, adopting a start-up mentality can be a good way of keeping on top of innovation. But it needs to be done with realism at the forefront: a lot of tries will not convert. Breakages must be paid for, so smash the B&M plates, not the Ming vases.  

The people surveyed by Emergn (CEOs, COOs, CTOs, and transformation, project, and change professionals) work for companies in “finance, IT & telecommunications, manufacturing & utilities, healthcare, retail, and education.” I am sure a few insurance execs are in there, but I expect the results for a similar survey focussed entirely on our sector would be at least the same, and possibly more so. The industry’s track record with IT initiatives hasn’t been stellar, and they may find it more difficult than most to adopt the technology trial-and-error method.

Either way, many companies are trying. Project failures are inevitable, but they cannot be allowed to cost 2.4% of sales, regardless of whether senior executives have a handle on technology.

Put it this way: if I produced a new widget that would guarantee to shave 2.4 points off the expense ratio, how would insurance executives react? If I became a governance consultant, suggested a new group policy that would have the same impact, and flogged it for a million a pop, how many big risk carriers would sign up? What about a new sales too that could multiply the top line by 1.024% instantly and almost without effort? Or maybe I sell them a news claims management tool, proven to reduce annual claims costs by 2.4 points, and charge 10% of the savings.

Any of those, I bet, would let me give up work even faster than Super-Phoenix.

Guillaume Bonnissent is CEO of Quotech.