AI pothole repair could help councils tackle road maintenance far more intelligently. From predicting which roads are likely to fail next to prioritising budgets and routing repair crews more efficiently, AI can improve the decisions behind maintenance work. But while the technology can strengthen planning, it cannot solve the deeper funding and policy problems that keep Britain’s roads in poor condition.
There are, according to the RAC, around a million potholes scattered across Britain’s roads. That’s roughly six per mile, which means that somewhere out there is a perfectly maintained stretch of tarmac hoarding everyone else’s share, and the rest of us are driving over what feels like the surface of the moon.
We are, as a nation, furious about this. And not in the abstract. A YouGov poll taken just before the recent council elections found that the local issue voters cared about most wasn’t the cost of living, wasn’t the NHS, wasn’t immigration. It was potholes, congestion and road maintenance. There is something almost poetic about a country that will calmly absorb a constitutional crisis but lose its entire mind over a crater on the B6362.
The numbers behind the rage are real, mind you. Compensation claims against councils for pothole damage jumped 90% in the three years to 2024. In February 2025, more than three times as many drivers blamed potholes for their breakdowns than a year earlier. Even the Transport Secretary, Heidi Alexander, reportedly got her Mini wedged in an Oxfordshire “moon-crater” and had to be towed out. It is a sentence that should be carved above the door of every council highways department as both warning and motto.
So here’s the question we keep getting asked, in one form or another: can’t AI just sort this out? New York’s mayor claims he filled 100,000 potholes in 100 days. Surely the country that gave the world the steam engine and the spreadsheet can throw an algorithm at a hole in the ground?
Let’s take the question seriously. Because the honest answer is: partly. And the reason it’s only “partly” is more interesting than a straightforward yes.
Where AI is genuinely brilliant at this
The dirty secret of pothole management is that filling potholes isn’t really the job. As the people who actually run council highways departments will tell you, a pothole isn’t the problem; it’s a symptom that the whole road underneath is failing. One highways chief puts it nicely with a windowsill analogy: look after it and it lasts a lifetime; ignore it and it rots, cracks, and eventually costs you the whole window. The DfT’s own figures suggest that fixing roads properly and early pays for itself more than four times over within a decade.
This is exactly the kind of problem machine learning was born for. Not “where is the hole” (your suspension already knows that), but “where will the next hole be, and which road is quietly rotting before anything visible has happened at all?”
A few concrete things AI can do today, not in some shimmering future:
- Predict failures before they appear. Feed a model the age of the road surface, traffic loads, drainage data and, crucially, rainfall, and you can rank which roads are about to fail. Given that water is the single biggest cause of potholes, and that our winters are getting wetter, a model that watches the weather is watching the actual culprit.
- See the whole network cheaply. Computer vision applied to dashcam, bus and bin-lorry footage can survey thousands of miles of road continuously, instead of relying on a once-a-year inspection or an irate resident with a tape measure and a grievance.
- Spend the money where it bites hardest. Councils filled 1.9 million holes last year, roughly one every 17 seconds, and it still wasn’t enough. When you’re that overstretched, the difference between a good prioritisation algorithm and a bad one is measured in millions. AI is very good at “given £3m, where does it do the most good?”
- Route the crews. Boring, unglamorous, enormously valuable: optimising which crew fixes what, in which order, on which day, before which rainstorm.
Notice that almost none of this is about the dramatic bit: the hi-vis, the hot tarmac, the photo op of a party leader gamely holding a shovel. It’s about the unsexy intelligence layer underneath. Which, conveniently, is the part humans are worst at and computers are best at.
Where AI runs straight into a wall (or a windowsill)
Here’s the thought-provoking bit, and it’s the part the “just use AI” crowd tends to skip.
AI cannot print money.
Take a typical mid-sized city council. Even after a welcome injection of extra cash from the Department for Transport, a highways boss might have a few million pounds to spend this year against a genuine need of three times that, just to stop new holes forming. Scale that up and the picture is stark: councils in England and Wales reckon clearing the existing repair backlog would cost £18.6 billion. The cleverest predictive model in the world, handed a fraction of what £18.6bn would require, will simply produce an exquisitely optimised list of roads it cannot afford to repair. That’s not a solution. That’s a beautifully formatted apology.
There’s a second, subtler trap. A lot of pothole funding comes with strings: spend it on potholes, spend it by this date, and publish your numbers or lose the cash. It sounds accountable. In practice, as the Institute for Government has pointed out, it can lock councils into patching symptoms rather than fixing the underlying road, the exact short-termism the windowsill analogy warns against. You can build the smartest preventative-maintenance AI imaginable, and a funding rule that says “spend it on visible holes by March” will quietly overrule it.
This is the genuinely important lesson, and it’s not really about roads at all. AI is a force multiplier for good decisions and a force multiplier for bad ones. Point it at “how do we maintain this network for the lowest whole-life cost” and it’s transformative. Point it at “how do we fill the maximum number of holes before the next photo opportunity” and it’ll do that beautifully too, and the spiral that experts warn about, where decaying roads eat ever more of the budget just to stay roughly safe, carries on regardless.
So, can AI fix Britain’s potholes?
It can fix the intelligence problem, and that problem is real and expensive. It can tell us where to dig, when to dig, and where a stitch in time saves nine. It can turn a reactive, complaint-driven scramble into something that looks suspiciously like planning.
What it can’t do is fix the will problem: the funding cliffs, the annual budget cycles, the political reward for visible patches over invisible prevention. That part is stubbornly, irreducibly human.
Which is rather the whole point of how we think about AI at Trimontium. The technology is rarely the bottleneck. The bottleneck is whether an organisation is willing to act on what the technology tells it. Get those two things aligned, and you can fill an awful lot of potholes. Get them misaligned, and you’ve just bought a very expensive, very accurate map of your own decline.
The good news? Filling six potholes a mile is a tractable problem. The honest news? The hard part was never the holes.
Author: Deborah Holmwood, Client Change & Transformation Partner.
Follow our LinkedIn company page to stay up to date with all our new blogs!
