Britain’s roads have long been a national joke. Every winter, social media fills up with photos of craters deep enough to swallow a wheel, and every spring, councils publish repair budgets that never quite stretch far enough. But something genuinely interesting is happening underneath all that grumbling: local authorities across the UK are quietly deploying AI pothole detection systems that can spot damage days or even weeks before a human inspector would ever notice it. Machine-learning cameras on council vans, drones scanning rural B-roads at dawn, algorithms flagging micro-cracks before they become tyre-wrecking holes. I find this stuff genuinely exciting, and the results so far are more promising than the usual council tech story.
How AI pothole detection actually works
The basic idea is elegantly simple. A camera-equipped vehicle drives its usual route, and rather than waiting for a pothole report from an angry motorist, the onboard system is continuously analysing the road surface in real time. Software trained on thousands of images of road defects can classify damage by type, depth, and urgency, then pin it to a precise GPS coordinate and push it straight into a maintenance management dashboard.
Hertfordshire County Council has been running one of the more mature versions of this, using a system called Verizon Connect (formerly known under the Gaist brand) that analyses imagery from cameras mounted on council vehicles as they make routine journeys. The footage is processed by machine learning models, and the council ends up with a continuously updated map of every pothole, cracked kerb, and surface failure on its network. No need to wait for a resident’s report. No need to send a separate inspector out. The van doing the school run data collection has already done the job.
Drones add another layer, particularly useful for rural roads where vehicle access is tricky or traffic volumes don’t justify running a data-collection van through every fortnight. Durham County Council trialled drone surveys on rural sections of its network in 2025, and the ability to capture high-resolution imagery from above helped surface issues on verge edges and drainage channels that ground-level cameras miss entirely. You get a proper bird’s-eye picture of the road’s condition rather than a worm’s-eye one.
Which councils are doing this, and what are they finding?
It’s not just the big metropolitan authorities. Norfolk, Oxfordshire, and several Scottish councils have all run or are running AI-assisted road survey programmes. Transport for London uses a variant for its managed road network too, though the sheer density of London’s traffic makes the data volumes involved pretty staggering.
What they’re all finding is that early detection genuinely changes the maths. A road surface that gets treated at the micro-cracking stage costs a fraction of what it costs once it’s opened into a pothole and the sub-base is exposed to water. The UK’s roads already cost the economy an estimated £3 billion a year in vehicle damage according to the RAC Foundation, and a significant chunk of that comes from defects that were spotted too late. Earlier detection means cheaper repairs and, in theory, shorter backlogs.
I’d caveat that “in theory” carefully. A few councils have been honest that the detection technology is improving faster than their repair capacity. You can have a perfect map of every pothole in your county and still not have enough gangs to fix them all. Staffordshire County Council, for example, has been transparent about the fact that its AI survey data has actually revealed a larger backlog than previously estimated, because the system finds damage that manual inspections used to miss. That’s useful information, but it’s also uncomfortable when the budget doesn’t grow to match.
Are repair backlogs actually shrinking?
This is the honest question, and the honest answer is: it depends entirely on funding. The Local Government Association has been warning for years that councils need billions in additional funding just to clear existing backlogs, let alone keep pace with new damage. AI detection doesn’t magic money into existence. What it does do is help councils spend what they have more efficiently, prioritising repairs by actual risk rather than by whoever phoned the complaints line most recently.
Oxfordshire ran an interesting pilot where AI-prioritised repairs were compared against a control set of roads managed the traditional way. The AI-managed roads showed a measurably slower rate of deterioration over 18 months, largely because preventative surface dressing was applied earlier. That’s a genuine win. Whether it translates to shorter backlogs depends on whether the preventative treatments keep happening at scale, which circles back to budget.
There’s also a data-sharing opportunity that’s barely been touched. If councils share their road condition datasets with each other and with Highways England (now National Highways), there’s a much richer picture of network-wide deterioration patterns. Satellite-based interferometry, used to detect millimetre-level ground movement, is already being applied to infrastructure monitoring by companies like Rezatec. Road surfaces aren’t far behind. My take is that the councils doing this well right now are building a foundation for a genuinely smarter network over the next decade, even if the immediate backlog reduction is modest.
It’s worth noting that infrastructure monitoring is becoming a broader obsession. The same impulse that has councils putting sensors on roads has domestic engineers thinking about monitoring the condition of rooftop kit like TV Aerials before faults develop into bigger problems. Early detection is just a smarter way to manage anything that degrades over time, whether it’s a B-road in Staffordshire or a Yagi antenna in a January gale.
The privacy and public trust angle
Camera-equipped council vans driving every road and drones buzzing overhead do raise questions, and I think it’s worth taking them seriously rather than dismissing them. The ICO’s guidance on public space surveillance applies here, and councils need to be clear with residents about what footage is captured, how long it’s retained, and what it’s used for. Most of the systems currently deployed are processing imagery locally and discarding raw footage, keeping only the defect classification data. That’s a sensible approach, and councils should be communicating it clearly rather than letting the tech roll out quietly.
There’s also something worth celebrating in all of this. The same algorithmic thinking that’s shaking up everything from AI personal training to sorting second-hand clothes in warehouses is now being pointed at genuinely unglamorous public infrastructure. Potholes are boring until one of them wrecks your front suspension on the A419 at 06:30 on a Tuesday morning. Then they’re infuriating. Anything that helps catch them earlier, even a little bit, is fine by me.
What needs to happen next
The technology is good and getting better. The real bottlenecks are funding, repair capacity, and cross-council data sharing. AI pothole detection is not a silver bullet, but it is a genuinely useful tool that shifts councils from reactive to proactive management. The Department for Transport has been nudging councils toward digital asset management for a few years now, and the 2025 Roads Investment Strategy included provisions for encouraging AI-assisted maintenance planning.
If you’re curious about your own council’s road condition data, most authorities now publish a public register of reported defects. Some have gone further and published their AI survey results in open data formats. It’s worth a look. You might find the pothole that’s been annoying you for months is already in the system, flagged amber, waiting for a repair gang to get to it. Whether that gang arrives before your next appointment with a tyre fitter is, unfortunately, still a question that no algorithm can fully answer.
