I got back from NYC this week to a country that had, in the space of a few days, abolished its science and technology department, launched a national centre for AI in policing to great applause, and produced roughly four hundred LinkedIn posts about “innovation.” I read them during AO’s nap times and on the plane, mostly out of curiosity about how much vaseline one industry can get through in a single week (answer: national reserves, and we may need to import). By the time we landed I knew I was going to write this, because I have been playing the keep-calm-and-carry-on game for a long time now and I’m tired of it.
So, no keeping calm today. Carry on though, please, this one’s worth it.
The dream part (yes, there was one)
When we started working with policing, it honestly felt like the reward for years of grind. We were finally building something our mothers – and, let’s be honest, we ourselves – could be proud of. Something that helps a real person on the worst day of their life, not another dashboard nobody opens after the demo.
And we chose this. We could have made far easier money consulting, or running yet another software house renting out engineers by the hour, or joining the brain drain like every sensible adult quietly advised us to (thanks, sensible adults). We wanted to build a legacy. Cool, hard problems. Ambitious, relentless people. All the struggle finally paying off.
Then you run headfirst into the machine, and the machine does not care about your dream.
How did it go, you ask?
Deep breath. The calm version first, because it earns me the right to the angry one.
I run a small tech company. Over the last couple of years we did something most of the current noise about AI in policing is still only promising to do. We put a working tool inside a live police force. We let an independent university evaluate whether it genuinely helped the people using it. We connected it to the systems forces actually run on (yes, including the legacy ones built back in the 90s) rather than a demo environment built to impress a room.
There is no shortage of AI ideas in British policing. What is missing is delivery – working things in real hands that someone has actually paid for. There is an enormous amount of talking, strategising, consulting, launching and re-launching, and unsurprisingly, very little of it ends with an officer using a tool that made their day better.
Look at just the last few days. The Home Office announces AI to summarise disclosure material, and PoliceAI will pilot the tools with a view to scaling across all forces in 2027. A thoughtful account director at a large integrator writes that forces should start with the problem and not the technology, which is correct. The Nuffield Foundation warns that AI is being adopted across justice faster than its impact can be independently evaluated. All reasonable. Most of it I even agree with. And almost none of it touches the thing that actually decides whether good technology reaches the public: whether anyone in the building can say yes, on a timeline, and pay for it.
The machine doesn’t ask “does it work?”
The Nuffield warning is half right and half backwards. Yes, in places deployment is running ahead of evidence. But the problem I have lived is closer to the opposite. Independent evaluation is exactly the thing the system does not know how to reward.
We brought one. An actual academic study run across real operational sites, not a vendor slide claiming ninety-something percent of something nobody can check. In a healthy market that’s a fast lane. In this one it counted for far less than it should have, because the study answers a question – does it work, is it safe – that no part of the machine is built to ask. The machine asks whose budget it is, which framework applies, which of the many people with a veto has to sign, and whether the national centre has stood up yet. Evidence is not the currency in that machine. Process is.
Well, to be fair to the January whitepaper (and I do try), it diagnoses this better than most of its critics do. It says out loud that the current structure runs on voluntary collaboration, that eighty-six separate decision-makers can each hold things up, that pilots perform beautifully in controlled conditions and then stall, that forces buy the same things forty-three different ways. All true. The cure it proposes is consolidation: fewer and larger forces, national procurement, and a national centre, PoliceAI, sitting inside £115m of AI investment over three years to identify, test and scale the technology.
I want that to work. I really do. But watch what it does to a company like ours in the meantime. Forces are told, sensibly enough on paper, not to go buying AI independently while the centre is stood up. So the signal to the whole market becomes… wait. Wait for the centre, wait for the framework, wait for the standards, wait for the merger that reshuffles who your buyer even is. The very centralisation that is meant to speed adoption up has, for now, become the reason nothing moves. PoliceAI was announced in June with a large number attached and a 2027 horizon, and the delivery that matters is already sliding, which is the ordinary fate of these things. Great in theory, announced for years, quietly underdelivered.
And this is not just a small supplier grumbling from the sidelines. My dear colleague Martin Leven, who as CIO led the technology behind merging eight forces into Police Scotland, looked at the same reform and warned that England and Wales are attempting in one leap what Scotland delivered in careful phases. Without national data standards underneath it, in his words, PoliceAI will be a series of pilots, not a national capability. He backs the ambition and still calls the sequencing wrong. When someone who has actually built this before tells you it will stall, that carries more weight than any vendor’s enthusiasm, mine included.
And the reflex runs right to the top. One of the new Prime Minister’s first acts, in the middle of naming his Cabinet, was to abolish the Department for Science, Innovation and Technology altogether and fold it into a new super-department for business, innovation, science and trade. The single government front door that innovators, researchers and AI practitioners actually dealt with? Gone. Scattered into a bigger machine at the precise moment the sector was finally showing momentum. The backlash was immediate – breaking up bodies like the AI Security Institute, the Sovereign AI Fund and UKRI risks stalling a sector growing at ten percent a year, science risks becoming a junior partner in its own department. Even DSIT’s former Permanent Secretary spent the day of the abolition warning that confusion and uncertainty are now the enemy, and that brilliant people are being moved to their third department in two years. She was describing the inside of the building. Out here, where suppliers live, the reshuffle changes none of the things that decide whether good technology ever ships.
The instinct at every level, from a police procurement office to Number 10 itself, is the same: redraw the box rather than ship the thing inside it.
Limbo charges rent, and guess who pays
The space between the announcement and the delivery has a name – limbo – and its cost is carried by exactly the people who didn’t design it. A large systems integrator can wait out a two-year procurement. They have the balance sheet, and they frequently get paid for the “foundations work” whether or not anything ever ships (nice work if you can get it). A company like ours cannot. Every month a decision is not made is a month of runway spent proving ourselves to people who cannot buy, for a budget that has not been released, against goalposts that move whenever the org chart does.
We were held in that state for a loong, looong time. And the people running this system should worry about it more than I do, because limbo is not a pause button for the rest of the world. While a good idea sits in a British holding pattern, other people act. Forces in other countries treat innovation as something to move on this quarter, not this decade. Private buyers who feel the same pain simply say yes and get started. Larger competitors who can afford to wait are perfectly content to let a smaller company do the hard proving, then arrive at the end to collect the contract. Advantage does not wait politely to be claimed.
None of which is a plea, by the way. We are not sitting by the phone. We build where we can move at the speed the problem deserves, the evidence we generated does not expire, and the police work remains exactly the proof it always was. I would just rather this country used what companies like ours can do than pay, in ceded ground, to watch it get used somewhere else first.
The friction I’ll defend, and the kind I won’t
Some of the friction was legitimate, fair enough. This is public money and public trust. There are real questions about bias, data protection and accountability, and they deserve genuine care. I never resented answering them. The safest, most ethical thing you can do with a new AI tool is exactly what we did: test it properly, independently, on real cases, with a human accountable, before you scale it anywhere.
That is not the friction I’m complaining about. The friction I’m complaining about is the other kind. The unclear ownership. The emails left unanswered for weeks – and no, ghosting is not a negotiating tactic, it does no good to either side and above all it is simply fucking unprofessional (oh, pardon the language, swearing is not befitting a professional? Then what do we call promising to “get back to you by Friday” and vanishing for a quarter. Or “the payment is being processed” four months running. Or a supplier learning from LinkedIn that their contact has moved on, and being asked to start the whole dance again with the new person. I’ll mind my language when you mind your calendar, deal?). The quiet institutional assumption that a supplier’s time is free and infinite. It is neither. Responsibility is not a document you publish, it is a practice you commission and pay for, and right now the incentive is to announce responsibly and test almost never.
And before anyone reaches for the tiny violin, I know how this can read, poor little startup, world so unfair, somebody call their mums. That is not the genre of this piece. We’re fine, we’re building. The frustration is on behalf of every small company watching the system clip the wings of exactly the people who deliver. I’m simply upset about the mythology around it, because I refuse to believe this obstacle course is some deliberate school of toughness meant to temper us. You are not Miranda from The Devil Wears Prada, terrorising people into excellence. You are, more often than not, a simple consultant on a day rate whose entire personal interest is to drag things out for as long as possible, keep your ass covered, and do nothing. And let’s remember that Miranda, for all the cruelty, shipped an issue every month.
I’m fairly sure almost everyone in my profession who has dealt with the calcified end of the public sector gets the same gag reflex I do (not always – sometimes it’s even digestible!). They just won’t say it out loud, because you don’t burn a bridge you might need one day. Fair enough. I have very little to waste, so I’ll say it for the both of us. And maybe, just maybe, one public soul with a bit of conscience and an appetite for fighting the status quo reads this and does something.
Meanwhile, next door
The contrast with the private sector is not subtle. In a private business, the person who feels the pain usually controls a budget, or sits one desk from the person who does. You can show them something on a Monday and be running a paid trial by the end of the month. The barrier to entry is lower, the feedback loop is faster, and a no arrives as a no. In the public sector the person who wants the tool cannot buy it, the person who can buy it does not feel the pain, and the money sits behind a gate that has not opened. That is not any individual’s moral failing, it is a structure. But it is a structure that quietly selects for large suppliers and slideware over small companies and working software – and then everyone stands around wondering why innovation in policing feels stuck.
What it would actually take
So if the reforms are serious about “unleashing technology,” here is what working with companies like ours actually takes. Not warm words about an innovation ecosystem. These specific things:
- Pay for the work, and pay when you said you would. Proving a tool inside your environment is paid work, not a free audition. A payment promised in thirty days that lands in a hundred and twenty, after the supplier has chased it more times than anyone should have to, does real damage even when the money is genuinely coming. Late payment is not an admin nuisance to a small company, it is a solvency problem you are quietly handing to your most useful suppliers.
- Give a fast, clear yes or no, and keep to your timeline. A fast no we can act on. It is the slow maybe that kills companies. If a decision honestly needs six months, say six months and then mean it.
- Name one person who can actually decide, with a budget behind them. Do not make a supplier navigate dozens of stakeholders and a shared veto and call that engagement. Your own whitepaper names this exact problem, so fix it at the point of contact, not only on the org chart.
- Reward real evidence. If a company turns up with an independent evaluation and a live deployment, that should move it to the front of the queue, not into the same one as a rendered demo and a glossy brochure. If you want responsible adoption, make responsibility something a company can actually win on.
- Do not let the gatekeeper also be the competitor. When the same body controls the budget, runs the evaluations and is building a rival tool of its own, that is a conflict no amount of good intention resolves. It does not need naming to be obvious, and a reformed system should be designed so it simply cannot arise.
Give us the sand
None of this is a complaint about individual officers, or the many good people inside these organisations who wanted our tool to work and fought internally for it. It is a complaint about a system that has learned to perform innovation faster than it can deliver it. I did not start a company to write flattering emails to procurement teams or applaud another strategy document. I started it to build something that works and see it used by people doing a genuinely hard job.
And this is bigger than our invoices. If we cannot break this spiral, then we – the UK, Europe, whoever we want to be when we grow up – will never get anywhere near what San Francisco was in its glory days. We will not produce innovation at scale. We will produce strategy documents about producing innovation at scale, and congratulate ourselves on the formatting. Give the smaller players a real chance and you would be surprised what creativity it sparks. We want the hardest problems. We have the people. We have the tools. We even have the sandbox. We are just missing the sand.
The reforms might change this, and I hope they do – hey, hope dies last. If PoliceAI becomes a real front door for real evidence, paid on time and decided quickly, nobody will be happier than me and I will say so gladly, in writing, right here. But hope is not a plan, and the pattern so far is the familiar one: a big announcement, a long horizon, and slippage nobody quite owns. Meanwhile the work carries on, ours and everyone else’s, and the advantage keeps moving to whoever is willing to act.
If you are one of the people this reform is aimed at, you have more power to break this than you think. Not another framework. A yes. A date you keep. An invoice paid when you said it would be.
Come on. Let us play, for fuck’s sake.
