In summary
- Organisations need evidence of real-world impact before scaling AI, focusing on outcomes rather than technical performance alone.
- Building trust in AI requires continuous evaluation, clear governance and the ability to learn from implementation.
- AI Labs and structured experimentation play an important role in testing, validating and refining AI use cases.
- Long-term success depends on developing organisational capability, including skills, governance and shared best practice.
- The most effective AI solutions enhance human judgement, reduce administrative burden and allow people to focus on higher-value work.
- While the discussion focuses on policing, the lessons apply to any organisation looking to adopt AI responsibly and at scale.
Over the past few years, I've had countless conversations about AI. Most begin with the technology. Which models should we use? What can generative AI do? How quickly can we adopt it?
Increasingly, I think those are the wrong questions to ask.
The launch of PoliceAI represents an important shift for UK policing. Backed by significant government investment, its ambition is clear: to help forces adopt AI safely, responsibly and at scale.
And that ambition extends well beyond policing. Across every sector, organisations are moving past the question of whether AI has potential. The focus is now on how to deploy it in ways that genuinely improve outcomes while maintaining trust and accountability.
That was the focus of our recent panel discussion, AI in policing: from experimentation to real-world impact, where I was joined by:
- Professor Barak Ariel from the University of Cambridge, a leading authority on evidence-based policing;
- Daniël Stuart, who established and leads the Dutch National Police AI Lab;
- Tanya Shackleton, who brings more than 25 years of frontline policing experience and supports over 30 UK police forces through the STORM CAD platform;
- Craig Dibdin, whose 30-year policing career now informs his work helping public safety organisations adopt new technologies.
Together, we explored what it really takes to move AI from isolated experiments into trusted operational capability.
If you'd like to watch the full discussion, you can watch it here: https://youtu.be/Uwf26MjX-NQ?si=1Rl4kI8TQdvREt9Y
Although our conversation centred on policing, it was clear that the lessons are relevant for any organisation adopting AI. Different sectors face different challenges, but the same questions keep emerging. How do you identify the right opportunities? How do you know AI is genuinely creating value? And how do you build the confidence to scale it responsibly?
I believe that these are the right questions to ask.
Throughout the discussion, there was a key theme that emerged again and again. Successful AI adoption is driven less by technology, and more by evidence, organisational capability and a clear understanding of the problems you're trying to solve.
Start with the right problem
It's easy to become distracted by what the latest models are capable of. The panel repeatedly brought the conversation back to a much simpler starting point: what operational problem are you trying to solve?
Craig highlighted opportunities to reduce administrative burden, improve resource planning and help officers spend more time supporting communities. Tanya described how AI could support control room staff by bringing together information that already exists across multiple systems, helping them assess risk more efficiently while allowing them to focus their attention on the person at the other end of the phone.
Those examples all have one thing in common. They begin with operational challenges rather than technical capability.
Daniël Stuart also shared an important perspective from the Dutch National Police. A model can perform exceptionally well from a technical perspective, but if it doesn't improve policing outcomes, it hasn't solved the problem it was built for. Equally, experimentation often uncovers benefits that weren't part of the original objective, reinforcing the value of testing solutions in real operational environments rather than relying solely on technical benchmarks.
That shift in thinking is relevant far beyond policing. Organisations who start with understanding where people are losing time, where decisions are difficult or where existing processes create unnecessary friction will mean AI has far more impact when it is designed around those challenges.
Prove value before you scale
If there was one idea the panel returned to more than any other, it was evidence.
Professor Barak Ariel compared AI adoption to clinical trials in healthcare. New treatments aren't rolled out because they appear promising. They're tested rigorously to understand whether they genuinely improve outcomes. AI deserves the same level of discipline.
That means looking beyond measures such as model accuracy, response times or technical performance. Those indicators tell us something about the technology, but they don't necessarily tell us whether the organisation is performing better.
Instead, organisations need to evaluate the outcomes that people actually experience.
- Has AI reduced administrative effort?
- Has it improved consistency?
- Are professionals able to spend more time applying their expertise?
- Are services becoming more responsive?
- Are people having a better experience?
Those are the questions that create confidence to scale.
The discussion also highlighted that evaluation shouldn't stop once a pilot ends. AI operates in complex environments where technology, people and processes constantly influence one another. Continuous evaluation helps organisations understand not only whether something works, but why it works and whether it continues to deliver value over time. In the case of policing, officers testing and feeding back in the field will prove value more robustly than any test environment.
That evidence becomes the foundation for trust. It gives leaders confidence to invest further, helps the workforce understand where AI genuinely supports their work and provides reassurance that new capabilities are improving services.
Build the capability, not just the technology
One of the biggest misconceptions around AI adoption is that success depends primarily on choosing the right tools.
Our discussion suggested something different.
Technology is only one part of the equation. Organisations also need the capability to evaluate new ideas, develop internal expertise, establish appropriate governance and share learning as AI evolves.
This led us into a conversation about the role of AI Labs. Their purpose extends far beyond building prototypes. They create an environment where organisations can test use cases, benchmark technologies, validate outcomes and develop the evidence needed before solutions reach operational teams. They also help organisations develop the internal expertise needed to make informed decisions rather than relying entirely on external vendors.
The conversation also explored the balance between local innovation and national consistency. Teams closest to operational challenges are often best placed to identify opportunities for AI, but successful approaches need common standards, shared governance and mechanisms for learning across the wider organisation.
Those challenges aren't unique to policing. Every organisation adopting AI needs people who understand both the technology and the context in which it's being applied, governance that evolves alongside rapidly changing tools and processes that allow successful ideas to be shared rather than repeatedly reinvented.
For example, if one force successfully uses AI to reduce the time officers spend reviewing digital evidence, the real value comes when that approach, the lessons learned and governance behind it can be adopted and adapted by other forces rather than starting from scratch. Without that capability, even the most advanced tools will struggle to deliver lasting value.
AI should make people more human, not less
Perhaps the most memorable insight from the discussion was that the purpose of AI is ultimately about people.
Throughout the session, we kept returning to professional judgement, public confidence and human interaction.
Tanya described how AI could reduce the time spent navigating systems and completing administrative tasks, allowing officers and call handlers to focus more fully on victims and the decisions that require experience, empathy and judgement.
Professor Ariel reinforced that point by sharing research showing that while automation often improves efficiency, people continue to value human interaction, particularly when they're vulnerable or experiencing difficult situations. Technology can speed up a process, but trust is often built through communication, explanation and reassurance.
That feels like an important principle for any organisation adopting AI.
The same principle applies across all organisations. AI is at its most valuable when it removes repetitive work, reduces cognitive load and helps professionals apply their expertise more effectively, and crucially creates the capacity to think about how services could be reinvented. It should strengthen human judgement and give them agency rather than distance people from the individuals they serve.
Looking beyond policing
Looking back on the discussion, what stays with me is the consistency of the message from people approaching AI from very different perspectives.
Whether coming from policing, academia or technology, they kept returning to the same themes: evidence, people and operational outcomes.
Those principles extend far beyond policing. Every organisation exploring AI has decisions to make about where to invest, how to measure success and how to build trust. Technology will continue to evolve quickly, but those questions are likely to remain the same.
The organisations that create lasting value from AI won't necessarily be those that adopt it first. They will be the ones that understand the problems they're solving, evaluate solutions carefully, build the capability to learn as technology evolves and keep people at the centre of every decision.
For me, that's the real lesson from this discussion. Moving from an idea to real-world impact is about deploying it with purpose for the benefit of your people and the people they serve.
If you'd like to watch the full discussion, it is available on demand here: https://youtu.be/Uwf26MjX-NQ?si=1Rl4kI8TQdvREt9Y