Adam Boyse started his session at Modernising Criminal Justice 2026 with two words: "Show me." As Director of Technology at the Metropolitan Police, he's focused on bringing practical AI and technology to frontline officers - not concepts or theory, but tools that officers actually use in vehicles with blue lights flashing, handling multiple information sources while managing incidents in real time.
Frontline policing is cognitively overloaded. Officers in vehicles or walking neighbourhood beats need information about their area - missing persons, wanted offenders, court warrants, vulnerability markers. Traditionally, they get briefed at the start of a shift, memorise and make notes, and recall it when needed. Adam's approach: put that information in their hands, live, on a phone.
But this isn't just about technology. It's about data that's trustworthy, processes that work, governance that moves at technology speed, and a cultural shift towards experimentation - being willing to fail, learn, and iterate.
Officers draw a geofenced area on their phone. The system searches available data and auto-filters by role. A neighbourhood policing officer sees missing persons, known offenders, warrants, stop-and-search data for their area - live. A response officer heading to an address gets not just information about that location but context: is there a history of knife crime, drug dealing, violence? It derisks the approach and gives officers as much information as possible before they arrive.
Getting this live requires integration into legacy systems without modifying them - a significant technical challenge. So the Met targets one key improvement every six weeks. Get missing person images feeding through the system? That's six weeks' work. Do that properly, then move to the next thing. Speed matters less than pace you can sustain.
Most people think of translation as a phone app. This is different: bulk translation of material seized in serious organised crime investigations. Thousands of pages from mobile phones, Telegram, Snapchat, documents, audio. Traditional human translation might handle a couple of hundred words per hour. Machine translation can process 150,000 words per minute.
But here's the critical bit: offenders use slang, texting shorthand, street language that commercial translation tools miss. The Met trains the system on that language dictionary. It's not removing human translators - those professional translators are still essential for evidential accuracy - but it gives officers a rapid gist of material so they can decide what to focus on, what to charge suspects with, what to hand over for professional translation.
The output is a flavour, not the final product. But from 6,000 pages translated in the first week, you can immediately see the power.
This is where Adam thinks AI has genuine potential. Hours of body-worn footage, CCTV, or other video needs redaction before release - faces, number plates, house numbers, phones, notepads. Officers used to drag through frame by frame, blurring manually. That's tedious and hugely time-consuming.
AI can search through video using object recognition (not face recognition, for clarity) and identify everything that needs blurring. It then tracks that object through the entire footage, applying redaction automatically. What took officers hours now takes minutes. The tool saves about 70% of time compared to manual work. Officers review the output, but the grunt work is gone.
Adam stressed: this isn't about AI for its own sake. Some applications don't need AI. Some benefit from it. The principle is to use available technology efficiently before buying something new. Don't bolt on solutions for each use case - that creates a Frankenstein architecture. Build a modular, clear stack that lets you assemble solutions relatively quickly.
Crucially: keep humans in the loop. AI removes tedium, it supports judgment, it surfaces information. It doesn't replace human decision-making or make operational choices. The officer decides. The system informs.
Adam outlined where this could go. Situational awareness could evolve into supervisor-level visibility - a map showing all officer locations, vehicle positions, crowd density at public order events. Video searching could move beyond redaction: find me the person wearing the grey hoodie. Where was the white Toyota on Friday at 5pm? It's searchable video footage, not face recognition, but it's powerful for missing persons, crime investigation.
Cross-criminal justice case file processing is another opportunity. As a case moves from police to CPS to courts to HMCTS, AI could summarise progression, flag similar cases, surface prosecution angles, assess case file quality against comparables. It could streamline that entire pathway.
Finally, AI-enabled data analytics coupled with natural language search. Once data is captured and structured, letting officers search using plain English rather than complex database queries could unlock value locked in case files.
Adam's takeaway was simple. Yes, there's technology. But it works because:
- It solves problems officers actually face
- It reduces cognitive load, not adds to it
- It respects that human judgment is the point
- It iterates quickly - six weeks, not two years
- It's designed around how officers actually work, not imposed on top
The "show me" approach means testing with real officers, watching what they do, adjusting based on that feedback. Not building something in a lab and hoping it works.
The goal isn't flashy AI. It's giving officers the information and time they need to do their job properly.