4 min readAI, Delivery
Start with the problem, not the AI
A policing proof of concept cut about five hours per case to about thirty minutes, and set the problem-first approach behind TAGD and Cordexa.
By Dave Tormey
For a few years now I have been applying AI to real problems, first at Comtrac and now through TAGD. One early conversation changed how I think about it more than anything else.
It was about a policing problem. Without going into specifics, there was a very high volume of incidents each year, and frontline officers were spending hours on each one working through process, documentation and compliance. It was in the order of five hours per incident.
A colleague and I spent a couple of hours unpacking it, not from a technical angle but a practical one:
- Where is the time actually going?
- What decisions are being made?
- Which parts are repetitive?
We ended up designing a proof of concept that used AI to:
- listen to recorded interviews,
- identify whether certain criteria had been met, and
- draft the required documentation based on that outcome, for an officer to review and approve.
Nothing theoretical. Just the technology applied where it made sense.
What stood out was not that it worked. It was the size of the impact: work that was taking about five hours could come down to about thirty minutes. That was the point where AI stopped being interesting technology for me and became something practical.
Where AI actually belongs
Since then my focus has been less on what AI can do and more on where it should be used. That has led to work surfacing internal knowledge for tenders and compliance, building decision support for real-time environments, and turning large datasets into structured findings.
Each one reinforced the same idea. AI is most useful when it solves a real problem, not when it is added because it is new.
From TAGD to Cordexa
The same thinking runs through Cordexa, which I am building now. It is less about chat and more about getting work done: connecting systems, supporting decisions, and taking effort out of document-heavy work. It is still a work in progress, but it rests on the same principle.
Start with the problem. Then work out where AI fits. Not the other way around.