From pilot to scale: turning AI ambition into public value

Nobody in government is asking whether AI is worth building into the state’s tech stack anymore. The money's been found and ministers want results.

So why do so few pilots end up as something the public actually uses? Usually, it isn't the technology, it's how the next stage gets bought.

If we want AI to improve public services at scale, we need to think carefully about what government puts out to market, how it measures success and where responsibility for delivery sits. In short, we need to improve procurement.

Start with the outcome

The starting point should be focusing on who a service is supposed to support. What’s needed to serve them better? How will we know whether it has? And who is accountable for making that happen?

For a patient, a successful AI implementation might mean a shorter wait time for clear correspondence. For a civil servant, it might mean less time on admin and more time helping the public. Those are the outcomes that should shape the programme, rather than a focus on the technology itself.

These considerations should also shape the commercial relationship. Buying specialist expertise has its place, but paying for time alone is not the same as paying for progress.

I believe fixed-price, outcome-based contracts have an important role to play here. Where the scope and responsibilities are clearly defined, suppliers should be prepared to stand behind their delivery commitments and carry the risks they can control.

Demonstrate what works

Greater weight should also be placed on practical evidence during procurement. For example, organisations pitching for the public sector contract can showcase a demo version of the end solution, with the client having people in the room who know enough to test it properly.

I'd love to see that become the norm for AI implementations. You learn a lot more about capability from an afternoon of watching a system deal with real cases than from a 200-page bid. It also means time to value is faster and scaling up is less of a gamble, because the pilot has effectively been part of the evaluation from the start.

It's worth looking at Denmark for inspiration. The Ministry of Finance there brought state IT buying together under one approach. That meant an independent IT Council, a project model every department has to use, standard contracts built around outcomes. People hear that and assume it means more red tape, but that’s not the case. The point is that every department measures success the same way, so what works in one instance can be repeated. Increases in contract award are also directly tied to past performance, which is incredibly sensible.

Suppliers must take responsibility

Scaling pilots isn’t simply a challenge for government. Suppliers have a role to play, too.

We should be willing to prove our approach, accept appropriate delivery risk and measure success in terms that matter to the public – not hours billed. We should also help build capability within departments, leaving teams better equipped to manage and develop their services.

Above all, public sector deliveries need to be partnerships with clear accountability, not a transfer of responsibility from one organisation to another.

The ingredients for implementing AI in the public sector for civic good are already here. What's left is the building and buying. Get that right and pilot-to-scale stops being a slide in a strategy deck and becomes how things happen.