Using the AI Challenge to test constraints, not just technology

The constraints

They were about the policy decision made years ago that hasn’t been revisited; the governance step that exists because of an old audit finding;  the data sharing arrangement that was scoped for one purpose, but which now blocks another or the handoff between two teams that was designed around a paper form.

The technology was the reason for the conversations last week, but the constraints were what many people who we met with really wanted to talk about.

A consistent pattern in our conversations

That pattern has been showing up in many conversations we’ve had in the last few weeks with Scottish public sector organisations, and it’s one of the most useful inputs when thinking about the Futurescot AI Challenge this year. Proof of concepts do matter. The integration is rarely the hard part. What a proof of concept often exposes is everything the existing service, if there is one, has been trying to get around.

We are not the only ones seeing this

The Stanford Digital Economy Lab published research* in April looking at 51 enterprise AI deployments and reached a sharper version of the same conclusion: "The difference was never the AI model. It was always the organization."

77% of the hardest challenges in those deployments were invisible costs like change management, data quality, process redesign, rather than technical ones. The most frequent source of internal resistance was not end users but staff functions like legal, risk, HR and compliance. The model is the easy part. Everything around it is the work.

The value of the proof of concept

This is what’s so exciting about the Challenge. Getting to proof of concept stage is about having permission to cut across the parts of a service that no one had a mandate to challenge. It lets a team discuss and test a workflow assumption without committing to full redesign, surface the friction that an individual or a team or even an entire department have been absorbing for years, and put a specific question in front of governance – is this control protecting an outcome, or just preserving a process? Those are the types of questions, at proof of concept stage, that can highlight how a service can be truly transformed, adopted and scaled.

The organisations who get the most out of the Challenge tend to do one thing differently. Before anything gets to proof of concept stage, they think about current service constraints and issues that might emerge – whether they be around governance, data, operational or even organisational culture issues.

If you are scoping a submission in your organisation right now, it’s worth spending time to ensure you’re thinking about the constraints you might encounter along the way. Everything, including the development of the proof of concept, perhaps an eventual business case for your organisation, becomes easier when you have.

The Futurescot AI Challenge helps Scottish public sector organisations move AI ideas into practical proofs of concept through structured experimentation, supported by us here at Storm ID. But the real value of these prototypes is not simply proving whether AI can help drive efficiency savings, but also to test the constraints that influence public services.

Got an idea? Want to chat?

If you were at the Public Digital AI event last week and we didn’t get a chance to talk, or you weren’t there but busy scoping an AI Challenge idea and would benefit from a chat before submitting, drop me a line at ai@stormid.com. I’d be happy to catch up.

The deadline for submissions is 17 June. Enter online at aichallenge.scot

* Pereira, E., Graylin, A. W., and Brynjolfsson, E. (2026). The Enterprise AI Playbook: Lessons from 51 Successful Deployments. Stanford Digital Economy Lab, Stanford University, April 2026.