Stop Buying AI.
Start Fixing Services.

There is a danger that we are asking the wrong question about AI in the public sector.

The conversation is often: what AI should we buy? Where can we deploy a chatbot? Which teams should have copilots? How quickly can we say we are using AI? But I think there is a much more important question: what are we actually trying to fix?

AI Is Not the Transformation

I recently read an article arguing that government needs to move beyond "bolt-on AI." The research behind it is particularly interesting.

This doesn't surprise me. We have reached the stage where deploying AI can actually be relatively easy. Transforming a service is much harder. And the two things should not be confused.

You can introduce an AI assistant without changing a process. You can put a chatbot on a website without redesigning the customer journey. You can deploy generative AI while the information employees need remains spread across PDFs, emails, spreadsheets, paper documents and disconnected systems.

Technically, you have "adopted AI." But have you actually transformed anything?

Start With the Service

My view is that the sequence matters. Technology should solve the problem. The problem should not have to justify the technology.

That is very different from starting with an AI product and searching for somewhere to deploy it.

Fix the Foundations First

One of the most interesting findings in the research is that the people delivering public services and the people receiving them appear to agree on this point.

Almost nine in ten public-sector workers surveyed also said their organisation was not yet fully able to take advantage of AI. That matters, because AI is incredibly powerful, but it is not magic.

Give AI fragmented information and it has to operate around fragmented information. Put it inside an unnecessarily complicated process and the unnecessarily complicated process still exists. Connect it to poor-quality data and you have not solved the data problem.

The Outcome Should Be Visible to the Citizen

There was another finding in the research that I found particularly important. When asked what they wanted from AI, public-sector workers were particularly interested in efficiency gains and cost savings. Citizens thought differently — their leading priorities included faster services, better public safety and fraud prevention, and easier-to-use digital services.

That distinction should influence how we think about transformation. A council might measure a successful transformation programme through reduced processing costs. A citizen experiences it differently.

Both perspectives matter. But perhaps we should start with the second one. If a transformation makes services dramatically easier for citizens while simultaneously reducing the cost of delivering them, that is a much stronger outcome than simply deploying another piece of technology.

Don't Automate the Bureaucracy

There is a wider principle here that applies well beyond AI. We should be very careful about automating processes simply because those processes already exist. Before automating something, ask why it exists.

If we automate all of those things without challenging them, we risk creating extraordinarily sophisticated bureaucracy. The objective should not be to make bureaucracy faster. It should be to remove unnecessary bureaucracy altogether.

From Transactions to Journeys

This is particularly important in public services because citizens rarely think in terms of individual systems. They think in terms of life events and outcomes.

Behind those statements may sit multiple departments, databases, forms, processes and organisations. But none of that is the citizen's problem. The opportunity for digital, automation and AI is to hide that organisational complexity from the person trying to achieve an outcome.

That is where I think some of the most exciting opportunities for AI exist. Not simply answering questions.

AI Should Be Inside the Process

The article I read made the point that AI creates much more value when it is connected directly to data, workflows and operational processes rather than sitting outside them as a standalone tool. I think that's right. But I'd take the argument one step further.

That's an important distinction, because five years from now, the AI models we use will undoubtedly be dramatically more capable than those we have today. But citizens will still want broadly the same things: simple services, fast answers, clear communication, fair decisions, less repetition, and confidence that when technology is involved in an important decision, appropriate accountability remains in place.

Stop Starting With AI

So perhaps organisations considering their next AI initiative should start somewhere different. Don't begin with "where can we use AI?"

That change in sequence sounds small. I think it could make an enormous difference.

We do not need more AI for the sake of AI. We need better services. And if AI helps us create them, then we should use it aggressively, intelligently and responsibly.