Building AI in Government

Insights from 133 Algorithmic Transparency Records

AI is now on every public-sector leader's agenda. The pressure to raise productivity, cut administrative burden and improve services is real — but moving from pilot to production is proving hard, and the constraints that decide whether AI scales are easy to underestimate

What’s inside

This report offers a new way to understand the AI tools government is building and how they are being delivered — through a structured reading of all 133 Algorithmic Transparency Recording Standard (ATRS) published on the registry. The report covers:

What government is building — A plain-English analysis of what has actually been built – from eligibility calculators and fraud-scoring models to chatbots and aerial-image mapping.
The technology underneath — how much is generative AI and how fast the mix is shifting.
Who builds it, and how — the split between built in-house, bought off the shelf, and hybrid — and why the hardest builds come from just a handful of teams.
The data foundation — how grounding tools in government data drives the greatest transformation – and what it takes to make that data usable.
What it means for leaders — the constraints most likely to catch you out on the way from pilot to scale

This is of course just a window into government’s use of AI, not the whole room. Not every AI project in in government has an entry on the registry — but those that are are rich in detail — and on those terms, it tells us a great deal.

Get the report

Enter your details and we’ll take you straight to the download — a 14-page PDF.

By submitting this form, I agree that my data will be processed in accordance with Kompass Consulting’s privacy policy

WHO SHOULD READ THIS?

Senior leaders and SROs

You're the ones being asked to turn AI enthusiasm into delivery. This shows the constraints that decide whether a pilot ever reaches production.

Chief Data, Digital and AI Officers

A benchmark for how your peers are really building, buying and resourcing AI — and where the scarce skills sit.

Data and data-science teams

Honest about the unglamorous truth that most of the work, and most of the cost, is in the data, not the model.

Policy professionals

See how AI actually lands in services that make decisions about real people, and what that demands of the data underneath.

Local government leaders

Councils can't sustain in-house data-science teams, so the buy-versus-build picture looks very different. This shows how.

Getting the most out of AI

AI has enormous potential to improve productivity, decision-making and service delivery. Realising that potential, however, requires more than deploying new technology. Organisations that achieve the greatest returns from AI start with the problem they are trying to solve — then take a strategic approach to investing in the capabilities that surround AI, to turn experimentation into lasting value.

To maximise the benefits of AI organisations need:

Clarity of intent

Start with the problem you're trying to solve, and be clear about what you'll build, what you'll buy, and what it will truly cost. Off-the-shelf tools are quick and cheap, speeding up what you already do; only AI built on your own data creates capability you didn't have before.

Data that's AI-ready

Government usually has the data it needs, but rarely in a state AI can read or learn from. The hard work — cleaning, linking and labelling — is what turns raw data into something AI can use to drive real value. It has to come first.

The right specialists

Getting the best from AI is like Formula 1: the driver, the designers and the engineers each master a different craft, and none can do another's job. Building, integrating and buying AI need different, scarce specialists too.

Guardrails not hurdles

Good governance isn't red tape — it's what lets you move quickly and safely. Proportionate guardrails, clear ownership, and a way to back the use cases worth pursuing. Too little invites risk; too much quietly kills innovation.

Call Us