AI and data-led planning: what does ‘good’ look like?

Posted on: 2026-10-23

By Harry Quartermain, Associate Lecturer in Urban Planning, University of the Built Environment

I’ve worked as a planner in England, and in Australia. During my career I’ve been in contact with other professionals who have experience working in the USA, Canada, New Zealand, and throughout the Commonwealth.

It seems that planning systems everywhere face very similar pressures: to deliver more valuable outcomes with rapidly diminishing resources; to balance competing interests in an increasing febrile and polarised political environment, to manage unequal rates of innovation in the public and private sectors, and to adjust to a constant push to ‘reform’ the whole system.

Many of these challenges are represented within the topic of artificial intelligence (AI) and “data-led planning”, which are being seen by some as a panacea – the silver bullet that will finally ‘fix’ the system. In this ideal, decisions would be informed by structured, shareable information rather than paper files and individual judgement alone.

But is it working?

The results so far vary widely. Some countries are using AI quietly to tidy up old records; others are putting it in the driving seat and letting it issue building permits. So what does good look like, and how would you recognise it in your own system?

Digitisation or digitalisation

Two words are often used as if they mean the same thing. They do not.

Digitisation is converting something analogue into a digital format: scanning a paper plan, or turning a PDF into data a computer can read. The process stays the same; only the format changes.

Digitalisation is using that data to change how the work itself is done: checking applications automatically, sharing live data between organisations, or redesigning a service around what the data makes possible.

It helps to think of this as a journey, with each stage building on the last:

1. Analogue: paper files and individual knowledge.
2. Digitised: the same information, now in digital form.
3. Digitalised: processes redesigned around structured, shared data.

If your idea of a digital planning system is one where you can email a pdf to someone, you’re barely out of the start gates with this process.

Most planning systems exist to create real world outcomes: protect the environment, create places that people want to spent time, encourage economic activity, provide homes. Not just keep people busy emailing each other.

Planners can only start to focus on these outcomes when we stop getting tripped up the process itself.

What AI actually does in planning

What about AI?

Strip away the hype and AI does five practical jobs in planning. It reads old documents and drawings and turns them into data. It checks proposals against rules. It predicts things such as traffic or housing demand. It helps to visualise places in 3D, or via a digital twin. And it summarises large volumes of public comments on proposals.  Much like the digitalisation spectrum described above, these processes exist on an adoption spectrum:

1. Digitise: turn paper and PDFs into usable data.
2. Assist: help a planner find the relevant policies, constraints and precedents.
3. Recommend: suggest a decision for a qualified person to check.
4. Decide: the system issues the outcome itself.

None of it works without good data underneath and a clear definition of what success looks like.

International case studies

England offers one of the clearest current examples of a national ‘digital planning’ programme. The planning system is “discretionary”: each application is judged on its merits by the local council, rather than approved automatically if it meets a fixed zoning code. This fact limits the possibility of full automation because every site and application is different, and could be interpreted differently.

In June 2026, the national government made Extract, an AI tool that converts old planning documents and maps into usable data, available to every council in England. Government estimates put the saving at around 250,000 hours of manual checking a year.

A second tool goes further. Augmented Planning Decisions, built with Google DeepMind and Faculty, is being trialled in three councils on “householder applications”: small changes to a single home, such as extensions and loft conversions. It drafts the assessment, but a qualified planning officer reviews and approves every decision. National roll-out is planned for 2027, which makes England a live test of the “recommend” position.

In Singapore the Urban Redevelopment Authority (URA) has spent decades building geospatial data (information tied to a precise location on a map) into URA SPACE and is now adding an AI assistant to help applicants navigate its development control guidelines, the detailed rules on what can be built where.

Dubai sits at the far end of the adoption spectrum. Its municipality is developing a system to issue villa building permits without human intervention.

When a decision rests on a simple check against building codes or fixed technical rules, the delivery of an automated system may be possible.

In a system that requires a subjective assessment about place or character, or a judgement that weighs impact against public benefit, is it possible to create a system that provides the efficiency of a data-led decision, and the flexibility of a subjective discretionary system?

And if so, where should the line be drawn?

Even when you have decided how the system should work, moving to a new digitalised system is not straightforward.

The Netherlands merged 26 laws into its 2024 Environment and Planning Act and moved every permit and local rule into one national digital system in a single “big bang” switch-over; the reform’s delays were repeatedly linked to that new system. In Toronto, Sidewalk Labs, a sister company of Google, planned a sensor-filled “smart” waterfront neighbourhood, but the project collapsed in 2020 after public concern over who would own and control data collected in public space. Together they show that adoption can fail on technology or on trust, and that both need planning long before launch.

Five questions to ask of any planning AI tool

No real-world problems in planning are truly unique. Differences across the world come from nuance in specific legislation, and national attitudes towards risk, and the role of public opinion in the process.

As planning systems around the world adapt to the innovation wave created by AI, it’s important to remember some fundamental questions:

1. What data is the tool built on, and is that data complete?
2. Who is accountable for the final decision?
3. Can the reasoning be explained to an applicant or a neighbour?
4. How can someone challenge the outcome?
5. Has the tool been independently evaluated?

These questions matter more as AI moves from back-office digitising into decisions that affect people’s homes and neighbourhoods. They are also the questions clients, government departments and the public will increasingly ask of professionals.

Want to learn more?

Find out more about the University of the Built Environment’s MSc Urban Planning

Harry will speak furthe on the topic of AI and data-led planning at a online CPD in November, stay tune!