How we’re building AI to help fix the English planning system

Today, we’re sharing how we’re building a prototype AI tool that will help speed up planning approvals across England.

This tool, which will be rolled out nationally from 2027, aims to halve the time it takes a local council to review and approve requests to build extensions and loft conversions - cutting wait times from eight weeks to four. 

As a result, it will give planning officers more time to do the work they joined up for - creating vibrant communities and getting homes built quickly.

The problem we’re trying to fix

If you’ve ever dealt with the planning system, chances are you will be aware of the confusing rules, long waits and general annoyances of the system. Now imagine how frustrating it is for the officials dealing with it every day.

The tool focuses on speeding up the work of the planning officer - an official who sits in one of 317 councils across England and tackles their cut of 326,200 planning applications each year.

These officers are needed for every new home for a young family, every extension, and every warehouse expansion to help a business grow. 

Whether the application is to build a new housing estate, or to install a new window, planning officers have to check every application against a long list of policies and precedents from the different tiers of government - be it a Neighbourhood Plan, a Local Plan, or the National Planning Policy Framework. 

They also have to check the development doesn’t do things like overshadow neighbours, cause parking concerns or breach environmental regulations. 

If that wasn’t enough, they also have to consider any complaints they receive - which are now being generated by AI tools and coming in higher volumes than ever before.

Some planning applications are complex, such as decisions over whether to build new factories, business parks or housing estates. These rightly need thorough analysis before a decision is made.

But the vast majority - 70% - are for small scale renovations. This might include converting a garage or an attic, adding a bedroom, or fitting another window. 

Yet even these simple applications can take around eight weeks to complete.

Their approval can and should be more straightforward. 

That’s where we’re building an AI tool to help - and cut this eight-week wait to four.

How it will work

Together with Google DeepMind, Google Cloud and the UK Government, we’re building an agentic AI system that will pull out key data from planning applications and analyse them against dozens of plans, frameworks and rules.

It will give each planning officer a team of AI agents doing administrative work in the background for them, including:

• Checking and validating details submitted by applicants, as well as cross referencing with local maps and characteristics of the area.

• Summarising comments received from neighbours and consultees in response to the application.

• Checking information received against local and national policies to make sure it is compliant.

The tool will give planning officers a step-by-step analysis showing whether it thinks a particular application complies with national, regional and local planning policies.

It will allow them to click into each analysis to see which information the system is using to make its recommendation - so everything can be validated and audited.

It will then recommend to the planning officer that the application be accepted, accepted with conditions, or declined. 

The planning officer remains the ultimate decisionmaker. They will be able to add to, amend or overrule any recommendations. They will continue to use their judgement to manage any trade-offs and overlaps in the policies that apply in a particular case.

By speeding up this process for smaller applications - like loft conversions and driveways - planning officers will have more time to focus on progressing the major developments that improve towns and communities across the country.

What happens next

The UK Government is working with Faculty, Google DeepMind and Google Cloud to build the tool, while working hand-in-glove with local councils throughout.

Initially, we will be working with local councils in Barnet, Dorset and Camden - which together offer a variety of planning challenges that stem from rural and urban environments. 

As Faculty, we will be bringing our decade of experience in co-designing and deploying safe AI in public services to make sure the tool is designed in a way that works for the people who use it - as well as helping councils use it effectively to speed up their work.

We’ll also be drawing on our knowledge of applying frontier AI models safely, which stems from our safety work with most of the major AI labs - where we test the AI models pre-release.

Once we’re confident the tool works as it should for the three councils we’re partnered with, we will roll out the tool through a series of more ambitious tests.