We’re working with planning officers to build an agentic AI system for them
Today, we are sharing an update on PlanAssist; the AI tool helping speed up planning approvals across England as it moves from its Alpha phase into a wider Beta phase expanding to 15 councils.
Just five months ago, we announced we were building an agentic AI decision support system with Google DeepMind, Google Cloud and the UK Government to help cut the time it takes local councils to review simple planning applications, like loft conversions and extensions, from eight weeks to four.
PlanAssist aims to do this by streamlining a confusing, sluggish system that sees planning officers spending excessive amounts of time checking that routine applications for home improvements conform with various rules, leaving homes unbuilt and construction projects delayed.
Now, we have completed the initial Alpha phase, working closely with planning officers in Barnet, Dorset and Camden to design, develop and test PlanAssist. The trial showed that the system can deliver on our ambition.
Our benchmarking and validation testing during Alpha showed a clear uplift in processing time for a simple household application. Key tasks are taking 58% less time, halving processing times for routine householder applications.
These results gave us the confidence and information needed to trial the system with a much wider group of councils and start to connect the tool directly into the systems they use day to day.
What we learned from planning officers in our first test
Alongside these promising time savings, working with planning officers in three very different councils - spanning both rural and urban environments - showed us that AI can meaningfully support the analysis behind a decision, while keeping the officer firmly in control of it. Their support and feedback following scenario testing with PlanAssist allowed us to shape the platform around their needs.
This replaced a step-by-step process, in which officers saw recommendations one stage at a time, with a more flexible approach that lets them choose what to look at and how. For example, officers can now view a site’s planning constraints, such as flood zones or conservation areas, side by side with an aerial view of the site.
We also made it easier for officers to decide which factors are material considerations for an application, such as loss of daylight to neighbouring homes, and to see them together in one place, which we call the evidence base. Once officers have reviewed and confirmed the evidence base, PlanAssist uses it to show how relevant planning policies apply.
Throughout, every recommendation comes with a clear explanation of the reasoning behind it. Officers can see exactly what the system has used to reach its conclusion, and overrule or amend it where their own judgment says otherwise.
Co-designing technology with the people that will use it is crucial to building tools that are actually helpful on the ground. It avoids the creation of systems that are disconnected from reality, and that are seen by users - in this case planning officers - as another administrative burden.
Instead, when it comes to rolling out a system like this to new users, they are much more likely to get immediate value from it - and take it up in their day-to-day work.
What’s next?
Over the next 10 months, we are expanding PlanAssist in a Beta phase to work with 12 more councils across England; a mix of district, unitary and county authorities, chosen to represent the range of planning challenges councils face nationally. We have two main goals throughout this Beta phase:
To prove PlanAssist works at scale, not just with three councils but across a genuinely diverse set of authorities, each with their own systems, policies and ways of working.
To move from manually feeding the tool static information to connecting it directly into a council’s own case management system so it can process live applications as they are received. We are starting this integration work with two councils, with a view to making this way of working available more widely as the programme continues.
Throughout this next stage, the fundamental principle stays the same as it was in the Alpha: planning officers remain the decision-makers. PlanAssist’s job is to do the groundwork so that officers can spend their time on the judgment calls that need human expertise: weighing up trade-offs, handling complex or contentious applications, and making sure communities get the homes and developments they need.
Towards the end of Beta, we will begin preparing PlanAssist for life as a production service, including the testing and infrastructure work required before any council can adopt it into their business-as-usual model.
Faculty will continue drawing on our experience designing and deploying safe AI in public services, working hand-in-glove with councils to make sure PlanAssist earns their trust and not just works in theory, but holds up in the reality of a planning officer’s working day.
