DeepMind's PlanAssist halves UK planning assessment time

Google DeepMind has published early results from PlanAssist, an AI tool for UK local authority planning officers. It has been piloted at Barnet, Camden and Dorset councils since June.

The tool targets routine householder applications. According to DeepMind, these make up as much as 70% of planning officers' workloads. They are usually simple to approve but take up a disproportionate share of officers' time.

DeepMind built PlanAssist with the UK government, Google Cloud and Faculty. It works in four steps:

• Validation: it checks application documents and architectural drawings to confirm the required information has been submitted.
• Constraint and policy checks: it reviews the application against heritage constraints, such as tree preservation orders, and against local and national planning policy.
• History: it checks whether issues or objections have been raised before, for example by local residents.
• Drafting: it writes fully cited assessment reports. Officers can edit these before making the final decision.

In the pilots, officer assessment time per application fell by more than half. Officers approved more than 70% of the generated reports with no edits. Those reports covered questions such as whether an application complies with local planning rules and whether all required documents were submitted.

Officers still make the decision on every application. DeepMind has not said how many applications the pilots covered, how the time savings were measured, or how often reports needed substantial changes. It has also not announced plans to expand to more councils.