DeepMind Builds AI to Halve UK Planning Times

Google DeepMind has partnered with the UK government on an AI prototype meant to halve the time planning officers spend processing householder planning applications. It's already being trialed in three local authorities — Barnet, Camden and Dorset — with a national rollout targeted for 2027. Built with Google Cloud and the AI firm Faculty, the tool goes after the paperwork that clogs routine permits: a single small application can have an officer cross-referencing policy PDFs for hours.

The prototype doesn't decide anything. It consolidates the data on a site, pulls the exact national and local policies that apply with citations, summarizes public consultation feedback, and drafts an initial assessment report with proposed conditions. A human officer reviews and edits every output, and the system keeps an audit trail of each step. Householder applications — extensions, loft conversions, the everyday stuff — make up nearly 70% of all planning applications in England each year, so shaving hours off each one adds up fast.

Why this matters beyond Britain

If you've ever waited months on a permit to build an extension, you know planning is where good intentions go to wait in a queue. The UK has a target of 1.5 million new homes by 2029, and the bottleneck isn't always bricks — it's officer time buried in document-shuffling. Pointing AI at the administrative layer, rather than at the judgment calls, is a sensible split: let software do the cross-referencing and summarizing, let people make the decisions. It's also a template other governments will copy, because the same pattern fits permits, benefits processing, and any system where humans drown in policy documents.

This isn't DeepMind's first move into UK planning. An earlier tool called Extract, built with Gemini, converts legacy paper planning documents into digital data and is already deployed to every English council, reportedly saving each one around 255 hours a year. The new prototype builds on that digitized foundation — you can't have AI read your policies until they're machine-readable in the first place. As Naisha Polaine of Barnet Council put it, the tool's ability to collect relevant information "has the potential to save significant officer time."

The genuinely smart part here is the restraint: the AI drafts, the officer signs. That's the version of public-sector AI worth wanting — speed on the boring parts, accountability kept with a named human. The risks are the familiar ones. An AI that summarizes consultation feedback can flatten or miss objections; one that cites policy can cite it wrong. The audit trail and human sign-off are the guardrails, and they only matter if officers actually scrutinize the drafts instead of rubber-stamping them. Early trials in three councils is also a long way from a national system millions of applications flow through.