Our Take
This is a competent public-sector LLM deployment with genuine workflow friction as its target, but the 50% timeline reduction is a pilot estimate, not a deployed fact, and human sign-off on every decision means the bottleneck has merely shifted, not broken.
Why it matters
UK housing shortfalls hinge on planning approval speed, not policy. If councils can actually clear householder applications (routine 70% of volume) faster, it unblocks officer time for the major developments that stall housing targets. The bet here is operational, not political.
Do this week
Public sector procurement leads: if your council is not in the Barnet/Dorset/Camden alpha cohort, file a readiness assessment now so you have documented baseline metrics before the 2027 rollout hits and vendors claim credit for any improvement.
England deploys two AI tools across all local planning authorities
The UK Ministry of Housing, Communities and Local Government (MHCLG) and Department for Science, Innovation and Technology (DSIT) are scaling two machine learning systems built on Google's Gemini foundation models to every local council in England by 2027. The first, Extract, has already rolled out nationally after trials across 20+ authorities. The second, Augmented Planning Decisions (APD), is currently in alpha testing at three councils: Barnet, Dorset, and Camden.
Extract automates data parsing. It converts unstructured planning records locked in legacy PDFs into structured datasets in minutes. Per operational data from trials, the tool eliminates roughly 255 hours of manual data entry per council annually (company-reported). APD operates as a planning officer's analytical assistant, consolidating application documents, flagging missing information, cross-referencing zoning laws with policy citations, summarising public consultation letters, and drafting evaluation reports. Human planners retain final decision authority; the system does not approve or reject applications independently.
The backdrop is real: householder applications (loft conversions, property extensions, routine domestic modifications) account for nearly 70 percent of annual submissions. Evaluating them manually requires officers to spend hours cross-referencing regional policies, historical archives, and unstructured PDFs. The government's stated target is a 50 percent reduction in decision timelines. The UK has also committed to building 1.5 million new homes by 2029, and planning approval backlogs are a documented constraint on that timeline.
The infrastructure is sound; the outcome is still provisional
The technical execution deserves credit. MHCLG and the government's Incubator for AI (i.AI) built Extract using Gemini models and hosted both tools on Google Cloud with active security controls to block prompt injection attacks and maintain data sovereignty over sensitive municipal records. The APD prototype logs its internal processing steps sequentially, creating an auditable chain of thought for every application. This matters because public sector workflows demand regulatory accountability, and an LLM that cannot explain its reasoning is unusable in a planning context where officers sign their names to decisions.
But the 50 percent timeline reduction is a pilot estimate, not a measure of deployed performance. Three councils in alpha testing are not three councils with sustained results. Moreover, the system does not automate final approval; it automates data prep and report drafting. The actual bottleneck may not be document reading time. It may be officer capacity, council risk appetite, or public consultation cycles. Shifting work from application review to report validation does not guarantee faster decisions if the validation queue is already clogged.
The 2027 rollout date is also a 3-year runway. By then, both the tools and the baseline council operations will have shifted. If housing targets accelerate before APD reaches all 300+ authorities, the AI's contribution to the win will be difficult to isolate.
Councils in pilot phase should measure before the national rollout
If you are a planning authority not yet in the APD alpha, treat the next 12 months as a window to establish rigorous baseline metrics. Document current decision timelines by application type (householder vs. major), measure how many hours officers spend on document review vs. other tasks, and track the rejection and appeal rates. When APD arrives in your authority in 2026 or 2027, you will be able to measure actual change. Vendors will claim credit for any improvement; your data will tell you whether they earned it.
For councils already in trials, publish decision timelines and officer hour allocation regularly. Transparency now sets the standard for the national rollout and prevents the common pattern of pilot success disappearing into implementation delays.