Can AI really cut planning delays for England's small...

A planning system under strain

England's planning system has rarely faced more scrutiny than it does today. Official figures for the year to December 2025 show councils granted 261,700 decisions, down 4% on the previous year, while the stock of applications sitting on hand stood at 105,700 at the end of the quarter — 1.4 times larger than the 76,300 applications received in that same period. Householder applications, which account for roughly half of all decisions councils make each quarter and nearly 70% of the circa 350,000 applications submitted annually, sit at the heart of this backlog.

The statutory determination periods are well known: eight weeks for most householder and minor applications, 13 weeks for major development, and 16 weeks for schemes requiring an Environmental Impact Assessment. In practice, these targets are increasingly theoretical. In the July-to-September 2025 quarter, only 41% of minor applications were decided within the statutory eight weeks, and just 19% of major applications hit the 13-week target. Performance agreements and extensions of time have become the norm rather than the exception — used in 41% of all decisions, and 78% of major decisions, in that same quarter.

For institutional investors, developers and lenders, this isn't an abstract administrative inconvenience. REalyse's planning pipeline data shows the average lag between permission granted and construction start has lengthened across most English regions over the past two years — the opposite of what planning reform was meant to deliver. A scheme granted consent today may not break ground for 18-24 months, and viability assumptions on build costs, sales absorption and financing can shift materially across that window. Against this backdrop, the government's push to bring AI into the planning process deserves serious attention from anyone underwriting UK residential exposure.

What the AI tools actually do

Two government-backed tools are central to the current rollout. The first, called Extract, uses Google DeepMind's Gemini model to convert decades of paper-based planning records — including handwritten notes and historic maps — into usable digital data in minutes rather than the one to two hours this traditionally takes. Trialled across 20 local planning authorities, Extract is now available to every council in England, with government estimating it could save around 255 hours of manual work per authority and roughly 250,000 hours of officer time nationally each year.

The second, a prototype known as Augmented Planning Decisions (APD), is more directly aimed at the householder backlog. Built with Google DeepMind, Google Cloud and AI specialist Faculty under an £8.2 million contract, APD triages incoming applications, summarises key information and drafts an initial assessment for the planning officer to review. The stated ambition is to halve average decision times on householder applications — extensions, loft conversions and similar routine work — from eight weeks to four. Alpha trials began in May 2026 across Barnet, Camden and Dorset councils, with expansion to up to ten further authorities expected later in the year and national rollout targeted for 2027.

Crucially, every one of these assessments is reviewed and signed off by a qualified planning officer before a decision is issued. The government has been explicit that this human-in-the-loop design is deliberate, partly to preserve planning judgement and partly because UK GDPR's Article 22 restricts decisions based solely on automated processing where they produce legal or similarly significant effects on individuals. Separately, a handful of local authorities — Lewisham among them — have piloted AI-based site-identification tools to flag small sites with latent housing potential, with one pilot identifying over 3,000 such sites, though officers stressed the outputs required manual review before being acted on.

Where the time actually gets saved

The case for AI here rests less on reinventing planning judgement and more on compressing the administrative layer around it. Much of the current delay in routine applications isn't about contested planning merit — it's about validation checks, document handling, policy cross-referencing and report drafting, all traditionally done by hand. Industry data suggests roughly one in five householder decisions is issued on the final day of the standard eight-week window, which points to process bottlenecks clustering at the end of the cycle rather than genuinely difficult cases dragging on throughout.

Where digital tools have already been deployed at pilot scale, the time savings look material. Case studies published by MHCLG point to East Hertfordshire cutting call-for-sites processing time by 60%, North Tyneside saving 50% of staff time on consultation analysis, and Greater Cambridge achieving 50% time savings on site assessment. None of these figures relate to full decision-making automation — they relate to the unglamorous middle stages of the process: collating evidence, checking policy compliance, and preparing material for an officer's judgement.

This distinction matters for how developers and agents should read the current rollout. Even a successful APD programme is targeting a halving of householder decision times, from eight to four weeks — a meaningful but incremental gain, and one confined for now to the simplest end of the application spectrum. For context, REalyse data on England's wider planning pipeline shows the real drag on delivery sits further upstream, in major and minor schemes where determination times regularly stretch past 300 days in some authorities and only around a fifth of major applications meet the statutory 13-week window. AI triage for householder work may free up officer capacity, but on current evidence it is not yet positioned to touch the structural backlog in major residential consents that drives most institutional underwriting decisions.

The safeguards institutional users should be watching

For lenders, developers and agencies relying on planning data to underwrite deals or advise clients, three safeguards stand out as preconditions for trusting AI-assisted planning outputs at scale.

The first is data governance around sensitive information. Planning application data can include personal details about applicants' health, family circumstances and financial arrangements submitted in support of applications. The involvement of global technology partners in processing this data invites legitimate scrutiny of data processor agreements and international transfer arrangements under UK GDPR, particularly as these tools scale beyond a handful of pilot councils.

The second is consistency of outcomes. England's planning system already shows striking authority-to-authority variation — approval rates for major schemes have been reported ranging from the high-20s to the high-90s by percentage across councils, reflecting differences in staffing, local policy stance and political appetite for development. AI tools trained or calibrated differently across authorities, or deployed unevenly as some councils adopt early and others lag, risk entrenching rather than narrowing this variation unless rollout is managed with consistent national standards.

The third is the erosion of officer training pathways. Industry commentary on the APD pilots has flagged a genuine concern: junior planning officers traditionally build judgement by working through householder and minor applications, learning how to weigh material considerations on relatively low-stakes cases. If AI absorbs that layer of work before officers have developed the underlying expertise, the profession risks a skills gap further down the line, even as current-generation officers benefit from reduced admin burden today.

Outlook

The direction of travel is clear: AI is being positioned as an administrative accelerant for England's planning system, not a decision-making replacement, and the government's own framing — human sign-off on every case, audit trails for every AI-assisted assessment — reflects an attempt to balance speed against accountability. For routine, high-volume householder work, the tools in testing look genuinely capable of compressing weeks of waiting into a faster, more consistent process.

The harder question for institutional stakeholders is whether this translates into the metric that actually matters for development economics: the gap between permission granted and homes delivered. On current evidence, that gap has been widening, not narrowing, even as application volumes hit a multi-year high. AI triage for small schemes may ease pressure on planning departments and free up officer time for the larger, more consequential applications that drive housing supply — but closing the delivery gap will likely depend on how that freed-up capacity is redirected, and whether similar digital tooling can eventually be extended to the major-scheme backlog where the real viability risk sits.