Whitehall says its AI will take householder planning decisions from eight weeks to four. We measured the starting line.
Quote freely with a link to this page · text & data licensed CC BY 4.0
In February 2026 the Ministry of Housing, Communities and Local Government awarded a contract worth £8.2M (£6.9M contract-notice value) to Google Cloud, working with Google DeepMind and delivery partner Faculty, to build Augmented Planning Decisions: an AI tool that triages householder planning applications, summarises the key information and drafts an initial assessment for the planning officer. Every assessment is reviewed and approved by a qualified officer, and officers remain the decision-makers. Alpha trials began in Barnet, Camden and Dorset in early May 2026. The government’s stated aim is to cut decision times from upwards of eight weeks to around four, with a decision at the end of summer 2026 on extending the trial to as many as ten more councils, and wider rollout from 2027 if the evidence supports it.
PlanningLens is not part of the trial and has no access to the tool. We measure its results the way we measure every council: from the decision record.
According to PlanningLens analysis of 7,663 householder decisions, the three councils piloting the government’s AI planning tool took a median of 7.71 weeks per decision in the two years before trials began, measured from validation date to decision date with no allowance for extensions of time. The pre-trial baseline, and a matched panel of 16 control councils, were frozen in July 2026 and published in August, before any outcome existed.
Source: PlanningLens analysis of council planning portal records · snapshot 22 July 2026 · Quote freely with a link to this page
Instalment 1, August 2026: the starting line. The pre-trial baseline for all three pilots is published below, and the control panel is frozen. Instalment 2 follows the government’s end-of-summer decision on extending the trial: if a new pilot list is named, this page will baseline those councils within days of the list being published. The scorecard then updates quarterly through the planned 2027 rollout.
Three councils, three panels, one window: every householder decision from May 2024 to April 2026, measured from validation date to decision date with extensions of time ignored. This is the number the tool has to move.
Barnet recorded 4,482 timed householder decisions in the baseline window. Its control panel: Enfield, Croydon, Hillingdon, Bromley, Lewisham and Havering, matched on authority class, decision volume and approval rate, none of them running the tool. In the two years before the trial, the median Barnet householder decision took 7.57 weeks; the panel’s took 7.86.
| Council | Median wks | Mean wks | Within 8 weeks | Within 13 weeks | Decisions |
|---|---|---|---|---|---|
| Barnet | 7.57 | 7.51 | 82.5% | 93.1% | 4,482 |
| Enfield | 6 | 6.65 | 91.1% | 97.6% | 3,569 |
| Croydon | 7.57 | 8.46 | 71.5% | 87.9% | 3,035 |
| Hillingdon | 7.86 | 8.62 | 68% | 90.7% | 3,179 |
| Bromley | 8 | 9.36 | 72.5% | 87.5% | 2,174 |
| Lewisham | 8 | 8.67 | 75.3% | 92% | 1,803 |
| Havering | 8 | 8.63 | 71.7% | 92% | 2,666 |
| Panel pooled | 7.86 | 8.27 | 75.7% | 91.6% | 16,426 |
| Council | Median wks | Mean wks | Within 8 weeks | Within 13 weeks | Decisions |
|---|---|---|---|---|---|
| Barnet | 7.43 | 7.35 | 84% | 94.1% | 2,213 |
| Enfield | 6 | 6.7 | 90.4% | 97.2% | 1,795 |
| Croydon | 7.57 | 8.31 | 73.1% | 88.4% | 1,445 |
| Hillingdon | 7.71 | 8.45 | 68.3% | 91.3% | 1,504 |
| Bromley | 8 | 9.55 | 70.3% | 86.7% | 937 |
| Lewisham | 8 | 8.75 | 77.4% | 91.7% | 931 |
| Havering | 8 | 8.88 | 68% | 90.7% | 1,297 |
| Panel pooled | 7.86 | 8.26 | 75.5% | 91.5% | 7,909 |
Camden recorded 1,366 timed householder decisions in the baseline window. Its control panel: Hammersmith & Fulham, Southwark, Kensington & Chelsea, Islington and Hackney, matched on authority class, decision volume and approval rate, none of them running the tool. In the two years before the trial, the median Camden householder decision took 8.14 weeks; the panel’s took 8.
| Council | Median wks | Mean wks | Within 8 weeks | Within 13 weeks | Decisions |
|---|---|---|---|---|---|
| Camden | 8.14 | 12.29 | 49.5% | 71.3% | 1,366 |
| Hammersmith & Fulham | 9.93 | 13.13 | 34.5% | 69.5% | 1,186 |
| Southwark | 7.14 | 8.37 | 74.8% | 90.1% | 1,615 |
| Kensington & Chelsea | 8 | 9.76 | 54.3% | 84.3% | 1,161 |
| Islington | 9 | 11.47 | 43.2% | 73.8% | 752 |
| Hackney | 8 | 10.99 | 56.8% | 80.1% | 1,220 |
| Panel pooled | 8 | 10.52 | 55% | 80.7% | 5,934 |
| Council | Median wks | Mean wks | Within 8 weeks | Within 13 weeks | Decisions |
|---|---|---|---|---|---|
| Camden | 8 | 11.24 | 55.1% | 76.6% | 657 |
| Hammersmith & Fulham | 10.14 | 13.31 | 31.9% | 69.5% | 696 |
| Southwark | 7 | 8.4 | 75.7% | 89.5% | 793 |
| Kensington & Chelsea | 8 | 10.06 | 53.1% | 82.5% | 610 |
| Islington | 8.86 | 11.31 | 43.9% | 74.4% | 515 |
| Hackney | 8 | 11.45 | 56.6% | 78.5% | 655 |
| Panel pooled | 8 | 10.82 | 53.3% | 79.4% | 3,269 |
Dorset recorded 1,815 timed householder decisions in the baseline window. Its control panel: Babergh Mid Suffolk, East Riding of Yorkshire, Bath & NE Somerset, Wiltshire and Cornwall, matched on authority class, decision volume and approval rate, none of them running the tool. In the two years before the trial, the median Dorset householder decision took 7.43 weeks; the panel’s took 7.86.
Dorset’s panel includes Wiltshire, the neighbouring rural unitary running the same two years without the tool: the county with the pilot against the county without.
| Council | Median wks | Mean wks | Within 8 weeks | Within 13 weeks | Decisions |
|---|---|---|---|---|---|
| Dorset | 7.43 | 8.3 | 75.4% | 91.3% | 1,815 |
| Babergh Mid Suffolk | 7.86 | 9.17 | 69.5% | 87.4% | 916 |
| East Riding of Yorkshire | 7.57 | 7.76 | 90.1% | 95.8% | 1,469 |
| Bath & NE Somerset | 8 | 11.4 | 51.8% | 76.8% | 1,082 |
| Wiltshire | 8 | 10 | 56.9% | 82.5% | 2,202 |
| Cornwall | 7.71 | 8.71 | 72.8% | 91.9% | 1,833 |
| Panel pooled | 7.86 | 9.35 | 68.1% | 87.2% | 7,502 |
| Council | Median wks | Mean wks | Within 8 weeks | Within 13 weeks | Decisions |
|---|---|---|---|---|---|
| Dorset | 7.29 | 8.22 | 75.8% | 92.4% | 872 |
| Babergh Mid Suffolk | 8 | 9.24 | 67% | 87.5% | 463 |
| East Riding of Yorkshire | 7.57 | 7.55 | 93.4% | 97.7% | 740 |
| Bath & NE Somerset | 8 | 11.48 | 51.5% | 77.4% | 592 |
| Wiltshire | 8 | 9.94 | 54.1% | 82.7% | 1,119 |
| Cornwall | 7.71 | 8.47 | 73.7% | 93.3% | 915 |
| Panel pooled | 7.86 | 9.28 | 67.5% | 87.9% | 3,829 |
The tables include May to July 2026, the first weeks of the alpha. We attribute nothing in them to the tool, in either direction. Alpha trials test workflow before they change throughput; most applications decided in these months were submitted before the tool arrived; and officer-assist effects reach the decision record months later. The honest expectation for this first instalment is a result close to zero, and publishing it is the point. The scorecard starts before the story does.
| Council | May to July 2026 median | Decisions | Calendar 2025 median | Decisions |
|---|---|---|---|---|
| Barnet | 7.29 | 413 | 7.43 | 2,213 |
| Enfield | 6.43 | 541 | 6 | 1,795 |
| Croydon | 7.71 | 469 | 7.57 | 1,445 |
| Hillingdon | 7.71 | 301 | 7.71 | 1,504 |
| Bromley | – | – | 8 | 937 |
| Lewisham | 8 | 259 | 8 | 931 |
| Havering | 8 | 326 | 8 | 1,297 |
| Panel pooled | 7.86 | 1,896 | 7.86 | 7,909 |
| Council | May to July 2026 median | Decisions | Calendar 2025 median | Decisions |
|---|---|---|---|---|
| Camden | 9 | 131 | 8 | 657 |
| Hammersmith & Fulham | – | – | 10.14 | 696 |
| Southwark | 7.71 | 221 | 7 | 793 |
| Kensington & Chelsea | – | – | 8 | 610 |
| Islington | – | – | 8.86 | 515 |
| Hackney | 8 | 135 | 8 | 655 |
| Panel pooled | 8 | 441 | 8 | 3,269 |
| Council | May to July 2026 median | Decisions | Calendar 2025 median | Decisions |
|---|---|---|---|---|
| Dorset | 7.71 | 191 | 7.29 | 872 |
| Babergh Mid Suffolk | 8.29 | 99 | 8 | 463 |
| East Riding of Yorkshire | 7.86 | 120 | 7.57 | 740 |
| Bath & NE Somerset | – | – | 8 | 592 |
| Wiltshire | 8 | 221 | 8 | 1,119 |
| Cornwall | 7.86 | 139 | 7.71 | 915 |
| Panel pooled | 7.86 | 579 | 7.86 | 3,829 |
Some panel cells in these three months are blank, for two different reasons. Where a council publishes recent decisions on a delay, its May-to-July record had not fully arrived by the 22 July 2026 snapshot; we print nothing rather than a partial count, and those cells fill in later instalments as the record completes. Where a council’s published feed has itself stopped updating, the cell stays blank until publication resumes, whenever that is. The baseline tables above are unaffected either way.
Official statistics measure statutory compliance. MHCLG counts an application as decided on time if it was decided within eight weeks or within an extension of time agreed with the applicant, and by that measure most already are. PlanningLens measures elapsed time: validation date to decision date, per application, from each council’s own portal, with no allowance for extensions. The two rulers answer different questions. The official one asks whether the council met the agreed timescale; ours asks how long the applicant waited once their application was valid. Both are legitimate. Only one of them can show whether decisions get faster, because agreeing an extension moves the target, not the decision.
The contract has a second boundary worth reading. The tender’s reduction targets exclude the parts of the process the tool cannot touch: the statutory 21-day consultation period, site visits, neighbour notification, responses from external consultees and committee scheduling. On the contract’s own terms, success means faster officer processing. The press framing is end to end. PlanningLens measures end to end, because that is the wait a household experiences: the consultation weeks pass whether or not an AI drafted the first assessment. If the tool succeeds completely on its contractual measure, the change residents feel may still be far smaller than eight-to-four. That gap, between the officer clock and the calendar, is what this scorecard measures.
The tool starts with householder applications as defined in the 2015 Development Management Procedure Order: the extensions, loft conversions and other small works that make up just over half of the roughly 300,000 applications decided in England each year. That is the cut PlanningLens already ranks. The scorecard applies the Eight-Week Myth methodology, unchanged, to three councils: householder applications only, validation date to decision date, extensions of time ignored, decisions taking over 364 days excluded. No new measure was invented for this report; the yardstick predates the experiment it measures.
The baseline window is 1 May 2024 to 30 April 2026, the 24 months before alpha trials began. For each pilot and each control council we publish the median and mean decision time in weeks, the share of decisions made within 8 weeks, the share within 13, and the number of decisions behind every figure. A calendar-2025 cut is published alongside as a sensitivity check. The headline is the median: it moves only when the typical application moves, and cannot be dragged by a handful of stalled cases. The 8-week share sits beside it in every table, because that is the number the government’s own target speaks to.
Three councils improving proves nothing on its own. The trial years are also the years of the national scheme of delegation, the Extract rollout to every English council, standardised planning data and backlog recovery: everything is getting faster or slower everywhere at once. So each pilot is paired with a panel of five or six control councils. England only, so every council sits in the same eight-week statutory world; the same class of authority; similar decision volumes and approval rates; and no AI tool. The controls are exposed to every other change the pilots are exposed to. If Barnet improves by two weeks and its panel improves by two weeks, the tool did nothing detectable. If the pilots pull away from their twins, the gap is the effect. One pilot resisted a clean match: no single council pairs with Camden on both volume and approval rate, so its panel carries rate through three members and volume through two. The per-council tables make the mix visible.
The panel is named in this report, before any outcome exists, and it will not change. Controls chosen after results are known can be chosen to flatter the result. Controls frozen in public, in advance, cannot. Anyone checking the arithmetic in 2027 will be checking it against a panel fixed while the answer was still unknowable.
The short versions, for anyone arriving in a hurry. The full workings are in the sections above, and these are the same answers the page publishes to search and answer engines.
Augmented Planning Decisions, built for the Ministry of Housing, Communities and Local Government by Google Cloud, working with Google DeepMind and delivery partner Faculty, under a contract worth £8.2M (£6.9M contract-notice value). It triages householder planning applications, summarises the key information and drafts an initial assessment for the planning officer. Alpha trials began in Barnet, Camden and Dorset in early May 2026.
Barnet, Camden and Dorset, chosen by the government for a spread of planning environments and systems. A decision on extending the trial to as many as ten more councils is due at the end of summer 2026. This page will name them the day the list appears, and publish their baselines within days.
No. Every assessment the tool produces is reviewed and approved by a qualified planning officer, and officers remain the decision-makers. The tool drafts; the officer decides.
That is the question this page exists to answer, and it is too early to say. The government’s target is to cut householder decisions from upwards of eight weeks to around four. The measured starting line across the three pilot councils is a median of 7.71 weeks, and the scorecard updates quarterly as decisions made with the tool reach the record.
PlanningLens measures elapsed time from validation date to decision date, with no allowance for extensions of time, for every householder decision in the three pilots and in a matched panel of control councils frozen before any outcome existed. Official statistics count an extended decision as on time; this measure does not. The full method is on this page.
Not before decisions made with the tool reach the record in volume. The government’s own gate, at the end of summer 2026, concerns extending the trial, not proving the result. Officer-assist effects show up in decisions months after they show up in workflow, so the first instalment that can honestly attribute movement is unlikely to arrive before 2027. The scorecard will be here when it does.
Everything on this page is published to be used. Quote it, chart it, republish it. All we ask is a link to this page. Figures regenerate as council portals publish new decisions; the snapshot date above tells you exactly which vintage you are quoting.
Cite in an article
According to PlanningLens analysis of 7,663 householder decisions, the three councils piloting the government’s AI planning tool took a median of 7.71 weeks per decision in the two years before trials began. The baseline and a matched panel of 16 control councils were frozen before any outcome existed. Source: PlanningLens, https://planninglens.co.uk/ai-planning-scorecard.html (data snapshot 22 July 2026).
Cite in research
PlanningLens (2026). The AI Planning Scorecard: pre-trial baseline decision times for the MHCLG AI planning pilots and matched control panel. Data snapshot 22 July 2026. https://planninglens.co.uk/ai-planning-scorecard.html
Embed the headline stat
The Find a Tender award notice (contract scope, value and the excluded process stages); the MHCLG digital planning blog, 19 June 2026 (pilot councils, trial structure, officer approval); the gov.uk press release, 17 June 2026 (the £8.2M figure and the eight-to-four target); the i.AI planning page (Extract). All PlanningLens figures: council planning portal records, methodology above.
Writing about the pilot? Localised cuts, ward detail or the underlying rows for any of the nineteen councils on this page, usually same day: mark@planninglens.co.uk.
Planning consultant? The engine behind this page also runs refusal alerts and evidence packs for any published English council: mark@planninglens.co.uk.
Want this data inside your own product? Licensing: mark@planninglens.co.uk.
National averages set expectations; your ward, council and project type set reality. Check approval patterns for your exact postcode, free.
Check my postcode