More homes finished, with caveats
Recorded new home completions in Wales rose 11% to 5,159 in the year to March 2026, according to Welsh Government official statistics. The year before was revised to 4,640. Starts rose 12% to 4,238.
The figures are likely an undercount. They come from council building inspectors and the National House Building Council, and leave out other private approved inspectors. The Welsh Government describes the undercount as small but has published no size for it, though a Welsh Government feasibility study has tested other data sources. Gwynedd did not supply data, so its figure was estimated.
In our view, starts are the better early warning, because homes are started before they are finished.
On the same release’s longer series, the year to March 2026 was the highest total since the year to March 2023. It was still below every year from April 2016 to March 2020. The total has risen, but not to its earlier level.
The newest figures are weaker. Between April and June 2026, completions were 3% lower and starts 18% lower than in the same quarter of 2025. One quarter is not a trend, and the figures are not seasonally adjusted. In our view, starts are the better early warning, because homes are started before they are finished. That is our reading, not a forecast.
Need is measured differently
How many homes does Wales need? The Welsh Government’s 2025-based estimates, revised in August 2026, suggest between 7,300 and 9,800 additional homes a year from July 2025 to June 2030. The central estimate is 8,700. On top of that is existing unmet need of 9,400 homes, covering households that are both overcrowded and concealed, and homeless households in temporary accommodation.
Need and completions are different measures, so we have not subtracted one from the other.
The revised release says newer population data mean the estimates are “likely to be at the upper end of the range”. It also says they “should not be used as housing targets”.
Need and completions are different measures, so we have not subtracted one from the other. Need counts additional housing units over years running from July, which can include homes that are not newly built. Completions count new-build homes over years running from April, and are likely an undercount.
The Programme for Government 2026 to 2030 adds a separate commitment of “at least 20,000 additional social homes by 2030”. That is a policy target, not a need estimate. In our view, whichever measure is used, Wales needs a planning system that can handle steady demand without long waits.
Decisions on time, mostly
In January to March 2026, Welsh planning authorities decided 90.5% of applications within the statutory or an agreed timescale, according to the official quarterly survey. Only 47% were decided within eight weeks without an extension. Averaging the 24 reporting authorities’ figures, a decision took 114 days, the slowest of the five quarters in the official table, which starts in January to March 2025. Denbighshire’s figures were missing.
In our view, extensions are often sensible, but the headline rate says little about how long applicants wait.
About 8,000 applications were still open at the end of March 2026, in the same survey. That is a stock of live cases, not a count of late ones.
Of the 4,516 decisions reported in the quarter, 4,088 were on time, but only 2,130 came within eight weeks without an extension. So nearly half of the decisions counted as on time relied on an agreed extension or a planning performance agreement. In our view, extensions are often sensible, but the headline rate says little about how long applicants wait.
Staffing is part of the story. A study for the Welsh Government by RTPI Cymru, published in February 2026, found that 73% of responding planning authorities had vacancies. It put the short-term need at 161 more planners, which the Welsh Government called the minimum over five years. Planning fees rose by an average of 50% for most application types from 1 December 2025. Both are older figures, included here as dated background.
What England’s AI tools actually do
England has two different AI planning tools, and they are easily confused. The first, Extract, turns old planning records, such as tree preservation orders and conservation area documents, into structured data. It has been available free to every English council since June 2026. It is not used on planning applications.
The UK Government’s own evaluation of Extract is candid. It says the tool “is not suitable to be an automated solution” and that staff must review and correct its output. It expects only minor edits in around two-thirds of cases. Its estimate that the tool roughly halves the time spent is modelled, not measured.
A claimed saving on tasks is not the same as a faster decision.
The second tool, PlanAssist, drafts assessments of householder applications, such as extensions, for officers to check and edit. It is still a trial. On 30 October 2026 the UK Government announced it would be widened from three English councils to 15 for 12 months, used alongside normal processes, with the new councils joining in three cohorts. The UK Government intends to offer it to every English council by the end of 2027.
The UK Government says early testing cut the time spent on key planning tasks by 58%. One of its developers, Google DeepMind, says more than 70% of draft reports were approved by officers with no edits. We found no published method or independent evaluation for either figure. The UK Government also says the tool is helping to halve processing times for routine householder applications from eight weeks to four. In June 2026 it called this an aim, and it has published no end-to-end figures to show it. The trade site Planning Geek notes that the 58% figure measures individual tasks, not end-to-end decision times. A claimed saving on tasks is not the same as a faster decision.
Neither tool is offered to Welsh councils. Both are for English councils. In the public sources we could open, we found no Welsh council using an AI planning tool in day-to-day casework. That is the result of our search, not proof that none exists.
The case for and against, and who decides
The case for AI in Welsh planning is strongest on capacity. Councils have vacancies, and digitising old records or drafting routine reports takes skilled time. The February 2026 RTPI Cymru study recommended digital tools “including artificial intelligence, where ethical and appropriate” to free planners to do planning. If a tool takes admin off desks, scarce planners could spend longer on complex cases.
A tool can draft a report, but a person must own the decision.
Accuracy is the first objection. Extract’s evaluation found that the same document can produce different results. Run three times, it gave at least one good output for three-quarters of documents. A wrong record could mislead a later decision. That is why both English tools keep a trained officer checking the output.
Accountability is the second. People affected by a planning decision need to know who made it and why. In Wales, the appeals body, Planning and Environment Decisions Wales, says it uses the Welsh Government’s secure version of Microsoft Copilot to summarise large document submissions. It says inspectors “do not use AI to make decisions”, and that AI is not used to produce reports or decision letters. It also warns that people who submit inaccurate evidence may have to pay costs.
In our view, that line is the right one to hold. A tool can draft a report, but a person must own the decision. AI also cuts the other way. The Planning Advisory Service says councils are receiving AI-written representations with “irrelevant or fabricated references to planning policy or case law”. A co-founder of one objection tool told the Architects’ Journal in February 2026 that it helps residents without consultants “level the playing field”.
Planners’ jobs are a fair worry. The Welsh evidence, though, points to shortage, not surplus. A UK Government minister said at the June 2026 launch that the tools are not about replacing planners. We found no credible source predicting planner job losses in Wales.
Welsh language is the test most specific to Wales. The Welsh Government-commissioned digital planning report found that, where Welsh planning content exists, it “is often weaker than the English version”. In a policy statement dated July 2025, the Welsh Language Commissioner encouraged organisations to make sure AI systems are “specifically tested in Welsh before being deployed operationally”. We found no evidence that Extract or PlanAssist has been tested on Welsh or bilingual documents. That does not prove it has not happened.
Next steps for Wales: data before tools
The bigger gap may sit underneath the software. A report commissioned by the Welsh Government, published in July 2026, found “very little standardisation of core data formats” across Welsh planning authorities. Some still manage work by email or spreadsheet. Many council systems do not yet reliably process the data the Planning Portal can already supply. The report recommends common data formats and automated validation checks. It does not recommend an AI tool.
Digital change carries its own risk. Audit Wales reported in October 2026 that Denbighshire still faced “challenges with obtaining key monitoring data” from its new planning IT system, and that its quarterly returns had gaps. The council’s own April 2026 report said the move had caused a backlog of applications, before officers reported a return to business as usual. Digital change can slow things down before it speeds them up.
The Programme for Government 2026 to 2030 commits to reform the planning system and to support the “safe and ethical adoption of AI”. It names no AI planning project. The National Infrastructure Commission for Wales has proposed that a Senedd committee inquiry into the new consenting regime for major infrastructure examine “whether digital tools and shared data are being used effectively”. We have seen no confirmation that such an inquiry will be held.
In our view, Wales should put its planning data in order first. That means common formats, automated validation and reliable national performance figures, designed in Welsh and English from the start. It would help councils even if no AI tool is ever used, and it would make any future tool easier to test.
Once that foundation exists, Wales could run an open trial of an assistive tool. It should be tested on Welsh and bilingual documents, keep an officer accountable for every decision, and publish its results against a stated method. Wales finished more homes in the year to March 2026. Whether planning keeps up depends less on buying software than on the data beneath it.