Introduction: A planning system that’s broken, and an AI that might fix it
Imagine this: you’re a developer with a solid plan to build 500 affordable homes. You’ve got the land, the funding, the architects. But before a single brick is laid, you’ll spend three years navigating a planning system that was designed in the 1940s. In the UK, the average planning application takes 13 months to get a decision — and that’s just for minor projects. Major housing schemes can drag on for five years or more.
Now imagine an AI that can read a 10,000-page planning document in seconds, flag inconsistencies, check compliance with local policies, and even suggest design tweaks to speed approval. That’s exactly what Google DeepMind has been working on, and their latest research — published in June 2026 — shows that AI-accelerated planning could unlock tens of thousands of new homes without changing a single law.
This isn’t sci-fi. It’s a real, working prototype that’s already been tested on hundreds of real-world planning applications. And it could be the key to breaking the UK’s housing deadlock.
The scale of the problem: why UK planning is broken
Let’s start with the numbers. The UK needs around 300,000 new homes per year to meet demand. In 2025, we built just over 200,000. That gap isn’t about land or money — it’s about permission.
| Metric | Current state | Target |
|---|---|---|
| Annual new homes built | ~200,000 | 300,000 |
| Average decision time (major apps) | 18+ months | 6 months |
| Planning applications rejected | ~15% | — |
| Appeals lodged per year | ~20,000 | — |
Every year, thousands of viable housing projects are delayed or abandoned because local planning authorities (LPAs) are overwhelmed. They’re understaffed, under-resourced, and drowning in paperwork. A single major application can generate 50,000 pages of documents — from environmental impact assessments to transport studies to heritage statements.
Human planners can’t read all of that quickly. So they take shortcuts, make mistakes, and often reject applications for minor procedural reasons. Developers then appeal, which adds another 12–18 months. The system is a bottleneck.
What DeepMind’s AI actually does
DeepMind’s new system — currently a research prototype, not a commercial product — is designed to assist, not replace, human planners. Here’s what it can do:
- Automated document analysis: It reads planning applications (PDFs, scanned documents, maps) and extracts key information: proposed number of units, building height, density, parking provision, green space, etc.
- Policy compliance check: It cross-references the application against local development plans, national planning policy frameworks, and specific council guidelines. It highlights where the proposal meets or breaches policy.
- Inconsistency detection: It flags contradictions — for example, the transport statement says “no impact on traffic” but the environmental report shows a 20% increase in peak-hour flow.
- Recommendation generation: Based on past decisions and policy analysis, it suggests likely outcomes: “This application has a 75% chance of approval if you reduce the height by one storey.”
The system was trained on a dataset of over 1 million planning applications and decisions from across England and Wales. It uses a combination of large language models (similar to Gemini) and custom-built policy encoding tools.
In tests, the AI was able to process a typical 200-page planning application in under 30 seconds — a task that takes a human planner 8–10 hours. And it achieved 92% accuracy in predicting the final decision (approve/reject) when compared with actual outcomes.
Real-world impact: faster, fairer, cheaper
What does this mean for house-building? Let’s break it down.
Speed
The most obvious benefit is speed. If every LPA in the UK used AI-assisted review, the average decision time could drop from 13 months to 6–8 months. That’s not just faster — it’s transformative for cash flow. Developers would know sooner whether to proceed, saving millions in holding costs.
Fairness
Currently, planning decisions can be inconsistent. Two councils with identical policies might approve or reject similar applications based on local politics or workload. AI can’t eliminate human judgement, but it can ensure that every application is checked against the same rules, reducing bias and unpredictability.
Cost
The UK planning system costs over £2 billion per year to run. Most of that is salaries for planners. If AI can handle 60–70% of the administrative work, LPAs could redirect their limited human resources to complex cases, public consultations, and strategic planning. That could save hundreds of millions annually.
Unlocking hidden capacity
DeepMind’s research also found that many rejected applications could have been approved with minor changes. The AI’s “what if” suggestions — like adjusting density or adding affordable housing — could turn a ‘no’ into a ‘yes’ without compromising quality.
Potential risks and challenges
Of course, no technology is a silver bullet. Here are the main concerns:
| Risk | Mitigation |
|---|---|
| AI bias (e.g., favouring high-density urban projects over rural) | Continuous auditing and diverse training data |
| Over-reliance on AI decisions | Human-in-the-loop: AI recommends, planner decides |
| Data privacy (applications contain personal info) | Anonymisation and secure processing |
| Job displacement for planners | AI handles routine tasks; planners focus on high-value work |
DeepMind is clear: this is a tool for planners, not a replacement. The goal is to give them superpowers, not pink slips.
What this means for developers and tech companies
If you’re a UK housebuilder, this is the moment to start preparing. The technology is coming — and it will change how you submit applications. Here’s what you should do:
- Digitise your archives. AI can only process what it can read. If your past applications are paper or scanned PDFs, convert them to machine-readable formats.
- Standardise your data. The more structured your application (e.g., consistent naming of documents, clear maps, tagged sections), the faster AI can process it.
- Engage with LPAs. Ask your local council if they’re trialling AI tools. Some are already running pilots with vendors like ASI Biont, which offers planning-focused AI solutions.
- Think about compliance from day one. Use AI to pre-check your own applications before submission. That’s cheaper than waiting for rejection.
For tech companies, the opportunity is obvious. DeepMind has shown the proof of concept. Now the market needs commercial products that integrate with existing planning software, handle different UK jurisdictions, and scale to every council.
The bigger picture: AI and public infrastructure
This isn’t just about housing. It’s about how we modernise public services. The UK’s planning system is one of the most complex in the world — with layers of national policy, local plans, neighbourhood schemes, and case law. If AI can make sense of that, it can transform healthcare, social care, transport, and education.
DeepMind’s work is a template for how to do this right: start with a specific, high-impact problem; build a tool that augments human expertise; test rigorously; and publish results openly.
Conclusion: the bottleneck is cracking
The UK’s housing crisis isn’t going to be solved by AI alone. We still need land, materials, labour, and political will. But the planning system has been a wall that no one could break through — until now.
AI-accelerated planning won’t replace the debate about what kind of homes we build, or where. But it can remove the friction that stops good projects from happening. That’s how you unlock house-building at scale.
The question isn’t whether this technology will be adopted. It’s how fast we can deploy it — because every month of delay means another 10,000 families waiting for a home.
This article is based on DeepMind’s June 2026 research publication. Read the full report here.
Comments