AI planning tool aims to halve council decision times

An artificial intelligence prototype designed to support routine planning decisions is being trialled by three English local authorities, as the government looks to reduce delays and make better use of planning data.

The system is being tested by Barnet, Camden and Dorset councils and is intended to help planning officers assess householder applications more quickly. According to the government, the tool can sift application documents, extract relevant information and produce an initial assessment for an officer to review.

The project has been developed with Google DeepMind, Google Cloud, Faculty and local planning authorities. Ministers say it could help reduce average decision times from eight weeks to four, although qualified planning officers will remain responsible for approving every assessment and issuing the final decision.

Householder applications account for almost 70% of planning applications in England, making them a potentially valuable area for AI-assisted processing. These cases often involve large volumes of documents, drawings, local policy information and property records, much of which must currently be checked manually.

For data scientists, the trial offers a practical example of AI being introduced into a complex public-sector decision process. The challenge is not simply generating a summary. The system must identify relevant evidence, interpret information from different document types and present its reasoning in a form that officers can verify.

Turning historic records into usable data

The trial builds on Extract, an earlier government-backed project designed to convert historic planning records into structured digital data.

Many planning departments still rely on scanned documents, old maps and handwritten records. Extract uses artificial intelligence to identify and organise information contained within these files, reducing the need for officers to search through documents or enter data manually.

The government has said the tool could save councils hundreds of hours of administrative work by turning fragmented and unstructured records into information that can be searched, analysed and reused.

This stage of the process is particularly relevant to data science. Extracting information from planning records may require a combination of document recognition, natural language processing, computer vision and geospatial analysis. The reliability of the resulting data will also influence the performance of any later decision-support system.

Poor scans, inconsistent terminology and differences between local authorities could all affect accuracy. Standardising that information while retaining its original meaning will be central to whether the technology can be deployed at scale.

Keeping people responsible for decisions

Matthew Pennycook, the housing and planning minister, said the technology would help streamline planning applications and allow officers to make quicker and better-informed decisions.

However, the government has stressed that the tool will support professional judgement rather than replace it. Every AI-generated assessment must be checked and approved by a qualified planning officer.

That distinction matters. Planning decisions can affect property owners, local communities and the future use of land, so any automated recommendation must be explainable and open to scrutiny.

A useful system will need to show which evidence it relied on, how it reached its conclusions and where uncertainty remains. Officers will also need a straightforward way to correct errors, override recommendations and record why a different decision was made.

These interactions could provide valuable data for improving the system, but they will also need to be monitored carefully. A high number of corrections, for example, may indicate weaknesses in the model, the source data or the way local planning rules have been represented.

Measuring more than speed

The government’s ambition to reduce decision times from eight weeks to four gives the project a clear headline measure, but speed alone will not determine whether the trial is successful.

Evaluation will also need to consider the accuracy and consistency of assessments, the amount of officer time saved and whether the system performs equally well across different types of application and local authority.

Other questions include how often officers accept or amend the system’s conclusions, whether particular documents or application types generate more errors and whether applicants receive clearer or more consistent decisions.

The trial therefore provides a useful case study in human-in-the-loop AI. Rather than handing decisions entirely to an algorithm, the system is designed to handle repetitive analysis while keeping responsibility with trained professionals.

That model is likely to become increasingly common across public services. Its success will depend not only on the capability of the underlying technology, but also on data quality, transparent evaluation, staff training and governance.

For planning authorities, the immediate aim is to reduce administrative pressure and shorten waiting times. For data scientists, the project offers something broader: a real-world test of whether AI can work reliably within a regulated decision process, where accuracy, explainability and accountability matter just as much as efficiency.

References

https://roofingtoday.co.uk/ai-to-speed-up-planning-decisions/

https://www.gov.uk/government/news/pm-unveils-ai-breakthrough-to-slash-planning-delays-and-help-build-15-million-homes-6-june-2025