DfT turns to AI to tackle consultation data overload

The UK’s Department for Transport (DfT) has rolled out a suite of AI tools designed to dramatically accelerate how it analyses public consultation responses – reducing processing times from months to hours.

The department, which manages around 55 consultations each year, can receive more than 100,000 free-text responses per exercise. Historically, analysing this volume of feedback required teams of officials to manually review and classify themes over several months, creating a significant operational burden and delaying publication timelines.

Now the DfT’s AI and Data Science Team has partnered with Google Cloud and the Alan Turing Institute to develop the Consultation Analysis Tool (CAT).

Built on Google’s Vertex AI platform and powered by Gemini models, CAT can identify and categorise themes across large datasets in just a few hours. The system has demonstrated accuracy rates of up to 90 percent, according to DfT figures, and is expected to deliver annual savings of up to £4 million.

The tool has already been used to support analysis of responses to the Integrated National Transport Strategy and consultations on driving test booking rules.

Expanding AI across transport policy

The consultation tool forms part of a broader push by the department to embed AI across its operations.

Using Google Cloud services including Cloud Run, Cloud CDN and Firestore, the DfT has also developed a Connectivity Tool aimed at helping urban planners make more sustainable infrastructure decisions.

In parallel, an AI Correspondence Drafter is being used to generate initial responses to public enquiries. The system combines Vertex AI Search, which retrieves relevant policy information from secure internal databases, alongside generative AI models to draft replies for review by officials.

Human oversight

Despite the scale and speed of automation, the DfT emphasises that human judgement remains central to its approach.

All AI outputs are reviewed under a “human-in-the-loop” model, ensuring decisions are checked for accuracy, fairness and potential bias before being used in policymaking. The department says this aligns with public expectations identified through its own research.

While Google Cloud provides the underlying infrastructure and processing capability, DfT says final decisions rest with DfT policy experts.

The department argues that this combination of automation and oversight will enable faster, more transparent engagement with citizens while maintaining trust in government decision-making.

As the volume of public data continues to grow, officials say such tools could play a critical role in helping departments meet consultation response deadlines – including the government’s 12-week publication target – while improving the quality and speed of policy insights.