UK Government AI Tool Saves 75,000 Days in Annual Consultation Analysis

TL;DR:

  • Government’s Consult AI tool analysed 50,000+ responses to the Independent Water Commission review in approximately 2 hours
  • Tool demonstrated 83% agreement with human reviewers, outperforming inter-human agreement of 55%
  • Potential to save 75,000 days of manual analysis annually across government, reducing £20 million in staffing costs

The UK government has demonstrated practical efficiency gains from its Consult AI tool, which categorised over 50,000 responses to the Independent Water Commission review in roughly 2 hours at a cost of £240. Policy experts required just 22 hours to verify the results, enabling them to focus on analysis and recommendations rather than manual sorting of responses.

Context and Background

Consult forms part of the government’s Humphrey suite of AI tools, designed to accelerate policy development processes. In comparative testing, the tool agreed with one or both expert review groups 83% of the time, whilst the two human groups only agreed with each other 55% of the time. This represents an improvement from earlier deployments, where the tool achieved an F1 score of 0.76 on a Scottish government consultation with 2,000 responses.

The technology has now been deployed across multiple consultations, including analysis of 800 responses to the Digital Inclusion Action Plan. Digital Government Minister Ian Murray emphasised that by automating basic administrative tasks, officials can redirect their focus toward service improvement and policy action.

Market Impact: Government currently conducts consultations requiring 75,000 days of manual analysis annually. At £20 million in staffing costs, even partial automation represents significant efficiency opportunities for public sector AI adoption.

Looking Forward

The government has identified Consult as one of several “AI Exemplars” aimed at streamlining administrative processes across departments. Whilst the related Redbox tool has been discontinued in favour of commercial alternatives like Microsoft Copilot—which recent trials suggest could save officials approximately 2 weeks annually—the engineering team continues developing specialised tools for planning decisions, probation services, and healthcare administration.

The shift toward task-specific AI tools rather than general-purpose assistants suggests a maturing approach to public sector automation, prioritising measurable outcomes over technological novelty.

Source Attribution:

Share this article