In the spotlight: AI transcription tool frees up time for probation officers

Background

Probation officers spend much of their working day meeting people under supervision, assessing risks and supporting rehabilitation. But every meeting also generates paperwork, with officers required to record discussions and maintain accurate case records.

Across England and Wales, the Probation Service conducts approximately 3.7 million supervision meetings each year. Writing up those meetings takes time that could otherwise be spent working directly with people on probation.

The challenge has become more pressing as the service faces significant staffing pressures and increasing demand. According to the National Audit Office (NAO), only 74 percent of probation officer-grade posts were filled at the end of June 2026.

As part of its Our Future Probation Service programme, HM Prison and Probation Service (HMPPS) is seeking to free up operational capacity by 25 percent by April 2027.

One initiative supporting that ambition is Justice Transcribe, an AI-enabled tool developed by the Ministry of Justice (MoJ) and HMPPS to transcribe and summarise supervision meetings.

Challenge

Traditionally, probation officers took handwritten notes during meetings or relied on remembering important details before recording them in the national case management system. The process was time-consuming and could affect both the quality of records and officers’ ability to engage fully with the people they were supervising.

The idea of using AI to address the problem came from frontline probation teams, which had identified transcription and summarisation as an opportunity to reduce their workload.

MoJ and HMPPS recognised that even relatively modest time savings could have a significant impact when applied across millions of meetings each year. However, any solution would need to handle sensitive information, produce reliable summaries and retain human oversight of official case records.

Solution

Rather than developing an entirely new application, the MoJ’s Justice AI Unit adapted Minute, an existing AI-powered meeting transcription tool developed by the government’s Incubator for AI (i.AI), now part of the Government Digital Service.

Working with HMPPS digital teams and probation practitioners, the unit tailored the technology to the needs of the Probation Service. This included developing templates for supervision meeting summaries, refining AI prompts to capture relevant information and simplifying the user interface. The tool was also designed to work during virtual meetings and in locations without Wi-Fi connectivity.

During a meeting, a probation officer records the conversation using Justice Transcribe. The system generates a transcript and draft summary, which the officer must review and edit before saving the final record to the national case management system.

Safeguards were incorporated into the process. Audio recordings are deleted once the transcript and summary have been generated, while temporary transcripts and draft summaries are automatically deleted after 30 days.

The project was overseen by operational and technical teams, with input from legal, risk, commercial, data protection and data ethics specialists.

An initial pilot involving 40 probation officers across Kent, Surrey, Sussex and Wales took place between May and July 2025. Of the 30 officers who completed the final evaluation survey, participants reported an average 73 percent reduction in the time needed to write up supervision notes. Reviewing and editing an AI-generated summary took around five minutes.

The pilot also found that 25 of the 30 respondents reported better engagement with people on probation, as they could concentrate on conversations and non-verbal cues rather than taking notes.

Meanwhile, 29 of the 30 officers believed the quality and completeness of their meeting records had improved.

However, testing also identified risks, including the possibility of officers relying too heavily on AI-generated summaries without adequately checking them. HMPPS assurance checks found inaccuracies in 20 percent of the final case records examined, primarily involving misspellings of names and agencies. The tool also experienced difficulties understanding some accents and distinguishing between speakers.

Following further testing and development, Justice Transcribe was rolled out nationally to around 11,000 probation staff between February and March 2026.

Implementation was supported by regional teams, training videos, written guidance, on-site support and regular opportunities for staff to provide feedback.

Results

Justice Transcribe has delivered substantial estimated time savings across the Probation Service.

According to the NAO, between 7 October 2025 and 14 September 2026, the tool supported more than 1.6 million meetings. HMPPS estimates that this saved approximately 266,667 hours of staff time over the period and forecasts annual benefits of around 450,000 hours for frontline probation officers.

These figures are estimates based on usage data and assumptions about the time saved per meeting, rather than direct measurements of productivity gains. Staff feedback has nevertheless been positive. Justice Transcribe achieved an average rating of 4.7 out of five stars across approximately 63,000 in-app reviews up to 2 June 2026.

Probation officers also reported being able to spend more time engaging with people under supervision and less time on administration.

The programme is expected to cost £7.9 million between 2025–26 and 2028–29.

The NAO highlighted the importance of further evaluation to establish the full productivity impact and ensure that time saved on administration translates into improvements in operational capacity and service delivery. It also identified outstanding assurance challenges, including the absence of an established mechanism for monitoring the accuracy of original transcripts.

Next steps

Following national rollout, MoJ and HMPPS are working to embed Justice Transcribe into routine operational and assurance processes. This includes developing ongoing monitoring of performance, record quality and the accuracy of AI-generated outputs, alongside examining how staff behaviour and model performance may change over time.

The organisations are also considering how the tool can be incorporated into existing quality assurance arrangements, particularly given the importance of human review in ensuring that official case records are accurate.