Justice Transcribe Special

"I never cease to be amazed at how intelligent, resourceful, educated and aware probation staff are. They demonstrate ability, curiosity, flexibility, problem solving and emotional maturity every day on this blog and yet the powers that be have nothing to say. No answers, no evidence, no willingness to engage, nothing. Silence is golden, but power is strength and authority overcomes all resistance. They intend to simply bulldozer through their vision of the future and the plebs can either trail along behind or be crushed by their juggernaut." Anon

*****

Justice Transcribe in Probation

AI-powered transcription and summarisation helping probation staff spend less time on admin and more time on meaningful engagement.

Justice Transcribe uses speech recognition to transcribe, summarise, and structure records from probation supervision sessions and case interactions. What began as a pilot across Kent, Surrey, Sussex, and Wales is now being scaled, with over a thousand probation officers equipped to use the tool following an expansion announced by the Deputy Prime Minister.

Validated in real-world frontline settings, Justice Transcribe is reducing administrative burden, improving the quality and consistency of case records, and freeing up time for deeper engagement with people on probation.

*****

https://www.gov.uk/government/publications/justice-transcribe/justice-transcribe-data-7-october-2025-to-14-september-2026

"The statistics are derived from internal product analytics collected for service monitoring and improvement purposes. Data was extracted on 14th September 2026. Coverage period: 7th October 2025 to 14th September 2026.

Between 7 October 2025 and 14 September 2026, over 1,600,000 meetings were summarised using Justice Transcribe... Justice Transcribe may save around 10 minutes per meeting... Applying this assumption to 1.6 million meetings gives an indicative total of approximately 266,667 hours."

The artificial brain inside my machine says:

Justice Transcribe Overview

• Purpose: An in-house AI tool developed to automatically record, transcribe, and structure case notes from probation supervision sessions.
• Rollout: Piloted in late 2025 and scaled nationally by March 2026, equipping over a thousand probation officers.
• Impact: Surpassed 1 million uses by mid-2026, saving staff upwards of 160,000 hours and cutting administrative note-taking time by roughly 50%.
• Efficiency: Estimated to save an average of about 10 minutes per qualifying supervision meeting."

Miraculous! By 14.9.2026 the jt wunderkind has saved the work of 138 full-time equivalent staff.

"the exact multi-million pound development and contract costs remain a classified internal operational budget"

So we'll never know how much of the oft quoted £700million has been spaffed up the wall on this system... I'd suggest it will be considerably more than the £5million or so allegedly 'saved' in those assumed 260,000+ hours.

The govt has acknowledged that "The UK Ministry of Justice (MoJ) and Her Majesty's Prison and Probation Service (HMPPS) are investing up to £10 million to roll out AI-powered tools and digitize manual systems to prevent accidental prisoner releases and improve operational efficiency." But that's just the cost of "rolling it out". They also say "A broader investment of £20 million for 2026/27 targets the digitisation of paper-based prison processes".

Here's the cobbler's 2025 announcement:

https://www.gov.uk/government/publications/ai-action-plan-for-justice/ai-action-plan-for-justice

In 2021 over £100million was poured down the drain:

https://www.civilserviceworld.com/news/article/prison-and-probation-service-spent-almost-100m-on-case-management-system-only-to-scrap-it

"The case-management system was part of HMPPS’ Electronic Monitoring Programme, which sought to overhaul the electronic tagging system used to monitor curfews and conditions of court or prison orders."

And here's a 2025 article about govt AI spending:

https://www.thinkdigitalpartners.com/news/2025/09/04/uk-public-sectors-3-45-billion-ai-spend-but-its-still-only-a-fraction-of-all-it-contracts/

"UK public sector’s £3.45 billion AI spend – but it’s still only a fraction of all IT contracts"

*****
Most of the assertions you make are inaccurate and misleading. Justice Transcribe was a tool developed in-house and adapted by MoJ at what would be no appreciable cost to HMPPS. Its rollout is slow and steady with minimal disruption or fuss as far as I can tell. No waste etc. Other AI projects build upon ongoing modernisation projects so again something we might welcome. Use is careful responsible and proportionate. This type of scaremongering helps no one. There is a small friendly team in charge of this in HMPPS that you can probably contact if you need accurate details. Ive seen them present at various events. They seemed quite sensible to me and sensitive to the needs of practitioners.

*****

Since we seem to have attracted one of the team I point out. Yes, Justice Transcribe saves a lot of hours, but if the aim is simply to do less rather than improve probation practice, create more time in the room with clients, and bring a better probation system then all the IT development in the world isn’t going to make a difference. If pay, training, conditions remain low then what’s the point.

It’s only a matter of time before the current static response of “Justice transcribe was developed with an £8m initial investment” is put under scrutiny.

The key question to be put to MoJ is:

Please provide the original approved business case, investment case or equivalent financial assessment for Justice Transcribe, including the estimated development cost, implementation cost, annual operating/maintenance cost and total forecast cost over the life of the product.

Then separately:

Please provide actual expenditure incurred to date, including internal staff costs, external contractors/consultants, software and cloud services, AI/model/API usage, infrastructure, security/assurance, testing, implementation, training, maintenance and support.

And critically:

If expenditure relating to Justice Transcribe is recorded within the Justice AI Unit, wider digital/AI programme, or another cost centre and cannot be separately identified, please provide the expenditure for that programme/cost centre and explain how Justice Transcribe’s costs are accounted for within it.

We all know what happened the last time they “developed” IT.

news.bbc.co.uk/2/hi/uk_news/politics/8339084.stm

*****

Seems like the basis of an FOI. Go for it and report back the response.

*****

"Justice Transcribe was a tool developed in-house and adapted by MoJ at what would be no appreciable cost to HMPPS."

So why do enquiries lead to statements such as "the exact multi-million pound development and contract costs remain a classified internal operational budget"? MoJ/NOMS/HMPPS have no track record of success whatsoever:

See also:

https://modulardata.co.uk/insights/ai-in-probation-and-the-future-of-public-service-reform/

"Against the backdrop of budget constraints, rising demand and system strain, the Chancellor’s announcement of a new £3.25bn transformation fund marked a significant moment in the direction of reform. The first tranche of this fund includes investment in AI tools for the probation service—an area under growing pressure as the justice system teeters near gridlock."

That doesn't sound like homemade AI; more like the sticking plaster ego-driven heath-robinson nightmare that is oasys, or the massive wastes of public funds with c-nomis etc etc etc.

• 2004-2007: C-NOMIS was officially abandoned as a single, combined database concept in 2007 after its projected lifetime costs more than tripled from £234 million to £690 million... it was eventually scaled back into a revised NOMIS programme costing £513 million

• 2012-2021: HM Prison and Probation Service spent £98.2m on a new case-management system, only to scrap it before it was finished, the Ministry of Justice's annual report and accounts have revealed... It would have cost £30m to finish the project, on top of the £98.2m already spent, the MoJ said...

• and then there was the shameful vainglorious tr experiment which fleeced the public purse; maybe you heard about it?

Financial Breakdown of the Failure

• Additional Taxpayer Cost £467 million ; extra money spent by the MoJ to stabilize and prematurely end the contracts.
• Early Termination Fees - At least £171 million;
• the direct cost to taxpayers for scrapping the private contracts early.
• Fines Waived / Adjusted - £110 million; shortfalls owed by private Community Rehabilitation Companies (CRCs) to the MoJ for missing targets, which the government effectively allowed them to retain to prevent complete collapse.
• Through the Gate Services - £46million; extra stabilization funding injected to help support offenders leaving custody.

Total Contract Value Paid - Up to £2.3 billion; the final amount the MoJ paid to these companies under the failed framework.

*****

Justice Transcribe is estimated to save probation approximately 450,000 hours a year.
That seems an awful lot of hours, but with a quarter of a million people currently on probation, that figure equates too just under 2hours extra available per person per annum. That dosent seem so much. What niggles me, is that it dosent really matter how much time can be made available to the probation officer for each client if it can't be used to advance progress.

From May.

https://www.crimeandjustice.org.uk/do-we-want-high-tech-future-probation-service

*****

“Doing less” can be either a genuine improvement or a serious mistake. If technology removes administration and gives practitioners more time with clients, that could be transformational. If it simply allows them to carry another 30 cases, it hasn’t improved probation practice; it has increased the productivity of an overstretched system.

A better model would mean fewer cases, time focused where it is needed most, and technology used to support, not replace the probation officer. Success should be measured by the quality of supervision and where the client is at in a few years time, not simply its completion or compliance.

Probation cannot do this without accessible housing, employment, finance, health and community services, without it what can supervision realistically achieve? Perhaps then recall wont become the answer to every obstacle, and probation can become a system of credible supervision and support, rather than simply enforcement. For those that don’t need the supervision, then make it voluntary from the outset.

*****
In a nutshell, in a nutshell!

"Probation cannot do this without accessible housing, employment, finance, health and community services, without it what can supervision realistically achieve? Perhaps then recall wont become the answer to every obstacle, and probation can become a system of credible supervision and support, rather than simply enforcement. For those that don’t need the supervision, then make it voluntary from the outset."

And just on recall here's something to ponder on. Why is it that so few people breach their licence conditions on ROTL, when they have to return to custody anyway, whilst licence breaches on probation is astronomical?

*****
Because the probation officer is not included in the equation. We know this, and many have called to change, renew, rehabilitate what probation is to something it could be. The irony is they think those at the top are listening, and Justice Transcribe is progress. But when low- to medium-risk cases check in with a robot probation officer, when 18–25-year-olds transfer to youth justice services, and probation becomes a UK Corrections Agency focused on high-risk cases, restrictions of liberty and recalls — while the current wave of PQiPs, POs and SPOs are conditioned for that model — we’ll all sit there saying: we didn’t see it coming.

*****

I use Justice Transcribe and genuinely see the benefits, but I think there’s a much more difficult issue here than how many hours it saves. Quality of the generated case note does not necessarily equal quality of the interaction it records. That’s incredibly difficult to measure.

Justice Transcribe can turn a fairly ordinary ten-minute check-in into a beautifully structured record that reads as though a substantial piece of probation work has taken place. That’s not necessarily because anything in it is false; it’s because summarisation inevitably interprets, organises and gives prominence to things that may have occupied seconds in the actual conversation.

There’s also a potential accountability problem here. What happens if there’s subsequently an SFO and that polished record is examined months later? If the generated note suggests something was meaningfully explored when in reality it was mentioned briefly, who owns that record? Presumably the practitioner does, because they approved it. We therefore need to be very careful that saving ten minutes doesn’t inadvertently create a documentary version of an appointment that is stronger than the appointment itself.

Before Justice Transcribe, writing the record also created a natural period of reflection. You thought about what had just happened, what concerned you, what didn’t quite add up and what you needed to do next. Some of that time wasn’t “administration” at all. It was professional thinking. If AI removes the writing, we need to make sure it doesn’t remove the thinking with it.

Which brings me back to workload. Technology can remove unnecessary administration and I’m all for that. But I don’t believe technology can manufacture the thing probation is desperately short of: time. Time to know somebody, time to think about them, time to follow something up, time to speak to another agency, time to notice change and time to build the sort of relationship in which somebody might actually tell you what is happening in their life.

Ultimately, fewer cases per practitioner is what creates that space. If every ten minutes Justice Transcribe saves is subsequently converted into additional workload, we’ve gained efficiency without gaining quality.

The difficulty, of course, is the staffing crisis. You cannot simply reduce caseloads when there aren’t enough experienced people to carry the work. But that doesn’t make the underlying equation disappear.

If we want better probation work, somebody somewhere has to create the time in which better probation work can actually happen. AI can help protect that time. It cannot substitute for it.

*****
You have it in a nutshell with “Before Justice Transcribe, writing the record also created a natural period of reflection. You thought about what had just happened, what concerned you, what didn’t quite add up and what you needed to do next. Some of that time wasn’t “administration” at all. It was professional thinking. If AI removes the writing, we need to make sure it doesn’t remove the thinking with it.” It’s the nuance that is so vitally important that only the practitioner is aware of. The practitioner knows and understands the client and will pick up on those fine, subtle differences in meaning, tone and feeling which is often realised after the session during the period of reflection. It’s about not what is said, but how it is said which machines just cannot pick up on. Cue in body language here as well. AI and machines cannot recognise this.

*****
You do not appreciate it replicates the work and is a reliable assesment tool. It's consistent It's cheap and it doesn't argue . See what's coming.

*****
Oh I appreciate exactly what you’re getting at. That’s actually what concerns me more than Justice Transcribe itself.

I like Justice Transcribe as an administrative tool. I use it and can genuinely see the benefit. I become considerably less comfortable when “it saves me time writing my record” quietly becomes “it can interpret the interaction”, then “it can identify what matters”, then “it can support assessment”, and eventually “why does the practitioner need to make all of these decisions at all?”

Consistency sounds attractive until we remember that human beings aren’t consistent. Neither is risk, motivation, trauma, relationships or desistance. Sometimes the most important thing a probation officer brings to an assessment is precisely the bit that doesn’t fit neatly into the model.

And yes, “it doesn’t argue” may be the most important part of your comment. A machine won’t say, “I know what the framework says, but something doesn’t feel right here.” It won’t push back because it knows this person. It won’t tell an SPO that the proposed plan makes no sense. And it certainly won’t tell HMPPS that its latest model is wrong.

So yes, I see what’s coming. And I think plenty of practitioners see it too. The question is whether those with the power to change direction can see it, or whether they are simply too pleased to have found another plaster to stick over a haemorrhaging Probation Service.

Justice Transcribe may save time. AI may eventually save considerably more. Some of it may genuinely make our working lives easier and I welcome that. But none of it changes the fundamental problem if every hour saved is simply swallowed by impossible caseloads, insufficient staffing and ever greater demands.

People are already on their knees. No amount of AI can change that. Technology might make an overstretched service more efficient, but it cannot give practitioners the time to build quality relationships if every efficiency is immediately converted into capacity to carry more work.

AI might help stop some of the bleeding. It isn’t going to cure the wound.

*****
So on the one hand we have the "official" definition that justice transcribe is there to "transcribe, summarise and structure case records" contrasted with the reality that it "can turn a fairly ordinary ten-minute check-in into a beautifully structured record that reads as though a substantial piece of probation work has taken place."

Is this not disingenuous and dishonest? If justice transcribe is there to analyse and make things "look good" then poor probation practice, where it exists and for whatever the reasons that casuse this, gets turned into something it isn't. Surely this opens the floodgates to" deliver a fairly meaningless 10 minute check in is ok provided this is made to sound more than it actually was?"

*****
Justice Transcribe looks great from a writing point of view but it’s terrible if your job involves reading previous contacts which mine does. It’s added bloody hours of reading unnecessary shite to my day. I really don’t need every entry to start with an explanation that ‘the PP used a pro social manner throughout the interview’. Contacts are repetitive, overlong and rarely tell you what actually happened. The first SFO with Justice Transcribe is going to be interesting indeed.

*****
This is so worrying and depressing at the same time! I can see it now: in the not so distant future HMIP will pick this up with recommendations for "practitioners to complete e learning on using justice transcribe" when from the very start we identified this problem!

*****
Just got home tonight to a job offer which I will be accepting. Notice will go in tomorrow, apologies to some my colleagues picking up my caseload, to the others you made your bed siding with the junior managers expecting them to maintain your lower caseload, no child protection, no pd, no marac, no neurodiversity, drugs or sex offenders, no seeds with trainee SPOs who are scared of their own shadows and spend half the time at home and sending constant puerile emails with checklists and name and shame, What a fantastic learning curve you will be on!