Helping probation practitioners spend less time searching and more time supporting people

When probation practitioners take on a new case, they often need to quickly understand a person's history, risks and support needs. That information is stored in Delius, the main case management system used across probation, where more than 40 million contact logs are recorded each year.  

Finding the right information among thousands of records isn't always straightforward. Different practitioners may use different terminology, acronyms, abbreviations or spellings to describe the same thing, making it difficult to know exactly what to search for.  

To address this challenge, colleagues from the Delius team, Probation Data Science and Technical Architecture worked together to improve how search works within Delius, helping practitioners find relevant information more quickly and with less effort. 

Why existing search wasn't always enough 

In October 2021, Delius introduced a free text search function that allowed practitioners to search contact logs using specific words, phrases and acronyms. While useful, feedback showed that finding information often depended on knowing the exact terms used by the person who originally recorded the contact.  

For example, one practitioner might record information about "smoking" while another writes "cigarettes". Someone searching could try several related terms before feeling confident they had found everything relevant. The team saw this pattern repeatedly in service data, with users carrying out multiple searches using similar words to avoid missing important information.  

For practitioners managing risk, preparing for appointments or taking over cases from colleagues, that extra effort can add up.

Practitioners can be assigned new cases from colleagues due to resourcing, risk, and expertise. So, they are reviewing their own contacts, but sometimes contacts written by colleagues too. Over time, feedback from users showed clear pain points with the search:  

• Knowledge gaps in what to search for (e.g. legacy intervention programmes)  

• Assumed similarity bias: the false assumption that people write the same as us. This can lead to difficulty finding certain logs e.g. due to misspellings, acronyms, synonyms and practitioner experience  

 

These pain points were most clearly evidenced as “search flurries” - users typing different related terms to double check nothing had been missed in their search e.g. smoking, cigarettes.  Our teams were determined to improve this, so we brought together technical colleagues including Probation Data Scientists and Technical Architects to work collaboratively and investigate areas for improvement. 

After analysing the Free Text Search audit log, our team estimated these flurries, and search gaps more generally and found it was using up a meaningful amount of staff time.   

Probation Data Science then completed an evaluation of search models and ran a thorough ethical evaluation. The model recommendation was then passed on to the Probation Integration Team who deployed the updated search in October 2025.  

Building a better search experience 

The team explored different approaches and carried out detailed evaluation work to understand what would best meet users' needs. They ultimately introduced a new hybrid search experience that combines traditional keyword search with technology capable of recognising related meanings, common misspellings and probation-specific terminology.  

This new hybrid search offered three improvements:  

• The search was better at recognising misspellings and typos    

• The search was more likely to bring back relevant related terms and phrases with similar meaning to what you’re looking for e.g. “fire”, ”arson”, “burn”   

• The search could better recognise probation-specific acronyms like MAPPA (Multi-Agency Public Protection Arrangements)  

You can read more data science details about this project on MoJ: nDelius Contact Log Semantic Search - GOV.UK.  

Establishing trust from the start 

Responsible use of data and technology was a core part of the project  

Ethical considerations were built in from the start and reviewed throughout.We designed our product in line with the Ministry of Justice AI and Data Science Ethics Framework that was produced in collaboration with the Alan Turing Institute.  

We did a range of analyses to quantify the real impact of any potential bias on the model outputs. We investigated group associations (e.g. male / female) with concepts (e.g. violence / pacifism). We used both theoretical analysis (WEAT testing) and product-specific testing on 30,000 probation contacts, finding no overall evidence of systematic bias. 

We do ongoing monitoring and get additional feedback from users to mitigate any bias risk and improve the product. 

Putting the new search into practice 

After deployment, we did some initial monitoring for feedback from our users (probation practitioners) and the in-built thumbs up / down but those provided little context.  

Quantitative analysis revealed how difficult it can be to track time savings, due to difficulties defining and interpreting user behaviour (search flurries). The proportion of time lost on flurries and average time between consecutive searches saw small changes, although usage of Free Text Search has significantly increased over the last half year.   

 Instead, we relied on User Research to evaluate whether the semantic search model reduced search effort and cognitive load, whilst maintaining trust and confidence in the search results. We were able to validate reduced effort and load in most user cases, with trust remaining high. Although effort was not entirely removed, with it shifting to choosing the right terms and judging results, especially for broad terms.

The impact was most evident when practitioners began using the improved search in real-life situations. 

One Performance Quality Officer told the team: "I think it is amazing. It cut down that need to have to read everything." 

A Referral Hub Officer described the search as: "Pretty much instant, if the information is there." 

And a Senior Probation Officer working in Electronic Monitoring said: "Very easy to use. Our team would be lost without it." 

Three Reflections  

As with any service improvement, the launch provided valuable insights that will help shape future development.

  

1. Be agile, especially in User Research. Our initial post-deployment research discussion guide focused on staff recalling their recent searches, but we quickly adapted to observing staff do live searches to better understand their behaviour. This helped us observe more cases where the search was working well and when not, such as specific acronyms or misspellings.  

2. Benefits are not always realised immediately. It can take time for a product to be adopted and used in the way it’s intended, so metrics like time savings might be on a time lag. There was visually minimal difference in the new hybrid search compared to the exact-only search version. As a result, we saw some user search behaviour like pre-deployment.  

3. The value of good content design. We saw users struggle with the visually cluttered Free Text Search webpage, which resulted in low discoverability of help and guidance information. This is why one recommendation is co-designing the page layout for better information and feedback flows.  

The work doesn’t stop here

We jointly worked with the Delius team to prioritise next steps.   

Firstly, we are working with the Delius team to do a small-scale communications plan for promoting the free text search and particularly help and guidance as that was a shortcoming observed during user research; we will work with Regional IT trainers and Change Management teams to offer a suite of online resources and potentially in-person training.   

Secondly, we will work with Technical Architects to rollout the semantic search to a new probation digital service which launched nationally in April 2026 to help practitioners manage their caseload (Manage a Person on Probation, also called MPOP).  

We will continue to monitor usage, learn from feedback and look for new opportunities to improve the service. 

This project has shown how small improvements to everyday tools can make a real difference. When probation practitioners can find the right information more quickly and with less effort, they can spend more time focusing on what matters most: supporting people, managing risk and protecting the public.