2024/2025 Police STAR Fund: project outcomes
The Police STAR Fund is an annual innovation call, run by the Office of Police Chief Scientific Adviser (OPCSA). Each year it funds local forces to innovate and try new things to improve their service to the public. Collaboration is highly encouraged, with close partnerships with academies across the UK.
An exciting range of projects, pertinent to policing challenges, are wrapping up for our 24/25 cohort, and new projects have now been selected for 25/26. Click here for the full list.
If any of these projects strike your interest, please get in touch via:
policestarfundenquiries@npcc.police.uk.
South Yorkshire Police, Fuzzy Labs, Microsoft, Sheffield Hallam University
This project explored how Large Language Models could support police officers in preparing case files, a time-consuming administrative task that can detract from frontline duties. The aim was to reduce manual workload while improving reporting accuracy and consistency, offering the police force a cost-effective alternative to hiring more staff.
Method
A prototype was developed using open-source, licence-free Large Language Models (LLMs) tools, securely integrated within South Yorkshire Police’s records management system data. The AI assistant was designed to summarise case information, identify key individuals and events, assess evidence, and flag inconsistencies. It also supported officers in reviewing their own statements by analysing MG11s (witness statements) and highlighting areas that needed further development.
Results
The project produced a tool to technological readiness level (TRL) 4 which summarised complex case information in both detailed and brief formats, reduced time spent on case preparation and improved file quality. It maintained data security by operating within police systems and avoided reliance on commercial vendors. The researchers estimate that, with further development, the tool could halve the time officers spend preparing case files. The project has been packaged into a GitHub repository to support future development and deployment into other forces. Using funding provided by the National Science and Innovation Board (NSIB) the tool has been trialled in another force in partnership with Derbyshire Constabulary. Next steps include seeking funding to enhance transcription capabilities and validating the tool and productivity gains in an operational environment.
Dorset Police, Bournemouth University and University of Suffolk
Dorset Police’s High Harm Unit (HHU), an Integrated Offender Management (IOM) Unit, is designed to improve the identification, assessment, and management of perpetrators of repeat domestic abuse, stalking, and rape and serious sexual offences (RASSO). The HHU aims to enhance public protection through targeted engagement and enforcement beyond traditional investigative practice. This project aimed to identify best practices, suggest improvements, and offer recommendations for other forces planning similar initiatives.
Method
The evaluation used qualitative and quantitative methods, including interviews with officers, observations of triage and High Harm Perpetrator Panel (HHPP) meetings, and analysis of force documents and operational practices. The HHU employs a dedicated offender manager and multidisciplinary teams to broaden the scope of traditional IMO Units. The team assess and manage risk, using both engagement (e.g. referrals) and enforcement (e.g. protection orders).
Results
The HHU demonstrated clear effectiveness: six of eight tracked offenders showed reduced offending, with overall offences down 55%. Three committed no further offences. The triage process improved referral handling, and engagement levels were high. However, risk assessment remains complex and time-intensive, highlighting the need for a bespoke decision-making tool. The unit’s success also included safeguarding outcomes such as Clare’s Law disclosures and support interventions. The HHU is expanding its capacity and strengthening partnerships, with recommendations for broader monitoring and enhanced external support.
West Yorkshire Police
With over 785,000 illegal plants seized nationally in 2021–22, current disposal methods are both costly and carbon-intensive. This project explored sustainable alternatives to incinerating cannabis plants seized during law enforcement operations. The initiative investigated whether composting could safely degrade THC and produce usable compost, drawing on relevant research to inform its approach.
Method
In collaboration with the University of Leeds, the project team developed a methodology for preparing cannabis for composting, testing various organic mixes to optimise THC breakdown. Compost samples were analysed for microbial structure, gas emissions, and viability. Environmental impact comparisons were made with traditional incineration methods. Anaerobic Digestion (AD) was also assessed as a potential alternative.
Results
The findings demonstrated that cannabis composts effectively, with THC levels reduced to below 0.1%. Composting under optimal conditions proved significantly more sustainable than incineration, although suboptimal conditions did result in nitrogen oxide emissions. AD emerged as the most environmentally friendly option, with methane production offering energy recovery. Notably, the inclusion of cannabis in food waste AD did not affect methane yield, suggesting a viable, low-cost disposal route for police forces.
Gwent Police, University of the West of England, Cyfannol Women’s Aid
This evaluation into policing responses to sexual violence reported by sex workers in Gwent aimed to understand investigative practices, community dynamics, and risk management processes, with a focus on trust, disclosure, and safeguarding. The study brought together police, support services, and academic partners to assess how a force engages with sex workers, who are a highly stigmatised and vulnerable population.
Method
A mixed-method approach was used to explore three core questions: what is the nature of police–sex worker relationships; what are the barriers to progressing criminal investigations where the victim is a sex worker; and how effective are the tools used to identify, mitigate and manage risk of violence towards sex workers in Gwent. Data was gathered through interviews, multi-agency observations, and analysis of safeguarding forums.
Results
Findings revealed low trust in police among sex workers, though the Sex Work Liaison Officer (SWLO) played a key role bridging the gap between sex workers and the police, repairing the ‘trust-deficit’, and providing trauma-informed support. Officers expressed a need for more training and recognised the complex vulnerabilities of sex workers. Multi-agency coordination was valued, but risk management lacked a dedicated assessment tool. Recommendations include developing a bespoke risk assessment for sexually exploited adults and enhancing force-wide awareness and training.
NPCC SOC Portfolio, University of Portsmouth, Greater Manchester Police, Essex Police, Gwent Police, Devon and Cornwall Police
Despite ongoing efforts, representation within specialist areas of policing remains uneven, often focusing on surface-level characteristics without addressing deeper cultural and structural barriers. This project aimed to gather a clearer picture of the enablers and deterrents of accessing Serious and Organised Crime (SOC) roles and perceptions of inclusivity within these roles.
Method
The research employed a mixed-methods approach, including a Rapid Evidence Assessment, 74 interviews, and a survey of 510 respondents across four forces. It explored both surface-level diversity (e.g. gender, ethnicity, age) and deep-level diversity (e.g. values, personality, cognitive styles). The study examined perceptions of SOC roles, recruitment pathways, and organisational culture.
Results
Findings revealed a disconnect between external perceptions of SOC as elitist and inflexible, and internal experiences of flexibility, autonomy, and strong team dynamics. Recruitment was seen as opaque and reliant on informal networks, limiting access for diverse candidates. Officers face trade-offs between specialisation and promotion, while staff report limited progression routes. Leadership was identified as key to promoting inclusion, though concerns remain about superficial diversity efforts. The project team are working with the College of Policing, Regional Organised Crime Units, forces and various national working groups, to promote and implement the recommendations.
Yorkshire and Humberside Regional Organised Crime Unit, CENTRIC, Sheffield Hallam University, IIS
The creation and management of online accounts can consume significant police resource. This project aimed to identify evidence-based principles that the police force could adopt to improve the survival rate of these accounts, which are often just 2%. Project Orion gathered evidence and developed recommendations for online account management, improving policing capability and risk management across force operations.
Method
A comprehensive mixed-method approach used sequential and complimentary phases, across 12 sprints, gathering data from both police staff experience (surveys, focus groups and interviews); and legislation (policy and training analysis). The project used systematic process flow evaluation and testing regimes to explore account maintenance success rates, with multiple variables across various online sources. Workshops were used to design new wireframes for solutions to improve these rates.
Results
Findings revealed algorithms and verification requirements as challenges to account maintenance. Time (the ability to plan and execute layered account activity) was identified as crucial for account survival but often overlooked. Varied training, staff experience and lack of senior leader knowledge were highlighted as barriers to successful maintenance. The project provided policy recommendations and reference guides for leaders and practitioners to fill these knowledge gaps. In early stages, adoption of best practice recommendations has seen account survival rates increase to 40%. Police STAR funding for 25/26 has been allocated to develop law enforcement’s ability to utilise generative AI to support investigations.
West Mercia Police, Swansea University
The DRAGON-Spotter project addresses the growing challenge of child sexual exploitation (CSE), particularly online grooming (OG), by enhancing digital forensic capabilities and expediting data retrieval. Developed by Swansea University’s DRAGON Team, the tool uses AI to detect OG content in chatlogs. Benefitting from a novel integration of linguistics and Machine Learning, the tool also identifies language-in-context patterns in the chatlogs, rather than specific keywords, this can improve both the speed and outcome of CSE investigations.
Method
The project aimed to test the functionality of DRAGON-Spotter on operational police data at West Mercia Police force. Specifically, to test user interactions with the DRAGON Spotter UI to collect a small sample of training data that would be used to test continuous training and fine-tuning of the AI model, thereby allowing domain experts in digital forensics to influence and tailor Large Language Model (LLM) technology to detect OG effectively.
Benefits
A major area of progress was the integration of DRAGON-Spotter into existing digital forensic (DF) workflows. Significant effort was invested in ensuring the tool could ingest the correct data formats used in forensic investigations and present the information in a way that could be cross-referenced with validated forensic tools. This included adapting the interface to align with forensic methods and evidentiary standards, ensuring traceability and compatibility with existing forensic software. These developments were critical in making the tool operationally viable and for ensuring that outputs could be trusted and verified back against the wider case data.
West Yorkshire Police (Yorkshire and the Humber Regional Scientific Support Services), University of Staffordshire
This project investigated the evidential value of shoe uppers captured in CCTV footage, focusing on whether class and individual characteristics, such as brand, creases, or damage, can support the interpretation of forensic comparisons in criminal investigations.
Method
Researchers examined environmental factors affecting image quality, including lighting, distance, and angle. The team also analysed the commonality and location of features like stains and dents and explored how materials and dyes behave under near infrared (NIR) light to predict colourways.
Results
The study produced over 10,000 images from 1,000 shoes, identifying 7,600 annotated characteristics. Outputs included a footwear characteristics database, image quality tool, and forensic insights. Phase 3 will focus on operationalising findings, developing a machine learning tool, and ensuring courtroom robustness.
Avon and Somerset Police, IBM
This project examined the feasibility of developing generative AI use cases from a single core platform, IBM Watsonx, with the aim of realising cost savings and operational efficiencies while upholding data sovereignty and security.
Method
The project team developed an AI platform tailored to the requirements of UK policing, with particular emphasis on data sovereignty, security, and future scalability. The work addressed two critical challenges within policing. The first concerned the Contact Centre, where around 60% of non-emergency calls were triaged by the switchboard, yet reliable management information was lacking. The second focused on the categorisation of crimes in line with Home Office Counting Rules.
Results
The project demonstrated promising results across both cases. For the First Point of Contact (FPOC) case, the AI-driven voice-to-text solution categorised emergency calls and generated actionable insights from data that had previously gone unrecorded. The system processed 44,700 call recordings, surfacing usable management information that would otherwise have required over 13,000 hours of manual review. An AI-enabled crime classification solution analysed low-level crimes and categorised them in accordance with the Home Office Counting Rules. This model achieved an accuracy rate of over 96%, demonstrating the feasibility and the significant potential of generative AI to support policing, while delivering measurable efficiencies and benefits.
City of London Police, Crest Advisory
While there is a growing body of evidence on fraud victimisation among adults, far less is understood about the experiences of children and young people (CYP). This gap is significant, as emerging research, including Crest’s previous work, suggests that children are disproportionately at risk of becoming fraud victims compared with other age groups.
Method
A mixed-methods approach, combining both qualitative and quantitative analysis, was used to build the most comprehensive understanding of CYP fraud victimisation to date. This involved engagement with a range of institutions, direct consultation with young people and parents through an extensive survey, and analysis of National Fraud Intelligence Bureau (NFIB) data covering fraud reports involving victims aged 10–21 in England during 2023.
Results
According to the survey, 88% of CYP aged 13-21 had been targeted by fraud in the last year. Only 1% of children and young people (CYP) have never encountered fraud. Nearly a third (29%) of CYP have been a victim of fraud. Some CYP are more likely to experience fraud victimisation, including older CYP (aged 18-21) and CYP with Special Educational Needs and Disabilities (SEND). An Action Plan was designed to ensure recommendations can be adopted by key partners to strengthen responses to CYP fraud victimisation.
*Please note, Action Fraud and National Fraud Intelligence Bureau (NFIB) were replaced by a new service in December 2025.
Heddlu Dyfed-Powys Police (HDPP), National Crime Agency
One in four Rape and Serious Sexual Offence (RASSO) suspects has prior links to sexual offending (Stanko, 2022). While tools exist for managing convicted offenders, none guide decisions for those released under investigation (RUI) or with no further action (NFA), and bail conditions rarely follow evidence-based risk assessments. Officers lack effective frameworks to identify high-risk perpetrators, leaving a critical gap in safeguarding and offender management.
Method
The project team developed and tested a novel evidence-based framework to support police in prioritising responses to RASSO suspects and mitigating future risk. The method comprised three strands: EMBED (testing reliability, validity, and officer utility), EVOLVE (mapping links to existing projects), and ENGAGE (planning force and national rollout). Project goals were to test the framework’s ability to identify high-risk perpetrators, assess its impact on policing practice and officer confidence, and train HDPP officers in its application. Whilst the framework is being piloted within Dyfed Powys Police, national rollout is not being explored at this stage.
Results
The framework and its tactical menu were further refined and developed through testing. Key innovations included: creating and validating an evidence-based policing framework, embedding it within policing practice, integrating expertise across policing portfolios to address the management of non-convicted perpetrators, and linking investigative work with sex offender management.
National Police Wellbeing Service (NPWS) Oscar Kilo, Merseyside Police, Liverpool John Moores University
The challenge to improve the health and wellbeing of police officers and staff in England and Wales is a major issue for contemporary policing. NPWS in partnership with LJMU have developed the world’s first police specific, end to end biometric health & performance system.
Method
This comparative study took place with 120 participants using Wearable Health Technology (WHT) with data insights, education and peer support to explore exciting new ways of working that can be used to impact on this mission critical problem.
Results
It is estimated that a full wellbeing programme based on this approach could save Merseyside Police £3.9m annually through reduced absenteeism and staff turnover.
Essex Police, Greater Manchester Police, Marinus Analytics UK
The project accelerated the detection of sex trafficking by applying a new risk tool within the Traffic Jam platform to analyse vast volumes of Adult Service Website (ASW) data. The tool auto-flags high-risk indicators for victim vulnerability and potential criminality, addressing the challenge of organised sexual exploitation now hidden online. With over 15 million UK ads identified in 2023, manual investigation is unfeasible. This approach enables officers to uncover criminal activity, trace victims, and strengthen safeguarding.
Method
Using the Marinus Analytics Traffic Jam platform, data from ASW’s and sex user review sites was collected and risk-scored to prioritise indicators of trafficking. The risk model was developed with input from survivors, NGOs, and academics, then coded and embedded into the platform by software engineers. The tool was tested, refined through officer feedback, and applied operationally over eight months with interim evaluations to capture lessons and improve performance.
Results
The trial produced modest measurable outcomes but delivered valuable learning for future development. The tool demonstrated clear capability in rapidly identifying connections and automatically building networks within large datasets. However, converting these outputs into actionable intelligence requires dedicated staffing resources, which may limit uptake by police forces facing competing priorities. Greater value may come from the tools network-mapping functions, which provide clearer, more actionable insights for practitioners.
Next steps will see Marinus Analytics roll out the enhanced tool to UK police forces at no additional cost. Marinus Analytics will lead on the roll out of the tool, with training inputs delivered to end users to explain the additional functionality.
Hertfordshire Constabulary, University of Hertfordshire, OSIRT
PULSE aims to enhance digital law enforcement by identifying concealed online identities. PULSE addresses the gap in current methods of online identity verification and crime detection, which are manual, time-consuming, and limited.
Method
The project successfully developed a functional prototype designed to enhance the ability of law enforcement to detect and link pseudonymous online profiles. The tool integrated a structured intelligence data with open-source content. The tool applies content analysis, sentiment assessment, and stylometric techniques to dissect nuanced patterns in user-generated content. A key innovation within PULSE is its visualisation layer, which translates complex algorithmic outputs into interpretable formats for investigators. Building on principles from digital forensics and visual semiotics, the system offers visual clustering that makes it easier for users to identify relationships among accounts and content based on inferred authorship or behavioural similarity (Leone, 2021).
Results
Early testing showed strong analytical performance, with the system surfacing linkages that would typically take human analysts significantly longer to detect, especially in cases where usernames or profile identifiers were deliberately obscured. Some limitations were identified, particularly in relation to performance on media-heavy or short-form content platforms.
Counter Terrorism Police (CTP), Defence Science and Technology Laboratory (DSTL)
Police investigations rely on complex data from many sources, often stored in different systems that work in isolation. Analysts need expert knowledge to locate, process, and interpret this data before decisions can be made. While many tools exist to support analysis, they are often opaque, complex, and require significant expertise to use effectively. This project explores how AI could help by automating data analysis and connecting disparate datasets, while ensuring outputs are transparent, accurate, and meet strict security standards.
Method
The project team developed the government-owned Amulet system, a novel AI system designed to unlock rapid, safe, and transparent insights from complex datasets. Amulet was developed from an initial proof of concept demonstrating the potential use of LLMs within a transparent pipeline, to a robust product with extensive documentation and administrative capability. Amulet was built to intuitively explore lines of enquiry, embedding human expertise and intent at its core through a human-centred approach.
Results
Amulet was successfully configured to work with dynamic data sources and varied datasets, supporting system management and integration with open-source and locally deployed AI models. Further developments are taking place though CTP, including user studies and experiments in a fusion context with the CTP Data Science Team. We’re keen to test Amulet in diverse policing settings - please get in touch if you’re interested in collaborating.
