2023/2024 Police STAR Fund: project outcomes
The Police STAR Fund is an annual innovation call, run by the Office of the 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 academics across the UK.
Many outcomes from the 2023/2024 cohort have gone on to produce sizeable improvements in the way forces use science and technology. Others have tested ideas which have not proved successful, this useful knowledge has been shared across forces nationwide to prevent a waste of resources.
An exciting range of projects, pertinent to policing challenges, are wrapping up for our 24/25 cohort, and a new projects have now been selected for 25/26. Visit https://science.police.uk/opportunities/police-star-fund/ for the full list.
If any of these projects strike your interest, please get in touch via policestarfundenquiries@npcc.police.uk.
Norfolk Constabulary, Suffolk Constabulary
The primary objective of this project was to develop a machine-learning based approach for identifying drivers whose behaviour falls below expected standards and to provide a more nuanced understanding of risky driving behaviours. The former system counted the frequency of risky events and implemented interventions if the number exceeded the threshold stipulated by the Driver Standards Group. However, it was not capable of considering both frequency and context of risky events, for example, not identifying risky behaviours if overall driving frequency was low.
The project team designed an end-to-end automated process for processing large volumes of telematics data from police vehicles, whilst maintaining data quality and integrity. This involved ingesting and cleansing the data and running supervised and unsupervised machine learning models to analyse patterns within vehicle journeys.
The result of the data modelling was a reporting mechanism, run monthly, to highlight where a member of staff/officer has been identified for intervention. This involved designing an intuitive reporting interface and developing a concise communication strategy to ensure that insights were effectively conveyed to stakeholders. The Constabularies plan to refine the model by integrating road networks, improving driving qualifications usage, understanding vehicle types, and considering driver experience for more accurate risk assessments and informed decision-making.
Cumbria Constabulary, Cambridge Centre for Evidence Based Policing, Lancaster University
The project set out to tackle shortcomings in how front-line patrol officers conduct digital investigations of domestic abuse (DA) crimes and to provide training to address that gap.
The method involved reviewing 500 closed DA investigations to identify key gaps, which then informed the development of a digital toolkit - comprising instructional videos and briefing notes. The toolkit’s efficacy was tested in a randomised controlled trial with Cumbria officers, who were either given access to the toolkit or continued with standard practice.
Findings showed no significant difference in solved rates or investigation quality between the treatment and control groups, likely due to late referral to the toolkit and the limited number of response team officers available in Cumbria. Nonetheless, qualitative feedback through focus groups from officers was positive, with officers highlighting its usefulness and ease of access. They also liked the digital hygiene and safeguarding video which was designed to be shown to victim-survivors to demonstrate what could be done to address their digital vulnerabilities. The project underscored the need to future-proof the videos and is now expanding the toolkit to all response team officers and embedding it into probationer training.
Kent Police, Essex Police, BAE Systems, Canterbury Christ Church University
The Kent and Essex ILAS DA Pilot project aimed to address the growing demand for processing crime and incidents often requiring intelligence reports and partnership referrals to support safe-guarding. Current approaches to managing this demand rely heavily on staff time, with manual processes leading to frequent duplication and omissions.
The project piloted the Intelligence Led Assessment System (ILAS) over two months, an innovative tool created by BAE Systems, to proactively scan thousands of data points daily and flag individual risk elements for safeguarding and proactive targeting. The pilot’s objective was to assess the potential for the solution to identify potential victims and offenders. The method involved academic evaluation by Canterbury Christchurch University, which assessed the operational benefits and learning outcomes. The pilot ran from January 2023 until the end March 2024 with the evaluation following in June 2024.
The results indicated that the cost of implementing ILAS outweighed the limited benefits and efficiencies provided by current solutions.
West Midlands Police
101 non-emergency call volume is a challenging area for many forces in UK. The aim for West Midlands Police was to develop and implement a proof of concept (PoC) AI powered voice assistant named “Andi-Esra” to answer live 101 non-emergency calls. The project sought to enhance public service by improving response times and prioritising vulnerable callers.
Andi-Esra was designed in-house to take calls from members of the public, provide advice, allow people to ‘request a crime update’, prioritise vulnerable callers to a live agent and analyse category breakdowns of 101 calls. Once created, Andi-Esra was trialled in a controlled Proof of Concept, with live members of the public over a two-month period from December 19, 2023, to February 19, 2024. During this period, Andi-Esra handled 10% of 101 calls per day, amounting to approximately 200 calls daily. On days with low staffing, this percentage was increased to 50% of callers, providing advice, sending emails to Officers in Charge, and analysing category breakdowns of 101 requests.
The 2-month PoC achieved the following:
These results demonstrated significant improvements in service delivery and efficiency, particularly in prioritising urgent responses and providing valuable data for future improvements.
Humberside OPCC/Humber Violence Prevention Partnership
The project sought to address some of the barriers to effective information sharing by integrating local service information in a multi-agency app, which could be used for instant referrals and sign-posting in support of on-street public engagement. It also incorporated the launch of a public-facing web portal for young people, Youth Connect, which would provide direct access to some of the information collated. Building on Northumbria Police’s SOS app project, the Humber VPP aimed to enhance the app’s functionality to better guide individuals to support services in the Humber region.
Designed for police officers and partner agency staff engaging with the public, the app intended to serve as a comprehensive reference point, streamlining referrals to services and providing infor-mation on local amenities, safe spaces, and youth activities.
The app was designed to integrate data from various regional sources via a multi-agency database, allowing real-time updates by commissioning staff and partners. This would ensure front-line officers and staff always had up-to-date information, aiding local agencies in planning future commissioning activities.
In tandem, a new web portal (Youth Connect) was created for young individuals seeking guidance, assistance, and opportunities for personal development.
The project delivered learnings on the digitisation of processes, and has shaped future thinking on streamlined multi-agency apps.
The new web portal (Youth Connect) was successfully launched.
Hampshire & Isle of Wight Constabulary
This project aimed to articulate how Hampshire and Isle of Wight Constabulary was not appropriately supporting, attracting and maintaining neurodivergent officers and staff in the police service. The project captured the full ‘as is’ process for recruitment and promotion processes for neurodivergent officers and suggested a new ‘to be’ process design to better support officers with neurodivergent conditions.
The project undertook an intensive deep dive of all processes relating to recruitment, promotion and retention (including reasonable adjustments) to find out how data is managed and shared relating to neurodivergent conditions and how the processes support individuals.
The current ‘as is’ process for managing neurodivergent staff was captured using data from: HR Recruitment, HR Workforce Team, Promotion Team, Occupational Health and Positive Action Team and the Equality & Inclusion Team.
With a clear understanding of the problem, the project team created an improved ‘to-be’ process map for the recruitment, promotion and retention process for officers and staff with neurodivergent conditions. With an emphasis on detailing the events, tasks, transactions, interactions (systems & actors), decision points & outcomes/outputs to obtain, store and share information relating to an officer or staffs’ neurodivergent condition.
Essex Police, Essex Police Fire and Crime Commissioner, Nottingham Trent University, Essex County Fire and Rescue, Essex County Council
Project Minerva built on the successes of a previous project which developed a spatial mapping tool of VAWG hotspots and statistically significant neighbourhood-level and physical-environmental drivers of VAWG. This project aimed to refine and evaluate Minerva by adding new perceptions of unsafe spaces from public perception and VAWG incident data. Along with improving methods to identify and tackle VAWG in public spaces and improving problem-solving approaches for resource placement.
The Minerva Approach combined existing data sets to provide a one-stop-shop evidence base that transforms traditional hotspot mapping into a transferrable model for addressing various social issues. This has been developed through a multi-agency team, working with the Quantitative and Spatial Criminology Research Group at Nottingham Trent University. The project engages Essex communities to understand how their feelings connect with partners’ views, using a combination of data sources and policy analysis to support Community Safety Partnerships (CSPs). Tools were developed to identify where VAWG is and could occur, the drivers of spatial concentration of VAWG incidents, and where women and girls feel unsafe in public spaces.
The project has led to the creation of tools that identify VAWG hotspots across Essex, understand their drivers, and combine incidents with public perceptions of safety. It has also produced a category breakdown of 101 requests, which informs future improvements and better call-flow management.
The project’s insights have begun to change the way forces think about tackling VAWG and have been a catalyst for changing the way partners work together. The team is exploring how to scale the Minerva Approach across different challenges and locations, codifying how others can use this approach across policing and other sectors.
Commonplace Essex Fear of VAWG map identifying a participant’s reflections on a specific public space
Source: https://essexcommsafety.commonplace.is/en-GB/map/share-your-views-map
Greater Manchester Police, Design Against Crime Solution Centre - University of Salford
The POPLAR project aimed to enhance the effectiveness of Problem-Oriented Policing (POP) used in neighbourhood policing teams within Greater Manchester Police (GMP). The projects’ main objective was to develop evaluation tools and support materials that would streamline the POP process and address challenges identified in its application by GMP’s neighbourhood policing teams and Prevention Hubs.
The project employed a human-centred design (HCD) approach, involving direct engagement with frontline officers through requirements capture research and collaborative workshops. The project had initially assumed that the Assessment phase of the SARA (Scanning, Analysis, Response, Assessment) model was the main area in need of improvement. However, engagement with frontline officers revealed that challenges existed across all stages of the POP process.
This led to the development of a revised POP process, integrating simplified governance structures and quality assurance mechanisms at each stage of the Scanning, Analysis, Response and Assessment model. The project team developed a consolidated POP Plan form to reduce administrative burden, a proposal for a POP Support Hub to provide resources for problem-solving officers, and a training programme. The next steps involved seeking further funding for a pilot phase in one or more districts followed by a wider scale-up across the force.
Heddlu Dyfed-Powys Police, National Crime Agency, Bournemouth University, University of Suffolk
This project aimed to address a gap in managing non-convicted perpetrators Released Under Investigation (RUI’d) or No Further Action (NFAs) for sex offences. The project developed and piloted a structured tool to guide risk-based decision-making and inform the application of tactical options to disrupt sexual offending.
The method involved building a prioritisation framework to identify high-risk non-convicted perpetrators requiring risk mitigation and suggesting tactical options for risk mitigation. The evaluation focussed on whether this structured method changed policing practices, including the design, ability, and confidence in seeking ways to prevent offending.
Key findings confirmed the need for the framework and its practicality in an operational environment. Innovations included the development of a holistic and multifactorial evaluation framework, drawing on investigative policing expertise with multiple policing portfolios and linking traditional investigative work with sex offender management.
West Midlands Police (WMP), Middlesex University, Bournemouth University, University of Suffolk
This project aimed to investigate how to build and evaluate algorithms that prioritise and predict potential high-harm offenders from individuals suspected of stalking and harassment, or with previous stalking and harassment charges. The team also investigated how algorithms can be best implemented in force and developed a framework for prioritisation algorithms in policing.
WMP built different machine learning models that aimed to identify stalking and harassment suspects who may become high-harm offenders. Data was shared with university partners so the features and performance of each modelling approach, including the methods WMP had taken could be evaluated. This included replicating each model to assess its accuracy in identifying high-harm suspects.
The university partners also conducted an extensive literature review and interviews with members of WMP were performed to identify the rationale for developing the model, as well as the steps they had taken in building it, and plans for how it could be implemented within current practice. This formed the basis of the framework for developing prioritisation algorithms in policing.
A replication of the model produced by WMP showed that a high-recall algorithm identified 76% of high-harm suspects but included 74% lower-risk individuals. Combining the algorithm with human judgment in a triage clinic proved feasible for reducing harm.
The results from all three methodologies were synthesised into a report for publication, a technical report for WMP, and a framework that is publicly and freely available -
https://www.police-ml.com/.
The RUDI (Rationale, Unification, Development, Implementation) framework provides ethical guidance for algorithm implementation, supporting police in building a business case, integrating data sources, testing models, and managing implementation challenges.
Devon and Cornwall Police
Mitigating the risk of Repeat Sex Offending (RSO) suspects is imperative, with 60% of Rape and Serious Sexual Offence (RASSO) suspects already appearing on police systems. This project developed and evaluated a series of behavioural based interventions in Devon and Cornwall Police and produced national police guidance on how these interventions can work to better target better target RSO suspects.
The project took on three strands of work:
Further development of the Suspect Check Template, originally created in Operation Soteria, for officers to record the challenges and benefits of interventions, and their perceived efficacy of considering wider disruption options.
Creation of a Disruption Panel to identify RSO suspects, assess their suitability for further policing work and create intervention plans for incorporating a range of targeting and engagement activities.
An assessment of the national use of Behavioural Analysis to identify RSO’s through a series of interviews with 14 forces.
Initial evaluation of the Suspect Checks Template suggested that some officers could see the value in this tool, although the implementation of such a tool must be carefully considered to ensure buy-in and continued use from staff. Early indications suggest that additional disruption options, specifically the use of civil orders, increased.
The Disruption Panel provided a structure in which DCP could consider how to identify, assess, and disrupt RSO suspects. Recommendations surround inclusion of teams such as Integrated Offender Management (IOM) and Neighbourhood Policing Teams (NPT) and a six-monthly review process for RSO Perpetrator Plans. Formal evaluation will be conducted after multiple Disruption Panels have run independent of academic oversight.
The national review of the use of Behavioural Analysis in the investigation of repeat sex offences identified a lack of knowledge around the help available and extremely high workloads. However, the utility and future potential of forces engaging with academics and practitioners within the behavioural science domain is great.
Thames Valley Police, The Open University
The study aimed to examine the use of Evidence Led Prosecutions (ELPs) in domestic abuse (DA) cases by using both quantitative and qualitative data to further understand when, how, and with whom these prosecutions are utilised.
82 police case files were analysed from three police forces participating in the project. The study concentrated on common assault, actual bodily harm, and non-fatal strangulation offenses. It explored initial attendance statements, victim engagement, and reflections on ELPs from both victims and officers.
Victims were predominantly white (86%) females (87%) with a history of violence from the current perpetrator (82%). Most perpetrators were white (85%) males (85%), with 92% having criminal records and 73% had prior domestic violence offences. Incidents often occurred in public places (36%), shared homes (30%), or the victim’s home (25%). The risk level was high in 50% of cases, medium in 44%, and standard in 6%.
Victims’ emotional states were missing in 53% of files, but statements with emotions had higher charge rates (63%). Victim engagement decreased when third parties reported incidents. The study uncovered several useful insights into the use of ELPs in DA cases. Victims expressed positive sentiments about achieving justice without involvement, while negative sentiments included feelings of being taken advantage of and not having their choices respected. Victims suggested spending more time explaining the ELP process, increasing specialist DA officers, and enhancing police knowledge on ELP evidence requirements. Officers reflected on feeling like “baddies/bullies,” experiencing compassion fatigue, and vicarious trauma affecting their service to DA victims.
NPCC CCTV Portfolio, Digital Video Experts Group
An increasing number of smart security cameras have been submitted for analysis to Law Enforcement Agencies (LEAs) . The project aimed to research the data storage and transmission structure of Wi-Fi and Cellular VSS/CCTV cameras to support Forensic Labs in their extraction and analysis of these devices. This would provide a forensic capability to reduce examination times and missed opportunities to capture evidence at crime scenes.
The project team conducted technical digital forensic research to identify key issues relating to Smart Security Cameras, including methods of extraction of data from Wi-Fi and cellular enabled CCTV cameras, de-coding of encrypted data, vulnerability testing and interception of Wi-Fi data, exploitable connectivity options, access to cloud data sources and detection of concealed devices at the scene.
This research was used to support LEAs in exploiting the evidential prospect of these devices in line with forensic best practice and regulations
Published outputs for the project included data sheets for 35 of the most common wireless CCTV devices and dash cams, and a generic published data retrieval process for forensic teams. The second outcome was the creation a community of experts from across policing and government departments who specialize in video evidence and its underlying technology. This community includes police video units, digital forensic teams, intercept analysts and Technical Support Units, these experts discuss present and future issues regarding Wireless CCTV and provide operational advice on data retrieval.
West Yorkshire Police, University of Staffordshire
There is little literature documenting the analysis and comparison of footwear captured in CCTV, Body Worn Video (BWV) or mobile downloads. This limits the resources available that could help forensic practitioners determine the make and model of footwear recorded in footage or understand its evidential value. Therefore, the first aim of this project was to test CCTV systems to determine whether there is consistency in the images they produce. The second was to develop a cost-effective and efficient method that provides a 360-degree view of footwear in white and near infrared light that can be uploaded onto a searchable database.
Pre-existing hardware (cameras, turntables, near infrared light sources) and software (imaging capture including 3D modelling applications) were investigated to explore the most cost effective, efficient and most fit for purpose approach. 2940 images of footwear uppers were collected and analysed from 5 different cameras, this included 49 pairs of shoes from a range of brands such as Adidas, Nike and Puma.
The results were used to produce two separate standard operating procedures to create 360-degree images and 2D still images. The footwear sample was used to develop the search criteria for a footwear uppers database and to test its discriminatory potential. Future research will investigate the evidential value of footwear uppers as a form of identification from images, allowing practitioners to better understand both the use and limitations of this evidence type.
Forensic Capability Network, University of Warwick, University of Leeds
This project aimed to address challenges in investigating cases of Violence Against Women and Girls (VAWG) involving digital text evidence from mobile phone conversations. Current processes, reliant on manual review of communication data, are slow, potentially overly intrusive, and ill-suited to the fast-changing nature of digital communication. These challenges risk compounding harm to victims and undermining the justice system.
A Natural Language Processing based platform was developed to detect threatening and abusive language in communication with victims. Key innovations include an integrated analysis platform that combines a browser-based interface with NPL models tailored for the analysis of sexist and violent content, along with a chatbot powered by Large Language Models (LLMs). Other innovations included conditioned summarisation techniques that focus on violence-related messages, LLMs for synthetic data generation, and persona profiling to understand the behaviours of involved individuals. The project faced challenges relating to data privacy, technological integration, and user interface design, which were addressed through synthetic data sets and iterative design improvements based on user and subject matter expert feedback.
The results indicated that the NLP technologies facilitated faster reviews of communication data, with the potential to improve accuracy and standardising investigative procedures. The next phase will focus on refining AI models to boost predictive accuracy, broadening data analysis to improve persona profiling - including image content - and enhancing the user interface for more dynamic interaction.
Cleveland Police, Teesside University
Project Guardian aimed to provide an understanding of the scale, nature, and dynamics of police pursuits in the Cleveland Police force area, and review the current procedures and tactics used to bring those pursuits to a conclusion.
The project analysed 727 pursuit cases recorded between November 2020 and November 2023, using Cleveland Police data, focus groups, and interviews with Roads Policing Unit (RPU) officers, Force Incident Managers (FIM), and Tactical Advisors (TACAD).
Motorcycles were identified as a significant and growing challenge in relation to police pursuits with current tactics being largely inadequate, and pursuit numbers increasing. Geographical challenges such as urban settings limited the efficacy of current APP tactics, and it was recommended a comprehensive review of the existing tactics should be conducted and new approaches developed. Risk aversion at all levels resulted in fewer serious injuries and limited vehicle damage, but the negative was that high levels of suspects were lost as a result. Training for drivers was deemed good, but for Force Incident Managers (FIM) inadequate, with both agreeing that a better mutual understanding of each other’s roles was needed to improve risk management and decision-making during pursuits. Pursuits in the force area were often connected with criminal activity and showed a downwards trend in suspect ages with suspects’ median age being 27.5 in 2021, whereas by 2023 the median age dropped to 23.
Kent Police, Canterbury Christ Church University
The aim of the project was to trial and test an evidence-based template to objectively measure the value of a Covert Human Intelligence Source (CHIS). The measurement would supplement professional judgement and provide meaningful data for practitioners undertaking reviews of CHIS value.
Phase one used semi-structured interviews seeking practitioners’ views to improve a devised template from a previous exploratory study. In Phase two, the template was completed for 120 registered CHIS over a 2-year period within Kent Police (2021 to 2023). Phase three asked relevant Kent Police practitioners to rate CHIS independently. Finally, CHIS were rated by a panel of experts. The data was analysed using descriptive and inferential statistics. Phase four involved the development of a cost/benefit ratio score which was used for each CHIS to provide details of CHIS for decision-makers, and the development of a red/amber/green system.
Phase one results showed there was support from practitioners for an objective measure to assess the value of a CHIS, with the caveat that professional judgement was still important as a purely data driven approach could miss the nuances of a situation.
Phase two related to the completion of the template for all 120 CHIS. Results were analysed and related to information such as: number of intelligence reports emanating from the whole cohort, number of contacts, time taken to produce information relevant to a priority crime, total value of seized property, drugs etc.
Handlers tended to rate CHIS higher than supervisors and in turn supervisors tended to rate higher than controllers. There were inconsistencies among the different raters at all levels. However, the template results correlated at a statistically significant level with all the different raters, meaning it can be shown to reflect the general direction of CHIS ratings. In phase four, following creation of a cost/benefits ratio, CHIS were then rated against a RAG system where the red/amber/green labels were attached to each CHIS. Decision makers could then review the amber and red cases and apply their professional judgment.
