Global progressive policing - To register a FREE account Click here
INNOVATION:

Answering the call: Lancashire’s voice to text analytics project

Voice to text

A review of Lancashire Constabulary's Force Control Room, which receives more than 1.2 million calls a year, led to the development of an award-winning voice to text system which will significantly improve the force's call handing and service delivery to the public. Project leaders Chief Supt Ian Dawson and Insp Andy Doran explain the background to the project, the actions already completed, and the potential for further advances.

In 2016 Lancashire Constabulary commenced a detailed ‘systems thinking’ review of its centralised Force Control Room (FCR).

In line with Lancashire Police and Crime Commissioner’s (PCC’s) Police and Crime Plan and the force’s priority of the delivery of core service, the work was developed with significant support from the PCC. It was delivered in partnership by the Constabulary’s FCR and Futures teams, with the programme led by the chief officer team.

The review considered demand by type and frequency, leadership, our processes and structures, and how the work flowed into the different pathways available for wider service delivery.

The aim was to improve the FCR’s performance by enriching our understanding of demand and capacity, while also improving our capability and service quality to the public.

The review considered demand by type and frequency, leadership, our processes and structures, and how the work flowed into the different pathways available for wider service delivery. This piece of work is now in the delivery phase, with the FCR currently decanted into three temporary rooms whilst the central FCR is re-designed.

Underpinning this change is the significant ICT investment, leadership and staff development, geographic pod alignment, process improvement and insight, through a comprehensive set of measures.

Understanding the problem

Lancashire Constabulary receives 1.2 million calls per year into the FCR. Typically, these are split between 900,000 101 calls and 300,000 999 calls. Out of the 900,000 101 calls, the Constabulary creates 300,000 logs with the remaining 600,000 calls sitting as unlogged audio files.

During the systems thinking review, a team of eight spent one and a half months listening and categorising 1,200 ‘unlogged’ calls, to understand demand in detail. This insight was essential in helping to shape the new operating model for the FCR; however, it was suboptimal and unsustainable.

This method uses an expensive resource and takes a long time to gain insight, which is immediately historic and offers no real-time view of demand. Ultimately, this limits our understanding of the internal or external factors influencing demand.

We found that channel shift in its current design simply creates more avenues for failure demand to enter the system, without addressing the root cause. Consequently, this reduces the capacity within the digital channels to deal with value demand.

This initial work, however, did provide insight and evidence for some of the large call categories. As an example, the statistical sample estimated that the force receives 209,000 calls each year from members of the public asking for an update to their crime or incident. In real terms, this amounts to one in five calls of the overall demand, 572 calls a day, one call every two minutes and 50 seconds, and 57 staff employed to answer them, at a cost of approximately £2,000,000.

While the Constabulary had already invested in significant digital options for public contact, it was clear that this was not creating ‘channel shift’. Although a much talked about premise in policing, we found that channel shift in its current design simply creates more avenues for failure demand to enter the system, without addressing the root cause. Consequently, this reduces the capacity within the digital channels to deal with value demand.

Currently, forces across the country utilise switchboards or call back facilities to screen out this type of demand, but voice to text technology provides the insight needed to deal with this demand effectively and prevent it coming into the FCR.

Finding a solution

In 2018, Lancashire Constabulary carried out extensive private and public sector research with the aim of identifying a solution to this problem. In response, the force embarked on a collaborative project with Hewlett Packard Enterprise (HPE) and Intelligent Voice (IV) to develop a real-time understanding of all unlogged demand through the development of a voice to text translation solution.

The project also aimed to utilise the wider datasets to help understand when these calls were received, the levels of vulnerability within the calls, and the impact on the wellbeing of our call takers.

Phase 1 focused on testing the effectiveness of voice to text technology in a policing environment, and demonstrated the significant potential that the technology has within policing.

With the support of HPE, we built an on-premises test environment, which leveraged high-performance hardware and software to process a sample set of 60,000 calls. The insight from these calls helped to create conceptual and logical data models, which eventually resulted in physical data models that helped us make the invisible, visible.

Designed in three phases, Phase 1 focused on testing the effectiveness of voice to text technology in a policing environment. This phase concluded in July 2019 and demonstrated the significant potential that the technology has within policing. In this phase, we built the categories based on word models, which went on to provide new insight into demand from the voice to text translation.

Phase 1 provided detailed categorisation of 21 categories of unlogged demand that could be broken down by time of day and day of week. They were also checked for accuracy through an assessment of a sample of calls by the team, reaching 98% accuracy for custody calls and 81% accuracy for case-related calls.

In total, this insight showed that we were spending 40% of our call-taking capacity on unlogged demand. An example of the insight on case-related calls is shown below.

The time spent on case-related calls

Voice to text Graph 2 V2

The daily breakdown of case-related calls

 

Phase 1 proved the concept of voice to text and how automated insight could revolutionise our understanding of demand. As a result of these findings, the Constabulary made immediate changes to its FCR delivery including improvements to the interactive voice response (IVR) messaging and electronic calling cards for front line staff.

Phase 2 sought to build on the learning from Phase 1, and concluded in the last quarter of 2019. The focus in this phase was to assess how effective the technology was at improving our understanding of mental health and vulnerability demand while also assessing the viability of an integrated live-time solution.

The voice to text data correlated with the academic assessment that low-risk cases present to the FCR in the afternoon, while high-risk cases present late at night. This also provided insight into our staff capability, and it was clear that they were logging the right incidents.

The Constabulary had recently undertaken a detailed assessment of its mental health demand with the University of Central Lancashire, and used the evidence from this learning to build a word model to search unlogged (and previously hidden) mental health demand. This identified that 3% of our unlogged demand related to mental health.

Furthermore, the voice to text data correlated with the academic assessment that low-risk cases present to the FCR in the afternoon, while high-risk cases present late at night. This also provided insight into our staff capability, and it was clear that they were logging the right incidents.

Not surprisingly, significant levels of calls which were not logged presented on Friday afternoons, the hypothesis being that the police were the point of contact when there was no wider service provision for mental health services. The below screenshot demonstrates this insight:

Voice to text Graph 3 V2

Mental health calls peaked on Friday afternoons

 

The achievements to this point were presented at the ADS Security and Policing event in March 2020, and the collaboration won the National Innovation Award.

The operational system

Phase 3 started in early 2020 and focused on moving from a closed-loop system to a sustainable integrated model within the force’s ICT architecture. This system takes live data and converts every call (101, 999 and primary channels) into text, delivering live-time automated transcription and categorisation capability.

With the COVID-19 travel restrictions, the delivery of Phase 3 was delayed. However, through the collaboration between HPE, IV and Lancashire Constabulary’s ICT, we were able to continue to build the system remotely – an outstanding piece of collaborative working.

Over the last seven weeks the system categorised upwards of 70,000 calls. Data analysts have connected the ICT architecture to the Force Business Intelligence solution and developed a dashboard to display a live-time product of all calls presented to the contact centre. The high-level entry dashboard is demonstrated below:

Voice to text Graph 5 V2

Lancashire Constabulary’s new live-call dashboard

 

An exceptional partnership

This architecture is now the platform for the delivery of service solutions providing automated insight on every call. It will enable tracking of changes to demand presenting to the FCR, through external factors and through internal systems and process changes.

It will also enable next steps in digital responses to physical contact improving the victim’s journey. Our next steps will deploy emotion recognition, behavioural analysis and voice intent technology to assist in identifying hoax calls and in wellbeing interventions for the workforce. The technology can be applied outside this environment in all voice data content, which will help us to grow the capability of this technology.

This has been an exceptional partnership between the public and private sector in developing a solution to a complex problem to improve the policing service to the public. The work has been shared by Lancashire Constabulary with other forces nationally through the National Contact Management Steering Group for adoption of this technology nationally, and with the Home Office for consideration in developing the solutions for the re-contracting of 101.

About the authors

Voice to Text - Ian Dawson H&SChief Superintendent Ian Dawson has been the Head of Corporate Development and Futures at Lancashire Constabulary for four years. He leads the Constabulary’s Change Programme, including its approach to digital transformation, and the adoption and use of new technologies, automation opportunities, AI and associated data management. He has been the evidence-based policing lead for 10 years and is passionate about it providing the foundation to policy and decision making. He achieved an M.St. (Cantab) in Police Leadership in 2016.

 

Voice to text - Andy Doran H&SInspector Andy Doran joined Greater Manchester Police in 2006 having studied Criminology (BA Hons) and Applied Research and Consultancy (MA) at Lancaster University. He worked in a variety of roles (including response, neighbourhood policing, serious acquisitive crime investigation and CID), before transferring to the Lancashire Constabularyy CID in 2014, and subsequently achieving promotion to sergeant and then inspector. A passionate collaborative leader, Andy currently heads up a team of change practitioners within the Constabulary’s Futures Team, and is leading the voice to text collaboration with HPE and Intelligent Voice.

 


You must be registered and logged in to post a comment

Please LOG IN or REGISTER
Top