Developing contextual responses to the abuse and exploitation of young people


Hindsight to insight: Deploying predictive analytics to safeguard



Think Family is Bristol’s answer to the government’s national Troubled Families programme. It is designed to improve outcome for families whilst reducing cost and demand on the public sector. This is being achieved through a truly effective early intervention strategy, which delivers significant and sustained outcomes for thousands of families across the city of Bristol. 

One of the programmes key assets is the Think Family Database. This data warehouse draws together information from 30 data sources, about 54,000 families in the local authority area.  These datasets are built on the 6 headline criteria for the Troubled Families programme, Crime & ASB, School Attendance, Children Needing Help, Unemployment, Domestic Abuse and Health Problems.  The sources are matched together to give us the best understanding of the issues facing families in the city that we have ever had. 

In order to do this we needed to put in place extensive data sharing protocols.  This included notices for all of the city’s residents, and publishing agreements, privacy notes and impact assessments on our website.  The information sharing protocols and legal gateways for the project can be found here.

The data captured within the TFD supports a more strategic understanding of the city and the issues and risks experienced by our citizens. It also informs service planning and the more efficient allocation of resources. The TFD enables more effective information sharing, whilst also providing the data to drive improved analysis and predictive analytics.

Predictive modelling allows us to make better use of data, to understand the known issues affecting the citizens of Bristol now and in turn how these factors may develop in the future. Understanding future trends can inform resource allocation and decision making on both strategic and operational levels.  For example who is most likely to become NEET or which children are at risk of being placed in care?  The answer to these questions is held within our data and modelling techniques allow us to find it. 

Predictive analytics should be interpreted intelligently, the results of a model do not replace a lead professional’s assessment nor are the outcomes generated guaranteed. Instead the results are meant to be used as a tool to get ahead of the curve, this use of data supports an early intervention approach. 

In the case of the Bristol CSE model, partners from the Barnardos project BASE (Barnardo's Against Sexual Exploitation) were consulted while gathering evidence and information on known cases of CSE in Bristol. By defining the common factors/ indicators in the lives of known CSE victims, the predictive model was able to search the TFD for individuals who were statistically similar to these known cases.  The model includes variables about a person, such as frequency of being reported missing to the Police or school attendance rates, but also includes wider family context such as having a sibling in care or associations with other victims of CSE.  This combination of personal and contextual data gives a powerful basis for scoring.

Making the CSE search tool accessible for lead professionals and staff is vital. Initially the CSE model was available as a separate spreadsheet. Feedback showed that having to access multiple systems made staff less likely to use the model. As a result we moved the data from Excel to SSRS and developed an embedded front end for staff which is accessed via our main system. To ensure new and existing users are kept up to date with how to use the search tool and interpret the results training and awareness sessions are delivered to teams across the city.

Using the model to identify children and young people with the most heightened risk scores, we have allocated 243 cases to key workers over the last year. As a result these families have received targeted support, aiming to reduce the vulnerabilities affecting them. 

We believe the proactive targeting of support to those who most need it is the future of family support and enables us to intervene at the earliest signs rather than wait for referrals at the point of crisis.  The CSE model has given us some valuable learning in how to deploy this sort of tool and the workforce and culture change required to make it a success.  It has laid the path for future work.  We have since deployed several other models and are currently working on identifying opportunities to reduce the numbers of children going into care and improve targeting of parenting programmes within Early Help services.

For more information or to get in touch with the author please email  

Posted: 13 Sep 2017

Author: Delphine Peace

University of Bedfordshire logo