As Digital Transformation Manager for Housing Services at Hackney Council, I get to run some really fun and interesting projects. This week saw the start of a great one however, as we kicked off a data science project which we hope will enable us to identify tenants most at risk of falling into rent arrears (it also marks the start of what will no doubt become a remarkably infrequent blog, but that’s another story).
Arrears are bad news for everybody; they cause stress for tenants who may lose their home and have to move out of the borough away from their support networks, and they reduce the amount of money we have to maintain homes and estates. When people fall into arrears we set up repayment plans, but for people on limited incomes the size of the repayment we can realistically expect them to be able to pay consistently on top of their rent is quite low, making debt stubbornly difficult to clear.
But what if there’s a better way, and we can predict those most at risk of falling into arrears so that interventions can be targeted to prevent the problem before it occurs? Within the council we have a Financial Inclusion team that helps residents with financial planning and can point them towards training to help them get better paying jobs. If we could better target that team’s resources to those most at risk then this wouldn’t just help reduce rent arrears, but give our tenants more control of their finances and help tackle unsecured debts, or payday loans.
To try and answer this question we’re working with a company called Pivigo, that run a programme to train candidates with PhD’s in quantitative disciplines to be data scientists. As part of the programme they need organisations to present them with real world problems to solve, so this week I met with our team to discuss the challenge. The team comprises:
Over the next month they will be analysing our data using techniques such as clustering, decision trees, and time-series modelling and building a model to try and quantify the risk of a resident falling into arrears. This model should then enable us to play with certain parameters such as anticipated inflation or wage growth rates to see how this may impact on our residents in given scenarios. Microsoft have given us free access to their Azure Machine Learning platform for the duration of the project, but the algorithm we develop will be platform-agnostic and available on Hackney Council’s GitHub repository. We hope that other local authorities and housing associations will test their data against it also and that we can work together to build upon and refine it.
In a project dealing with such personal data as this privacy is of course incredibly important. Personally identifiable data is not needed to develop the model and so not included in any of the data sets used for analysis or testing. Similarly, should the project be successful and we create something that can be introduced to our working environment it wouldn’t be something that staff would be able to dig around in, but a tool that selectively highlights only those that may be considered vulnerable to the teams that can help them.
The project is scheduled to run until 7th September and I’ll post again on what we’ve learned, but the end point is really just the MVP and I’m hoping that we can work with other local authorities and housing associations to develop this further.