Knowledge Transfer Partnership with Liverpool University

From left to right: Dr David Hamilton, Philip Moss and Professor Leszek Gasieniec

By Dr David Hamilton

The social housing sector is data rich. Huge volumes of information are gathered every time a provider buys products, undertakes repairs, lets properties or asks about customer satisfaction. As a result, landlords hold a wealth of statistics in their housing, asset management and finance systems.

But the sector doesn’t do enough with this ‘big data’. Figures are often siloed, and many providers still rely on traditional data-processing methods such as spreadsheets to collect and manage large volumes of data. More needs to be done to clean, join up and analyse this information and draw meaningful insights from it.

For the past two years, I’ve been working with PfH to tackle this problem. Every month I’ve been examining 300,000 lines of spend data, from nearly 1,000 landlords, looking at how data analysis could improve procurement in the sector.

I’m a researcher at Liverpool University’s School of Electrical Engineering, Electronics and Computer Science and since 2016 I’ve been involved in a Knowledge Transfer Partnership with PfH. Machine learning and predictive analysis is my academic area and I’ve taken on the role of data scientist at PfH, working with the company’s growing IT team.

What is a Knowledge Transfer Partnership?

KTPs help businesses to innovate by linking them with research organisations like Liverpool University. Part-funded through government grants, these partnerships enable companies to bring in the latest skills and academic thinking to deliver a specific, strategic innovation project.

PfH is the largest spend aggregator in the UK social housing sector, so for them, that innovation project is showing landlords the potential of their data and helping them to manage and use it in the right way.

And that potential is huge. PfH manages over a million invoices a year – that’s £250m of spend. Over the next ten years it will process around 50 million lines data. PfH wants to use this unique position to provide sector insight to landlords and machine learning is key to achieving this.

What is machine learning?

Put simply, it’s about identifying patterns in historical data. Algorithms learn these patterns and then forecast future trends.

Machine learning is a type of artificial intelligence. AI is all about creating computer programmes to think and learn like humans. Machine learning is one of those computer programmes.

During this knowledge transfer partnership, I’ve designed machine learning models to help PfH analyse data. These models will power a range of new analytics-based solutions for members.

How is the KTP helping members?

The partnership has helped PfH to automate the categorisation of purchases made by PfH members, drilling right down into their spend data. Being able to classify information at such a granular level – for example telling a housing association what they spend on chrome plated bathroom taps, right across their organisation – can help landlords identify whether they are spending too much on taps compared to their peers, whether they are purchasing several types of tap unnecessarily or whether there are better value taps available.

Such low-level information can also be combined with public data sets to provide a more holistic view around packages of work. For example, is a housing association paying more than the average cost of replacing bathrooms in 1,000 three-bed social housing properties?

Machine learning has enabled us to categorise both on-catalogue and off-catalogue spend. So, if a member wants to know what proportion of their materials spend is non-compliant, PfH can tell them immediately.

Having provided these insights, PfH’s aim is to work collaboratively with members to explore the opportunities available to them, help them to save even more, boost quality further and increase overall value for money.

Wider use of machine learning

Working with PfH has given me a real insight into the social housing sector. The potential for machine learning to support providers is significant. Here are just a few examples:

  • Predictive analysis could link housing providers’ repairs data to price indices to indicate the best time to buy certain products such as building materials.
  • Machine learning could be used to recommend comparison products for social landlords, for instance a boiler that is less expensive, has a longer warranty, has a smaller carbon footprint.
  • Data from technologies like IoT thermostats, windows sensors or smart boiler parts could recognise failure in advance and help organisations switch from reactive repairs to planned maintenance.
  • Landlords could use ‘emotion AI’ to analyse social media mentions about suppliers and combine this with data on contract performance, legal disputes or redundancies to build risk profiles.
  • Machine learning could help with letting decisions, using demographic and other data sets to calculate which customers are likely to leave early or create the greatest repairs demand.
  • Providers could use predictive analysis to identify rent accounts likely to default, allowing them to support tenants before they run into difficulties and avoiding eviction and resettlement costs.

This month marks the end of PfH’s two-year knowledge transfer partnership with the University of Liverpool, but it’s just the beginning for PfH’s data strategy. The relationship has enabled PfH and its developers to create a powerful data warehouse which underpins new services such as price-checking and VfM reporting and adds additional insight to existing services such as spend analysis.

For more information about PfH’s partnership with Liverpool University, please contact