Maintenance activities play an increasingly central role in the trade-off between the conflicting objectives of social service, efficient productivity and safe operation which drive the public sector. Knowing when and how to test, inspect, repair, renovate, and revert the components of a system is fundamental to reduce failures and unplanned downtime for safety or economic reasons, and ultimately lead to optimal safety and economy. This modeling approach provides the first step towards making better use of the available data that was collected at considerable expense.

The blockbuster software package was developed to reliably predict how the condition of the School Estate will change through time. The model takes as input; the reasonably big PDS data (too large for spreadsheets), which details the condition and quantity of each building component in every school building, the deterioration rates, and repair costs of those components and the amount of funding spent on the rebuilding and repair of those school buildings. Together these inputs are used to simulate the deterioration of the School Estate into the future under various maintenance and rebuilding spending policies.

Predictive modeling in Government Departments has historically been constrained by using data sets that are small enough to load into Microsoft Excel. The Property Data Survey (PDS) was too large (2.3 million rows) and thus was aggregated into just six numbers. These statistics were then used to support a failed Autumn Budget request for extra funding to rebuild and repair schools in England (the model was criticized). This compromise was a relic of skills and technical shortage, despite the data being available to answer the question with the desired precision.

This innovation is currently in the process of being used as part of standard business practice by the Department for Education (DfE), the output of which will form part of their budget request from the Treasury. This product has been developed through collaboration between data scientists and analysts, and now ultimately between the Government Digital Service (GDS) and DfE. The Strategy and Intelligence team with the EFA recognized a skill gap existed and recruited a data scientist to tackle this problem, enabling the data to be used in a more optimal way. Through a collaboration with GDS they were able to tackle a previously inaccessible data problem. The model was developed using Open Source software (including version control) as a package. The package enshrines all the business knowledge used to create a corpus of work in one place; including the code and its relevant documentation. This protects the department against staff turnover and improves the reproducibility of analysis.

Working with partners within the business, across both analytical and policy communities, it was possible to deliver this project, we gathered some feedback about this collaboration: “The model itself is a step change in improvement over the older model, particularly the level of granularity in which the blockbuster model operates at”.

The data science process involves a lot of back-and-forths—between the data scientist and stakeholders, and between the different stages of the process. Our key takeaways were that: A successful data science project involves more than just statistics. It also requires a variety of roles to represent business and client interests, as well as operational concerns. Make sure we have a clear, verifiable, quantifiable goal. Make sure we’ve set realistic expectations for all stakeholders.

DfE is now running the simulations on blockbuster to calculate the backlog cost of repairing the school estate with the latest funding assumptions up until 2020-21. They are now working with a policy team to provide outputs from the model to inform business cases to Treasury for the Autumn Statement. A key requirement is to help shape the value for money argument for investing in the school estate now rather than wait for the estate to deteriorate further which will then incur a significantly larger cost (which is not limited to pure construction inflation).

The granularity of the blockbuster model will be able to strengthen the value for money case with Treasury. The challenge for us now is to produce reliable forecasts of the cost of repairing the school estate for a variety of different scenarios and policy assumptions, with the aim of setting out a clear argument to HMT that we need to significantly invest in the school estate now and not in 5 or even 10 years’ time . This objective is clearly for Peter and me to take forward but your support and guidance on using the blockbuster model to the best of abilities will no doubt help our cause.

Getting sponsor sign-off becomes the central organizing goal of a data science project (to approve or acknowledge the outcome as meeting the goals of the project). At the outset of the project the understanding was that the blockbuster model was to be used to improve on the old spreadsheet model by simulating the expected deterioration of the school estate under simplified rebuilding and maintenance policies (a fixed policy), however based on the previous meeting the data scientist got the impression that the sponsor was more interested in the effect of different policies or identifying the optimal spending policy for school rebuilding and maintenance. Although a similar methodology may be appropriate it is a very different goal and should be clarified and distilled further.

It’s important that all work is being undertaken within an environment where changes can be made and expert opinion listened to. If stakeholders are unwilling to be challenged, in this case with reference to the old model, then it is very hard to achieve improved outcomes from a project. It is also vitally important that the technical skills required to further develop the project and iterate upon it are present within the department, or that the department is willing to upskill existing staff or bring in new skills to enable this.

This particular innovation represents a step-change in the way that government handles the costings and renovation of its estate. In this example, the data has been collected and a system has been deployed for the purpose of the school estate, however, this same approach could be taken and applied to any property-owning part of the public sector such as Defence, Prisons, Social Housing, and Hospitals.

In order for a collaboration like this to be successful, there are a number of things to consider. The parameters of success need to be closely defined; it is important to ensure that the sponsor, working level collaborators, and domain level experts, are all bought into the same vision for the project, this means ensuring that clear goals are set. This can be obtained through approaches such as directed interviews.