Demand Model accurately predicts care spend for Middlesbrough Council
Background
Demand Model is part of an intelligence tool kit that uses current and historic data to predict future trends in adult social care and support Local Authorities in more accurate forward planning, particularly in the sphere of social care placement spend.
One of the first clients to benefit from Demand Model is Middlesbrough Council, who have been using a demand model to inform Commissioning decisions, budget planning and workforce structures since 2012/13.
Middlesbrough is part of a cohort of Local Authorities in Tees Valley who collaborate to collect, share and analyse care transaction data, utilising our Data Hub. Therefore, when the Local Authority decided to embark on a project to develop a more flexible, automated and granular demand model, our solution offered the perfect fit.
Carl Johnson, Information Analyst at Middlesbrough MBC explains, “We use demand modelling to quantify our user base and provide a view of activity levels across a range of service areas. We also wanted to be able to predict what is likely to happen in the future, to aid us with Commissioning decisions and financial planning. In our earliest model, the data showed us movement between one financial year and the next only; and while the more recent version could account for quarterly movement, our ability to visualise changes in demand was limited - there was none of the granular detail that affects demand. What’s more, it took us two to three weeks every year to refresh the dozen or so categories and we would produce quarterly updates. So, we were looking for a better way forward.”
A better way forward
Middlesbrough’s brief was to develop a more automated system, able to provide regular updates as and when required, to be responsive to unexpected changes in activity so that the team could see change coming and react quickly, and finally, provide more granular forecasts, by age group and by primary support reason.
According to Tom Knight, Head of Data Insight at HAS Technology (an Access company), it was decided early on that rather than simply improve the existing model, a radical solution was required. “We went back to the ‘drawing board’”, he says.
Research brought Tom and the team to Time Series Forecasting, which in itself is nothing new. Using statistics, Time Series Forecasting allows you to look at historical trends and forecast what might happen in the future, but there are multiple models to choose from, with many different settings.
“In our proof of concept testing, we couldn’t find a model that fitted every nuance of social care,” Tom says. “So, it was clear that we needed to build something that would be highly adaptable.”
Intelligently Tuned Predictions
“Time Series Forecasting models often test the accuracy of their forecasts against a period of known activity. We realised we could write an algorithm to run as many models and configurations as possible and work out which was the best combination for the particular series of activities – some stable, some more volatile. From there, we could develop a tool that delivered a high level of accuracy.”
What would take an analyst at least a month, we could achieve in just a few seconds. Currently, Demand Model is running 630 separate analyses at any one time for Middlesbrough.
“Essentially, we’re using modern technology – namely robotic automation – to really improve performance in one area to get the right outcome,” Tom confirms.
Outputs - future predictions accurate to within 1%
Looking at long term residential care activity over a full year enabled Demand Model to make predictions accurately, with up to 99% accuracy for Middlesbrough MBC. Based on historic data, the Demand Model predicted in April 2018 a requirement for 726 social care beds. In fact, the figure in April 2019 was an astonishingly close 734. The forecasts can allow for and factor in seasonality too, identifying periods of increased activity, followed by decreased activity, for example in terms of social care requirement this may be particularly marked during winter.
Handling unexpected events
Social and health care can be unpredictable, with many factors, such as weather, triggering change in activity. So, a successful system cannot simply rely on past activity to predict future requirements, it must be able to react to unexpected and sudden changes in demand as well. Accordingly, Demand Model doesn’t just run a single predictive model, but four at a time, providing predictions that are right up to date, not only at year end, and allowing the team at Middlesbrough to spot when variance increases or decreases by more than 5%.
“5% change is the alert that demand is deviating from the norm, triggering the team to take action,” Tom explains. “If change isn’t recognised and addressed quickly, the consequential cost of bed shortfall for example, could run into millions of pounds.”
“Our Demand Model can’t solve the problems,” he observes, “But it gives the team at Middlesbrough the ability to identify a problem quickly as it arises in real time and it also provides predictive intelligence to support their reaction. It can make the difference between simply watching your market, or managing it.”
Unit costs are in the mix too, so that from actual figures for activity and spend, predicted activity can also be viewed in terms of budget required now and going forward.
Delivered benefits
- Demand Model takes less than an hour to refresh, and supports accurate monthly predictions.
- It delivers exceptional granularity, monitoring and tracking 632 areas of activity at any one time, with the ability to drill down into primary support reason and age groupings.
- Exception reporting means the team don’t need to track all 632 areas, but can focus resources instead on those that have been flagged as stepping out of the predicted line.
Amazingly, this complex and comprehensive predictive tool, accurate to 1%, is built on just six fields of required data, all of which are available on Local Authority Statutory Returns. These are week start date, service type/demand category, age group, primary support reason, unit of activity and average unit cost.
Customer feedback
“To be able to monitor this kind of activity and project it with the level of accuracy delivered by Demand Model is great for us,” concludes Carl Johnson. “In these times of austerity and increased pressure on health and social care, we really need this information. The system is efficient, processes data quickly and doesn’t take a great deal of resource to manage. As a result, we can see what is happening on a real time basis, which is very important. It gives us lots of intelligence about our local care market, which we’re required to have under the Care Act, but it also helps Commissioners better understand and plan the services they should deliver. Other benefits include the ability for us to make evidence-based decisions from consistent, accurate forecasts. The system provides more reliable data for setting our budgets, it allows us to react quickly to unexpected changes, improve the quality of our service to customers and derive efficiencies from automation and smarter commissioning. We can target our attention to where it is needed, rather than where we think it should be!”
Tom Knight adds, “The Demand Model demonstrates how data already being collected can be exploited for strategic planning, based on a reliable view of the road ahead rather than the road already travelled.”
Maximising the Demand Model
Having initially developed this predictive tool for Middlesbrough and the other Tees Valley sites, Demand Model has taken all the learnings from that work and created a more generalised product that can be implemented by any Local Authority. To date, in addition to the four Tees Valley Local Authorities, others already realising the benefits of Demand Model to include five LAs in the Liverpool City Region, and most recently, Birmingham.
For more information about Demand Model download our 'Guide to utilising health and social care data to predict demand'
