The Demand Model combines statistical forecasting techniques with artificial intelligence to produce a highly versatile management information tool. In stable environments, it can be used to make highly accurate predictions about future activity levels, but it can also be used to monitor and measure change whenever unexpected events occur.
Data is organised into demand categories, which typically consist of a service type, primary support reason and age group. Forecasts are made for each individual demand category but the individual results can be grouped to produce a combined forecast across multiple categories. This is particularly useful where there is a dependency between two categories.
Forecasts are generated with statistical time-series forecasting (TSF) models. These models look at past activity and identify trends and patters (e.g. winter pressures) and make forecasts about likely future use.
Each time we generate predictions, we use artificial intelligence to select the most appropriate model for each demand category. Some models perform better with stable categories and others handle volatile ones better. Our algorithm always selects the optimum model (or combination of models) based on the activity of each individual demand cateogry.
Whenever activity has low volatility (i.e. it is continually changing but over medium to long periods) the Demand Model can be used to generate accurate predictions about future activity. Locking in baseline forecasts enables us to measure how the actual outturn has tracked against the forecast. This enables us to monitor and trigger alerts if activity changes step outside expected thresholds. It is not uncommon to have over 500 demand categories so this functionality is important so that you can ignore the areas that are behaving as expected and focus on the areas that need attention.
There will always be some demand categories that deviate from traditional patterns – particularly if you are actively managing your market and trying to effect change. In these scenarios, the same prediction engine offers some really helpful tools for monitoring and managing that change.
It is simple to measure the actual changes for any category, but by setting a baseline prediction to run immediately prior to any major change, the variance value can be used to measure the impact of the change. Crucially, this is the difference between the actual outturn and the predicted value at that point in time had the change not occurred.
Once a change has established a new normal, alternative forecasts that heavily weight recent activity can be used to make initial projections about potential outcomes.
Actuals and forecasts can be combined together to view the impact of changes across the entire social care market. This is particularly useful for managing events that have a major cross-category impact (e.g. COVID 19 or Brexit). Changes can be broken down by service type, age group or primary support reason which provides visibility of key areas that have been impacted the most.
The Demand Model is available as part of our wider Social Care: Landscape product, or can be deployed as a standalone tool.