Prevention is better than cure, so the old adage goes. Many businesses and organisations, especially in the wake of an unexpectedly poor result, spend time poring over their data to try and glean some understanding of what has happened. Even if it is possible to deduce what went wrong, how can further nasty surprises be avoided, or at least anticipated? Analysis of past data and monitoring of current performance are both essential, but there is a third piece to the jigsaw: predictive analytics.
A growing number of our clients are beginning to increase their awareness of predictive analytics and the positive impact it can have on their outcomes. Predictive analytics utilises a wide range of data mining, analytical and statistical techniques to make forecasts about future events based on current and historical data.
Using predictive analytics means that businesses no longer have to rely on intuitions, guesses or assumptions, which although well-intended can often be misguided and over-optimistic. A more robust approach allows decision makers within organisations to proactively identify potential risks and opportunities, and to take appropriate steps towards securing the most beneficial results, whether this means increased sales, higher customer satisfaction scores, improvements in savings or any other KPI (key performance indicator).
One type of data that is particularly amenable to this approach is the time series; these are ideal for use in predictive analytics as they are highly structured. Time series are commonplace within many businesses; any KPI that can be measured quantitatively and has been tracked over time can be represented by a time series. For example, this might be historical sales figures, the number of patients with a certain disease, or the population at a location. There are numerous ways of analysing these time series, but our preferred method involves decomposing the data into its three component parts: trend, seasonality and remainder.
The trend indicates the long-term increase or decline in the average level of the data. Seasonality allows us to see when a series is affected by recurring seasonal factors; this could be the day of the week for time series measured daily, or the month of the year for longer-term time series. There may even be a combination of seasonal effects at work. These effects can be very subtle and are easily missed by the human eye. The final component of a time series is the remainder: this includes any fluctuations that cannot be accounted for from the trend or the seasonality, and depending on the level of significance might be ignored or analysed further to identify any important causal factors that can then be included in the forecast.
At Musgrave Analytics, we have considerable experience in predictive analytics, having recently undertaken forecasting projects for a range of companies in diverse fields such as social housing, manufacturing and healthcare. Here are a couple of examples:
This is an illustration of a tool we recently developed with a client who wanted a way of anticipating whether end-of-month sales targets would be met by tracking sales during the month. We analysed several years' worth of monthly sales figures to determine a normal sales pattern. As new data comes in each day, a forecast is built up that predicts the likely outcome based on the figures to date and calculates the probability of meeting the target by the end of the month.
This chart shows a simple graphical representation of a forecast. The three lines shown are the current performance of a KPI (in this case current tenant arrears), the forecast of the outturn of the KPI by the end of the financial year and the target value. Again, the forecast is updated throughout the year as new data becomes available.
In this article, we have explored how predictive analytics can make a positive difference to the outcomes of an organisation, whatever its goals. Good quality data, particularly in the form of time series, plays a crucial role in this.
by Dominic Nelson