Business leaders are still making many important decisions in an ad hoc way – gut feel, what has or hasn’t worked in the past, or what competitors are doing.
The huge amount of data collected by companies and the tools to analyse it are enabling a better way to make major decisions, allowing business leaders to gauge the likely outcome of their plans before they put them into action.
Machine learning and artificial intelligence are informing decision making and validating decisions by providing leaders with visibility about how the different options will play out.
There are four stages to building a predictive decision-making tool:
1. Describe what happened in the past by collecting and sharing historical data.
2. Analyse why it happened by looking at patterns and common features for different outcomes.
3. Use the data and the analysis to predict what happens next, and test and refine the tool on historical data.
4. Stage four is where the real value comes in. This is where we suggest what a decision maker should do to get a given result. If you can predict what’s going to happen because you know various characteristics and consequences, then if you pull some levers you should be able to change the result.
A major issue for subscription based businesses is that when they increase prices they need to understand the implications regarding their subscribers and identify buyer behaviours, so as to design targeted offers that minimise users departing the service. The starting point was to collect historical data about past price rises and resulting ‘churn’ – customers who cancelled their service. Then we went back through five years of data and looked at the characteristics of customers who cancelled their subscription, things like calling customer support to complain about the cost or quality of service or making late payments.
Next, we built a predictive model using the first three years of historic data and tested the model on the following two years of data. After we refined the model, we could predict with high accuracy which customers were likely to give up their service.
We now had the data to help make better pricing decisions. Rather than pushing through a standard price rise we were able to propose targeted offers to reduce customer churn and improve overall business returns.
As a result, our client gained a significant increase in revenue with very little churn, showcasing the value of data modelling for any business.