In my last post I made the case that the future of business will require more reliance on powerful, computer-based analytics that interact with (not supplant) skilled data scientists who bring both analytic and business context knowledge into the process. Modern, effective customer insight analysis requires this kind of efficient cooperation between the analyst and the analytic engine to meet the business objectives of our customer insight function.
While this cooperation between analyst and the analytic tools is critical to effectively producing useful customer insights, it is equally critical to getting those insights implemented within the business. This is especially true when attempting to apply predictive analytics to the way we manage and interact with our customers. We need a system that is designed around the people for whom the insight is produced – a system that allows the insights to support the decision-making of the people and processes that directly impact your customers.
I have been part of many predictive analytics projects over the last 5 years. Nearly all of them have successfully extracted a set of useful insights that have impacted the way companies think about or interact with their customers. However, very few of them have made an impact on the specific actions taken toward specific customers, which is necessary to experience the full power of predictive analytics. For instance, I once built a model that predicted whether or not a channel partner’s revenue to the OEM was going to decrease in the following year. It was a very accurate model, and it highlighted a number of key reasons behind declining revenue. The drivers of decline were immediately seized upon by channel management, and task forces were put in place to better understand and fix them. However, no one was able to find a use for the ability of the model to identify specific partners who were likely to decline in the coming year.
There are many reasons that customer (or partner) focused predictive analytics fail to impact specific decisions made about specific customers. A majority of the reasons I’ve encountered can be attributed to an excessive gap between the point of insight production and the point where the insights result in actions toward specific customers. In the pioneering book on enterprise decision management – Smart (Enough) Systems – Neil Raden and James Taylor dub this phenomenon the ‘insight-to-action gap.’
In my experience, there are two critical reasons that this gap exists for customer-focused predictive analysis projects:
-
The analysis project was not focused on a specific, customer-focused decision-point. Many insight projects start with the goal to see what we can find that will predict an increase in revenue. This is not a bad thing. Sometimes we need to start this way to understand where to dig further. Just don’t expect these projects to yield quick and specific action.
-
There was no forethought or buy-in to the process by which insights will be delivered to the front-line employee for action. This often happens when a customer strategy group gets access to previously unintegrated datasets about the customer and has been tasked with providing insights about a particular issue like low renewal rates. They, rightly, decide that predicting who will and will not renew is more actionable than just looking for drivers of renewal, and a predictive project is launched. I fully support this thinking except there is no definition around getting the ‘actionable’ insight into the hands of those who will act.
The reason both these situations yield a gap between the insight and action is illustrated in the figure above. After an insight is developed, the next defined step is developing a report and presenting it to group of managers or executives. We bring our amazing project and results, and lay it out for them to approve or disapprove of. The best case is that they approve and say, “We need to use this information to intervene and stop the predicted bad things from happening while capitalizing on the predicted good things!” Then what? Then a group of people, which may now include at least a manager of those who will take the action, is convened to create a strategy for operationalizing the insight. By the time the insight actually impacts a customer it could be months later.
Thankfully, there is a better way, and it’s all about beginning with the end in mind. Start by defining the specific customer outcome you want to predict – not something broad like customer revenue, but a specific aspect like customer spend on new products. The next step involves identifying the decision or interaction point at which you will intervene. This requires involving the group who will be responsible for the intervention and designing the process/system to do it. Then we move on to identifying and integrating data, running analysis, putting the insights into action, via the pre-defined and agreed-upon process, and validating the impact. An upcoming post will discuss this process in more detail, but the most important step is knowing exactly what to predict and what you will do with that prediction when it comes. Without that, your well-crafted insights are likely to stagnate. We need to mind the gap.
