Many people attend my training program before coming to the Predictive Analytics World conference. In this program, I show why it’s true that, if you can predict it, you can own it. Analytics provides the only data-driven means for an organization to predict. Really, that’s what analytics is for. For organizations eager to grow sales by taking an analytical approach, they need to understand the data that is available right now, lock down clear business objectives, keep the process in motion and on target, and measure the results.
I often work with marketing executives who are accustomed to using data only for looking at the past – the sales report from last year, last month and last week – which is quite a different thing from inducing what will happen. This is the topic around which all discussions are centered at Predictive Analytics World, several times annually. The truth is, business metrics do a great job summarizing the past. But if you want to predict how customers will respond in the future, that requires predictive analytics. By learning from your abundant historical data, predictive analytics provides the marketer something beyond standard business reports and sales forecasts: practical predictions for each customer. These predictions encompass all channels, both online and off, foreseeing which customers will buy, click, respond, convert or cancel.
Sourcing and Selecting the Best Data
The word data only sounds boring when you forget what it means – it is essentially a “history” of your organization. And like a written history, there can be different interpretations, different sources and different ways to look at it. If you were writing a history of computers and your best data source involved PCs, it would make sense to narrow your focus to that area. Likewise, your available data should dictate, to some degree, your first pilot projects with predictive analytics. You are more likely to gain success by limiting the scope in that way. The bottom line is that you can only train an analytical model to predict customer behavior about which you’ve already accumulated plenty of data.
As you gain more success in using your data, cast the net further and begin to plan analyzing textual data and understanding social patterns. A great example of the latter is a cellphone provider understanding which customer might need a retention offer based on the carriers of the people the cellphone subscriber is calling, and on whether those contacts are changing carriers themselves (defecting). This ties in to one of the latest analytics trends (link to “What are the hottest trends in analytics?” article): analyzing social data and social networks to improve predictive models.
Aligning Business Objectives
The most important foundational aspect of an analytics initiative is establishing a clear business objective. Questions you need to ask are: What value will be delivered? What business problem will be addressed? How will the outputs of the analytical process (the predictive scores – often one per customer) be acted upon? What type of operational decisions will be driven by this information? This is not an exercise solely for the IT department, a statistician or a programmer. If the project leader is satisfied by discovering, “Hey, this is cool,” or “these insights are very interesting,’’ there’s no guarantee the project will be valuable. There must be a tangible effect on the bottom line. Since this is about solving business problems, it is critical to gain buy-in from individuals throughout the enterprise, as well as help in forming the questions the data will be used to answer.
Keeping the Process Open
There’s an artistic balance when talking about analytics in terms of business value so that it is not just presented as a mysterious, opaque “black” box. Unfortunately, when you start talking about algorithms and models, it can start sounding math-heavy and arcane very quickly. It’s important – whether using analytics in-house or working with a third party – to make sure the organization can take a peek inside. The people building the models should be able to explain what data they are using and how they are deriving the suggested course of action. If they can’t, this is a huge red flag.
Planning for Success
One of the biggest mistakes companies make in evaluating the success of their analytics efforts is not asking a final question: What will happen when I deploy? For example, in traditional churn modeling, you are predicting the likelihood of a customer leaving so that you can pick the right customer for a retention offer (this being an offer that you can’t afford to send to every customer). But the actual response to the retention offer is not often measured. You need to measure both. You have predicted who is at risk of leaving, but without the second half you can only guesstimate the ROI of targeted retention. But with it, it will be possible to forecast how many customers you will retain, and translate this into a forecasted increase in revenue.
By following these four steps – 1) sourcing and selecting the best data, 2) aligning business objectives, 3) keeping the process open, 4) planning for success – organizations can go a long way to owning their market.
Watch this interview for more advice on ways to execute predictive analytics.