Time to Drill Deeper

No, this is not an article about fracking — drilling for gas and oil in shale. This is about “drilling down” into big data. We’ve been using the term for a long time and it provides a useful metaphor for data analysis. However, we’ve conditioned ourselves to think of drilling down only to a superficial degree, and that needs a rethink.

When data wasn’t big and analytics relied on less robust hardware, we were only able to scratch the surface of our data, a practice that survives to this day. “Scratching” often means looking for insights only at the end of business processes. So, for example, we look for signs of churn next week or the next best offer today, or we attempt to forecast the next sales deadline. All of this is valuable but not enough.

If we’re doing our jobs right, we should be using powerful analytics to perform root cause analysis in order to better forecast events, so we can either avoid them entirely or further enhance our likelihood of success.

Our Prediction Problem

What if you could go further into your data so that rather than simply discovering someone or some business that was about to leave your service (churn, nonrenewal) you could find those moments of incipient danger and correct a problem at the source? You can — but it requires change, not more hardware or better software. Those things are always welcome, but there’s a different way of framing the challenge in front of you.

Too often, we make assumptions about some aspect of business and then collect and analyze data about it. That’s a good approach, as long as the assumptions are valid and accurate, but too often they are not.

When we assume something, we are building an ad hoc model of what we believe reality is, and that’s a good thing. Modeling is the heart of all kinds of progress in any number of fields of human endeavor, but it’s not something we do particularly well in business — with some exceptions.

“We need to stop and admit it: we have a prediction problem,” writes Nate Silver in The Signal and the Noise: Why So Many Predictions Fail – but Some Don’t.”We love to predict things — and we aren’t very good at it.”

You might recall that Silver called 49 out of the 50 states correctly in the 2012 presidential election. This man does not have a prediction problem.

Retailers might be an exception; they model heavily and they do a good job. They collect customer and store data so that they can model the ways they set up stores and plan the assortments they stock. Those models mirror very closely the clientele and traffic for an individual store. When it comes to online business and B2B business, we aren’t there yet, because it’s both a different and a harder challenge.

Doing the Two-Step

Finding a solution in the online world starts with figuring out your model before you make any assumptions, and before you implement something. (This is not your business model but your approach to customers, which is part of the business model.) It’s surprisingly easy to do if you take a two-step approach to analytics.

Step No. 1: Build a realistic model of your business by asking your customers. I call this “discovering your moments-of-truth,” and I write about it in my new book, Solve for the Customer, which will be available shortly. As you know if you read this space often, a “moment-of-truth” is simply any time your customers expect you to live up to a promise — whether that’s a product, company or brand promise — irrespective of whether the promise is expressed or implied.

Step No. 2: Build in customer-facing processes based on your moments-of-truth. Your processes and supporting software will meet customers where they live, so to speak. The best way to do this is with journey-mapping software, because it lets you examine all the contingencies and define subprocesses as appropriate. It’s also the logical place to define metrics that will tell you if you are meeting your goals in your moments-of-truth.

For example, customer onboarding is a good example of a moment-of-truth, and there are many analytics vendors that focus on customer health as a function of how quickly customers get down your learning curve.

For instance, there is a direct correlation between customer longevity and how fast they onboard, people at Scout Analytics have told me. Knowing this, smart vendors deploy customer success managers to ensure that onboarding is swift and trouble free.

You can identify moments-of-truth like this throughout your customer lifecycle, and often those moments do not automatically require expensive human intervention. Still, having a moments-of-truth approach, plus good analytics rather than assumptions, enables a vendor to deploy resources where they’ll be most beneficial to both the customer and the vendor.

None of this is hard. In fact, once you change your frame of reference (aka your “paradigm”) from ad hoc assumptions to dedicated and conscientious modeling, it flows. When we move from random approaches to modeling, which incorporates a bit of statistics (and that is what analytics is about to a great degree), we pass from a framework of art to one of science. That’s what’s happening right now in many areas of front-office business, and it’s why I’m saying we’ve arrived at a new science, Customer Science.