The best big data strategy is the strategy that keeps adapting

If human behavior undergirds your big data strategy, you must constantly be on alert for a shift in how your target is behaving.

The winds of change can bring fortunes or wreak havoc on your big data strategy; however, you have more influence on the outcome than you might think. The newest trend in strategy is staying as adaptable as possible. Some modern-day sages suggest that, instead of building competence within a set of product/market combinations, you build competence by staying flexible around your markets and offerings.

I don't want to debate the merits of this approach today, but it's somewhat contradictory to a strategy that uses big data and predictive analytics to understand and anticipate your target market's moves. As much as I appreciate what the modern-day strategists are espousing, I'm still very much in favor of the latter approach - but there's a catch. If human behavior undergirds your big data strategy, you must constantly be on alert for a shift in how your target is behaving.

Hold still, I'm trying to market

As I've mentioned before, your predictive analytics are only as potent as the resolution between your sample and the population your sample is attempting to represent. What I mean by that is this: The data you're collecting and analyzing right now is a sample - a representative - of your past, present, and future outcomes (e.g., consumer behavior). All samples have error, but the most insidious sampling errors introduce themselves while you're in operation.

I've talked before about sample errors as they relate to size and representation; however, these are analytic design issues that are relatively easy to correct given the appropriate awareness. However, you can design a perfect analytic-based operation that's running nicely for a while, only to discover that your analytics suddenly lose their predictive power. This is an operational issue that's not so easy to catch or fix - but we'll see what we can do.

The truth is it's not typical for human behavior to change abruptly. Most of us are creatures of habit; it's what keeps us sane. If you experimented with a completely new routine every single day, the degree of uncertainty would likely drive you crazy. One ambivalent aspect of human behavior is that we need variety but we also need certainty and stability. This is important for your analytic strategy, because you can key on this need for certainty to extenuate sampling error. But first, you must know that a shift is happening before it impacts you, and that means keeping your eye on the right information.

Control issues aren't all bad

This may seem like a simple trend chart, but what makes it a Control Chart, is that it also monitors the stability or consistency of your average and variance over time. There are different rules that determine whether your information is following a stable pattern - your data scientists know how to configure this. The key for you as a leader is that if your Control Chart sounds an alarm, it may indicate that your predictive super-strategy is coming dangerously close to Kryptonite.

Noticing that usage patterns are dropping off is good; however, ideally you'd like some lead time before revenues start attenuating. I've recently been doing some change management work with Pacific Gas and Electric (PG&E), a utility company out here in Silicon Valley. When PG&E lays steel pipe underground, corrosion becomes huge concern. Obviously, it's not a great idea to wait until the pipes are fully corroded before a notification is generated, so they've installed an elaborate system called cathodic protection.

Cathodic protection not only extends the life of a steel pipe, but it also provides a way to monitor corrosion far in advance of the pipe actually falling apart. What you need is cathodic protection for your target market! Finding leading indicators is a topic for another time. The point here is that your Control Charts should be monitoring both performance and leading indicators.


Natural sciences and social sciences are two different animals - predicting social behavior is difficult and inherently risky. Any strategy that employs big data and predictive analytics to anticipate behavior has two subsequent challenges - getting it right, and then keeping it right. Of course there are no guarantees, but you're not categorically at the mercy of your market either. Put a control plan in place that monitors both leading and performance indicators, and execute this plan as religiously as your execute your marketing plan. You wouldn't want your market to corrode right under your nose, would you?