One of the big questions on the table for brand marketers is what to do with Big Data. The presumption is that more data means better marketing, but finding the path from more to better is the challenge at hand.
A big part of this challenge is that the flood of data is misunderstood. The term itself, Big Data, focuses brand marketers on the amount of data, an orientation reinforced by infographic hyperbole about the supernova of bits and bytes sweeping through the brand marketing galaxy. But more data matters only if it’s better.
Getting something better from Big Data goes beyond the data itself. In fact, it depends mostly on the ways in which data are analyzed. What the Big Data revolution has stirred up is less about amount and more about analytics, but this is not something that comes naturally to most brand marketers.
A recent survey of marketing professionals by the IBM Center for Applied Insights found that 40 percent are well behind the curve of the analytics required for Big Data. Another 37 percent are further along, but still “limited” and “struggling.” In other words, a little over three-quarters of brand marketers are not keeping pace with the analytics needed to ensure that Big Data produces better outcomes. As Ari Sheinkin, IBM’s VP for Client Insights, put it in an AdAge op-ed, brand marketers are “stuck in a time warp, channeling their inner Don Draper.”
Most worrisome is the finding from this survey that fewer than one in five of this three-quarter slice of brand marketers brings a “scientific approach” to research and analysis. Relying instead on gut and grit may explain why only 23 percent of all marketing professionals say they are “highly effective” at building value through new insights, only 25 percent at capturing new markets and only 32 percent at engaging individual customers. Big Data alone won’t improve any of this. Indeed, more data will make all of it worse if brand marketers put it to use unscientifically.
The first question brand marketers should ask about Big Data is not what to do with it, but what not to do with it. Knowing what not to do is also the best way to see what can be done.
Much of what’s being touted about Big Data nowadays is a fallacious trumpeting of amount as putting an end to the need for scientific rigor and precision. By giving marketers a seemingly authoritative excuse for continuing to be unscientific, brands are put at risk. Big Data requires smarter not cruder analytics.
The biggest misconception is that Big Data “makes the scientific method obsolete.” This was Chris Anderson’s inexpert headline in a 2008 Wired article in which he proclaimed, “Correlation supersedes causation, and science can advance even without coherent models, unified theories, or really any mechanistic explanation at all.” Anderson’s rash pronouncement has been echoed by Oxford professor Viktor Mayer-Schonberger and Economist editor Kenneth Cukier in their 2013 Big Data bestseller in which they spend an entire chapter cheering the supposed triumph of correlation over causation.
Much of what’s behind these jabs at the scientific method and causation is the misguided belief that science is needed only when data are incomplete. This misperception stems from the fact that the error range typically reported with research results reflects the error associated with sampling. With all the data in hand, sampling error (presumably) goes away. Since that’s what most people think of when they think of research error, it’s no surprise that many come to the mistaken conclusion that Big Data will eliminate all error, thus obviating any need for the scientific method or causation.
But it is naïve, or at least overly hopeful, to suppose that whatever patterns or connections the naked eye can see in a universe of data are free from error. More data is no cure-all.
As discussed in a prior Branding Strategy Insider post about the enduring need for random sampling, humans have a hardwired, built-in propensity for seeing patterns. While this helps us, it also hurts us. It’s not hard to trip up our eyes and minds, as even the simplest optical illusion reminds us. With experience, we learn that we have to be disciplined and careful as we navigate our lives. Yet even so, we are still inclined to see patterns or connections where none exist, a problem that is particularly perilous when we are working with large datasets to make important policy or business decisions.
This is the value of the scientific method. It is a systematic framework for evaluating data that, properly applied, substantially reduces the likelihood of seeing patterns or connections where none exist. There is no magic threshold of data volume above which the scientific method suddenly goes poof and disappears. Whether more data or less, a scientific approach is essential for marketing to continue advancing beyond the limits of success possible with gut and grit.
This is not to suggest that more data won’t make marketing better. It is just to caution that when processed unscientifically, more data can be worse than less data; indeed, much worse.
As Nate Silver of FiveThirtyEight fame, now at ESPN, noted in his 2012 bestseller about forecasting, The Signal and the Noise, dearth of data is not the reason why economists can’t predict their way out of a paper bag. The Federal government publishes 45,000 economic indicators every year; private firms, over four million! You can plumb this “stack of economic indicators as thick as a phone book” to your heart’s content, but all you’ll get are “spurious,” “coincidental” relationships, nothing “substantive.” Only three to four dozen of these variables really matter. They are well known already – more data isn’t needed to find them. And they are understood to matter because they are rooted in theory – wanton data mining won’t add or subtract from that.
But even these few dozen variables aren’t very good predictors of economic outcomes. So some economists try to remedy that by uncritically piling up a large number of predictive variables, eschewing theory and causation for more data and correlations. It doesn’t help. As Silver has put it, “[I]f you just look at the economy as a series of variables and equations without any structure, you are almost certain to…delude yourself…into thinking you are making good forecasts when you are not.” In summing up the story of one economic forecasting firm that made an especially erroneous prediction in 2011 using this approach, Silver noted that it failed because “[i]t had a random soup of variables that mistook correlation for causation.”
Even more pointedly, Silver adds, “[T]here isn’t any more truth in the world than there was before the Internet…Most of the data is just noise, as most of the universe is filled with empty space.” This bears repeating – more data doesn’t mean more truth; it just means more noise, and thus a greater need for the scientific method to keep us from seeing patterns or connections where none exist. Unscientifically piling on Big Data only degrades what we know by clouding our understanding with randomness, chance and coincidence.
The problem with economic forecasting is not a lack of data. It’s the poor quality of both data and theory, along with misaligned forecasting incentives that reward meager analytics. More doesn’t help; better is what’s needed. This should be the Big Data focus of brand marketers, too. It’s not about big; it’s about better.
What brand marketers should do with Big Data is bring it to bear in areas where more means better. Four such areas are noteworthy.
First, more data can help in areas in which little or no data are currently available. For example, Big Data has already had a big impact on pricing. In years past, data for pricing decisions was sparse or non-existent. But dynamic, real-time digital data is enabling brand marketers to capture value from pricing inefficiencies previously hidden from view by a lack of data. More data matters here, but it matters because more means better not because more means more.
Second, more data can help in areas in which existing data are flawed because of things like weak or indirect measures or untimely, slow collection and reporting. For example, Google search trends might be better early indicators of many brand marketing metrics ranging from consumer confidence to unemployment to economic slowdowns. Big Data can also be used to create tracking scales and risk indices that are impractical with existing data or to augment and improve existing measures. Again, Big Data makes a difference because it is better not because it is more.
Third, more data can enrich established, well-understood models of marketing engagement. One simple yet important example is personalizing the shopping experience of consumers. Personalization is a known, proven marketing model (e.g., One-to-One, CRM, mass customization, etc.). Big Data isn’t needed to unearth correlates about the value of personalization. What Big Data can do, though, is improve the ability of brand marketers to execute personalization. The more that brand marketers know about what consumers prefer, the better they can personalize what consumers get.
Finally, for some businesses, a surfeit of data facilitates more robust experimentation. Google gets lots of headlines about its fervent commitment to nonstop testing and continuous incremental improvement. But for other companies, too, Big Data opens up feedback loops and performance details that make trial-and-error easy, quick and affordable. As many experts have insisted, with experimentation so simple and accessible, A/B testing, not gut, grit or theorizing, should be the heart and soul of brand marketing.
But this is also where many observers get carried away. Running experiments and testing ideas on real consumers doesn’t mean an end to the scientific method and causation. The principles of good experimentation are the same as ever, and these principles constitute the scientific method. Figuring out if an A/B test result is a real improvement or a chance finding still requires sound scientific protocol and causal analysis. Big Data makes more experimentation possible; it doesn’t make experimentation less scientific.
Beyond these four areas, brand marketers should proceed with care. Big Data must be used for specific ends and particular purposes, not for fishing expeditions. The worst thing that brand marketers can do is poke and probe and play around just to see what turns up. As Silver noted, that’s the sort of mindless, indiscriminate data mining in which you will find that ice cream sales are correlated with forest fires because both are more frequent in the summer. But it would be a misguided marketer who tries to use that correlation to sell more ice cream. There are no new hypotheses or models about what drives ice cream sales. There are no new insights about engaging consumers. There are no new indicators of attitudes and behaviors. There is nothing here other than a coincidental correlation between ice cream sales and forest fires. Obviously, this is an extreme example, but it is illustrative of the very real danger lurking in the unscientific handling of Big Data.
Big Data is a product of the digital age. Everything big about Big Data is premised on the explosive growth in digital sensors, digital signals and digital trails. Digital is the biggest thing facing brand marketers, of which Big Data is but one part.
It is easy for brand marketers to get so wrapped up in all things digital that Big Data gets taken for granted. It looks to be part and parcel of a transformative revolution in which none of the old rules apply. But that’s not true of data. Whether big or small, data are data. The same rules and safeguards that apply to less data apply to more data, too.
For Big Data to add value it has to be better not just bigger. There is no value in it simply because it’s more. Big for the sake of big is sure to mislead. More without better is less. Big Data must improve brand marketing, and when it does that it’s not big, it’s better.