We're entering a new Age of Marketing Precision (think AMP) brought on by ever-increasing computing capacity. One result of that amped-up computing power is the ability to process complex sets of data faster, and outputs that more precisely predict the relationship between marketing inputs and human behavior.
In short, new modeling techniques can bring brand managers closer to achieving the dream of "push this button, get that result!"
Ad Age recently reported that Procter & Gamble has spent $15-$20 million for modeling that will give the company a real-time read on its marketing mix return on investment (ROI) so it can make faster adjustments and focus on communications tactics that produce results.
Not every marketer has that kind of budgets, and different kinds of affordable predictive models exist for different marketing needs, such as strategy development, product-concept optimization, product-line optimization, and media mix modeling, to name a few.
If you're thinking about jumping on the predictive modeling bandwagon, here are five points to consider.
1. Good model outputs depend on good inputs
Or, as the saying goes: Garbage in, garbage out. A company that specializes in predictive modeling can guide you on the types of inputs needed for a good model, and will supplement your own knowledge of category, competitors, and relevant brand attributes. Look for a modeling partner that understands business and can help you think through the model inputs that will lead to a robust and powerful model.
2. Try to be complete
One goal of modeling is to account for as many variables as possible that might influence the ways people behave. The more complete the data, the more complete the model and the greater its power to predict the things that will influence behavior.
Some areas are often overlooked. As David Rockland, partner and managing director at Ketchum Global Research, and author of the industry's Barcelona Principles, notes, " Public relations is often left out of marketing mix models because it's more difficult to estimate the often indirect impact it has on outcomes like product purchase or consumer attitudes. Yet if that impact is accounted for, it can make a large difference in the predictive accuracy of the model."
3. Patience is a virtue
Because many powerful models are able to incorporate multiple streams of data (e.g., economic conditions, ad spending, social media activity, purchase behavior), marketers sometimes believe that modeling is like creating a stew: Just throw it all in there, and it will come out delicious on the other end.
However, to make data sets talk with one another, it's often necessary to do upfront work (sometimes a lot of upfront work). That is particularly true if it is the first time a given model is run and the model incorporates different data streams from different sources.
But be patient. Once the data sets are aligned, magic will truly happen!
4. Be sure your modeling partner is up to date on academic work
Human decision-making is a hot topic in the emerging field of neuroeconomics. A wealth of academic work that's being done can sharpen a model's accuracy, and many of the tools in older, earlier modeling approaches have been discarded because they are outdated and less effective.
For instance, many early marketing mix models are based on linear regressions (and many companies still use those approaches because that's what their people know), which work well if there is a straight-line relationship between the marketing cause and the behavior effect. However, cutting-edge methods currently being developed will help account for the kind of situation David Rockland mentioned (see item No. 2), where the marketing activity may have an indirect or a lagged effect on behavior.
The modeling tools that are used can make as much of a difference as the quality of the inputs you've given your analytics partners in how accurate the model is. Some of those tools are obscure or very academic, but the members of your internal research department—or the folks in the modeling firm—should be able to explain what they use and why it makes a difference.
If they try to "wow" you with academic mumbo-jumbo, or if you feel you're being asked to buy into a "black box," trust your instincts and look for a partner who invites you into a clear and meaningful discussion about the approaches it uses and the reasons they are relevant.
5. Look for user-friendly outputs
If you're the user of the model, you really don't want a single "optimal" answer on a PowerPoint slide. Ideally, a model should be the "gift that keeps on giving." You should be able to easily play "what if" games. So, when the chief marketing officer comes in and asks "What will happen if we increase our price in Wal-Mart by 50%?" you'll be able to have an immediate answer.
You should look for the results to be delivered in an interactive tool that lets you adjust the inputs to see what impact they have on outcomes, such as market share, brand preference, revenue, or profits.
* * *
In short, think of predictive modeling as a hot, souped-up Ferrari. It's sexy; it's powerful; and, with the right knowledge, it can be a heck of a lot of fun to drive.
