Despite the ever-changing world we call marketing, the concept of media-mix modeling has withstood the test of time. Now and then—1989, when the idea was revolutionary—media-mix modeling is used to analyze sales data to determine the effectiveness of the marketing mix. This is accomplished by massive data collection, followed by the building of an elaborate statistical model to experiment with different levels of marketing spending via the various channels. When done right, a company can see where its efforts are most effective.
“Instead of asking questions to the customer, we started with scanner data,” said Gloria Rosenberg, founder and president of Market Fusion Analytics, a New York-based media-mix modeling consultancy, in an interview with CMO.com. “We were finally able to deal with large databases and performance data.”
In fact, the results at General Foods, which Rosenberg said was the first company to use the technique, and other pioneering companies were so successful that media-mix modeling spread rapidly. Today media-mix modeling is a standard tool within big consumer goods companies, as well as B2B companies, providing tighter control over the media budget, and offering an indicator of what works and what doesn't.
With the growth of e-commerce, media-mix modeling has become popular on the Web as well. According to Bob Gooze, a principal at Santa Clara, Calif.-based Customer Manufacturing Group, the Web offers some significant advantages for media-mix modeling. “The Web gives you intermediate feedback,” he explained to CMO.com. “When you do a direct marketing campaign, you don't know how many opens you get. With the Web, whenever they open it, you know it, and you know what else they did.”
With Knowledge Comes A Price Tag
Although the price of using media-mix modeling has dropped substantially with more powerful desktop computers and better software—making it an option for midsize and some smaller businesses—the capability still requires considerable investment because, for one, each company's model must be custom-built.
“If you figure the low end of media-mix modeling of $45,000,” Gooze said, “the reality is that if you're not spending 10 times that, you're not going to make that investment. As soon as marketing spend is above a quarter-million, media-mix modeling is something to consider.”
Building a good model requires a high degree of statistical sophistication. Take, for example, the matter of correlation versus causation. As they tell you in basic statistics, “correlation is not causation.” In other words, just because two events happen together doesn't mean one causes the other. Part of the difficulty of building an effective media-mix model is teasing the real cause-and-effect relationships from the mass of events that occurred simultaneously.
For example, if on the day you start a new test campaign for your soft drink you see sales in the test area drop off sharply, does this mean the campaign is a failure? Not if that's the day your test market area was hit with a massive thunderstorm. Unfortunately, most correlations aren’t so obvious, and it takes statistical magic to find the real effects of your marketing campaigns.
Suffice to say, the model must be very carefully constructed. “There are two or three conditions that need to be met,” said Dominique M. Hanssens, the Bud Knapp Professor of marketing at UCLA's Anderson School of Management, in an interview with CMO.com. “You need someone in the organization who understands statistics and econometrics. You need data on business performance, and you need data and software.”
Even before that, “It starts with internal discipline on keeping marketing records,” he added. “I've done it a couple of times for smaller organizations, and it has actually worked quite well.”
In addition, marketers must understand which elements they’re in control of, Gooze said. “It's important to understand what you're doing that could affect the outcome,” he said. “It's also important that what you are doing changes over the period of interest. Change what you're doing and measure it exactly.”
Modeling Under The Microscope
A model is a mathematical relationship showing the relation between various factors—in this case, factors affecting sales. It shows what happens when the enterprise changes one factor, and how it affects the other.
That's the simple explanation. Such models are usually examples of multiple-regression modeling, a technique that attempts to account for all the factors involved in influencing sales and how much of an affect each factor has.
The factors a marketer is most interested in are the ones the enterprise has control over, such as price and promotions. However, the model should also include factors that aren't under enterprise control if they could have an impact. In our hypothetical soft drink example, if the weather had been hot and sunny instead of a major thunderstorm when the product launched, then soft drink sales would have been higher.
“It seems extreme to worry about the weather, but it's important to understand macroeconomic data,” said Katherine deSesa, a statistician at Geomomentum Insights, a Chicago consultancy that specializes in helping chain stores, in an interview with CMO.com.
In addition to technical sophistication, the modeler has to know enough about the business to include relevant factors in the model. “It's an art, and it's also understanding business,” Rosenberg said of building good models. “To predict people's behavior, you have to build your model, but after you build it, you have to understand the 'so what?'” -- in other words, the various factors interacting in this business at this time.
If this kind of sophistication sounds complicated, you’re right. Most businesses hire a consultant to manage and update the model. For companies interested in beginning media-mix modeling there are several requirements, chief among them: data. If a company does not have sales data and marketing data to compare it with, then it's very difficult to build a media-mix modeling model.
Another all important requirement: Identify your company's internal champion, Rosenberg said. “It should be someone who is involved in key business decisions who can think strategically,” she said, followed by building support within the organization.
It's also important to understand that media-mix modeling results “aren't a report card--they're a business tool," Rosenberg said. "You need to use it not just as a reporting tool, but as a growth tool. Use it as a planning tool going forward.”