From the IIR Conference on Marketing Mix Models

2. Did old models emphasizing price ruin package goods profits? Expanding on the history of modeling at Kraft described by Ron Bates at last year’s WARC conference, I projected some of those results to the whole CPG field. I urged the modeler-attendees to think back to the time when scanner based data started flooding in showing the relationship of price and sales in great detail. It revealed clear relationships we had not seen before. However, when the early modelers only included GRPs to represent advertising, or limited their analysis to differences in sales between stores in the same market, advertising did not account for much of the observed differences. Many concluded price could make sales increase, advertising couldn’t. They increased price promotion, reduced advertising, and one of the economy’s most profitable segments became one of its least profitable. Even when stated in such simplistic terms, not a single modeler in attendance argued with that basic "doom’s day" scenario.

3. Many managers still trust their own judgment more than models. This isn’t exactly new, but the source was – Procter & Gamble. Henri-Jacques Letellier from P&G’s Paris headquarters cited this in describing his efforts to gain greater acceptance for modeling throughout his organization. He found the best solution was to start the process with a cross-functional team so everyone buys in and brings their expertise. A number of modelers from other firms agreed. He didn’t go as far as Kraft had last year in showing earlier skepticism on the part of management may have been justified, but he did cite the need to continuously up date and improve all marketing mix models, and increase management’s math literacy.

4. There was wide agreement measuring the volume of advertising wasn’t enough. Don Schultz of Northwestern derisively called models that just use GRPs, dollar expenditures, or other measures that only reflect total potential exposure, "Media Tonnage Models". They completely miss two key effects:

5. Two ways of including quality of advertising in models were used. One used the type of recognition-based tracking our firm specializes in, and the other used large databases and elaborate mathematics to ferret out quality. This is shaping up as the main area for future improvements.

6. Certain price changes can reduce overall prices, yet increase profits. Craig Stacey of DemandTec showed an example balancing a reduction in an elastic price against an increase in an inelastic price. He said opportunities like this could be spotted by his firm’s new Demand Based Management software. Neil Canter described new off-the-shelf software for do-it-yourself optimization of marketing mixes from IRI. Ross Link, founder of Marketing Analytics described how he partnered with IRI to automate the system so "non-Ph.D." marketers could safely use it. Both types of software utilized Bayesian Shrinkage, an elaborate statistical procedure that uses related data to make stable estimates for parts of the database where samples are thin. Both software sellers stressed its advantages. On the other hand, Gerard Tellis of USC, about to write a book on what we have learned from 50 years of modeling in marketing, took a somewhat different view. He stressed that virtually every calculation that is really needed in modeling can be performed on a simple spreadsheet. Letellier had earlier mentioned differences in P&G on a related issue. Overseas their modeling is done in-house. Here it’s out-sourced.

7. Models are being used for more than package goods. Rachel Kaufman & David Krysiek presented a case history from a brand new product category, XM Satellite Radio, with no history, but an urgent need to develop a media plan in a hurry. USC’s Gerard Tellis modeled ad effectiveness for a telephone referral service. Response time was measured in hours, not days or weeks. Sarah Darin from DDB Matrix and Matt Ragland from Hamilton Beach used test market data to model national sales of a unique new air freshener. Sunil Garga of Media Marketing Assessment compared the effect of online and offline advertising in generating traffic at 50 websites. He, like all the speakers who were optimizing media plans, found a mix of media proved most effective. Wharton’s Dave Bell showed how to identify the point at which the size of the purchase at a lower price store would over come it’s less convenient location. Modeling related to package goods was limited to P&G, and Miller Brewing where Shwetal Patel described how they used linear programming and adstock models to optimize media buying.

8. Agreement is growing on the basic shape of the curve showing advertising’s effect on sales. Everyone agreed you eventually encounter saturation where the curve levels off and further increases in advertising do not produce further increases in sales. Darin & Ragland tested levels that didn’t show any saturation effect but did show a minimum breakeven point. Thanks to John Philip Jones almost everyone now agrees advertising’s effect is greatest when it first appears. Kate Lynch from Starcom showed new brands have it easier than old brands. USC’s Gerard Tellis was the one exception. He showed curves where advertising had little effect until there was enough to pass a "threshold" level.

9. Getting data is a major problem for almost everyone. Stacey indicated the data on store prices needed by his DemandTec program is currently limited to what his clients can provide. All of the clients he listed were stores. He said, although he hopes to use them eventually, IRI and Nielsen take too long to clean up their data. The lack of syndicated data in other fields was the main reason Hamilton Beach and XM Satellite had to get the data they needed through test marketing. Tellis, Link and others made a point of mentioning the problems caused by the unusually large amounts of data required to use dummy variables and Bayesian shrinkage. Modeling a firm’s share of market, thereby including the firm’s competition in the model, was strongly favored. But it required a massive increase in the amount of information that had to be gathered. I said our ability to reflect complex relationships mathematically was a decade ahead of our ability to gather the needed data. I also mentioned that the volume of competitive advertising is fairly readily available, but it takes a recognition-based tracking survey, the type our firm specializes in, to get the needed measures of how good your competitor’s advertising is.

10. Financials are the bottom line. Don Schultz, Professor Emeritus at Northwestern, emphasized that future profits were the real goal, not sales or market share. He felt the next stage in marketing metrics is Value-Based Analysis. Craig Stacey described the "Financial Engine" in his new Demand Based Management software as a tool for achieving this goal: taking the volume predictions from Marketing Mix Models, applying cost data and showing the effect on profits.

These are only highlights of things that caught my interest. For more details, slides for most talks are in a Conference Workbook (# M1486) that can be ordered from IIR at 1 (800) 670-8200.

Don Bruzzone, March, 2002