Marketing Mix Modeling

By Jerry W. Thomas

Business magazines and dollars. To grow revenue and profits, corpo-

rate executives need to understand the types What exactly is
websites are abuzz with news of marketing investments that are most likely
marketing mix
about the value of marketing to produce viable, long-term revenue growth.
modeling? The
That is, what combination of marketing and
mix modeling as a way to advertising investments will generate the
term is widely
used and applied
help companies maximize greatest sales growth and/or maximize profits?
Eureka! Marketing mix modeling might indiscriminately
returns on their marketing provide some answers to these challenging to a broad range

investments (ROMI). problems. of marketing

models used to
Despite the currency of this topic in the What exactly is marketing mix modeling? The
term is widely used and applied indiscrimi- evaluate different
media, the concepts and tools of marketing
nately to a broad range of marketing models components of
mix modeling date back at least 30 to 40 years.
The topic is of growing interest partly because used to evaluate different components of mar- marketing plans,
of the corporate world’s interest in growing keting plans, such as advertising, promotion, such as advertis-
topline revenue. The last couple of decades packaging, media weight levels, sales force ing, promotion,
have witnessed unparalleled cost cutting and numbers, etc. These models can be of many
packaging, media
staff reductions among the Fortune 500 in types, but multiple regression is the workhorse
weight levels,
the U.S. The opportunities for further cost of most marketing mix modeling. Regression
sales force
reductions are diminishing in number and is based on a number of inputs (or independent
variables) and how these relate to an outcome numbers, etc.
scale, so the pressure for long-term financial
performance from public markets can only be (or dependent variable) such as sales or profits
met by renewed emphasis on new products or both.
and revenue growth. Once the model is built and validated, the
A second reason for the growing interest in input variables (advertising, promotion, etc.)
marketing mix modeling is the proliferation of can be manipulated to determine the net effect
new media (i.e., new ways to spend the mar- on a company’s sales or profits. If the president
keting budget), including the Internet, online of a company knows that sales will go up
communities, search engines, event marketing, $10 million for every $1 million he spends
sports marketing, viral marketing, cell phones, on a particular advertising campaign, he can
and text messaging, etc. No one knows how quickly determine if additional advertising
to accurately measure the potential value of investments make economic sense. But in a
these many new ways to spend one’s marketing broader sense, a deep understanding of the

Strategic Research  Analytics  Modeling  Optimization 1.817.640.6166 or 1.800. ANALYSIS • www.decisionanalyst.com

Copyright © 2006 Decision Analyst. All rights reserved.
variables that drive sales and profits upward is essential in  Industry data. What are the trends in the specific
determining an optimal strategy for the corporation. So industry? Is the market for the product or service
marketing mix modeling can assist in making specific mar- growing? What is the rate of growth? Is interna-
keting decisions and tradeoffs, but it can also create a broad tional trade affecting the industry? Are important
platform of knowledge to guide strategic planning. geographic differences evident within the industry?

From a conceptual perspective, there are two main  Product category data. What are the trends in the
strategies to pursue in marketing mix modeling. One is specific product category? For example, is the
longitudinal; the other is cross-sectional or side-by-side refrigerated soy milk category growing? At what
analysis. In longitudinal analyses, the corporation looks rate? How does this growth vary by geographic
at sales and profits over a number of time periods (months, region? What are the trends by brand?
quarters, years), compared to the marketing inputs in each
 Product lines and SKUs (Stock Keeping Units). What
of those time periods. In the cross-sectional approach, the
is the history of each major brand within the
corporation’s various sales territories each receive different
category? What new products or new SKUs have
marketing inputs at the same time, or these inputs are
been introduced, and when, for each major brand?
systematically varied across the sales territories, and are
What is the history of private label brands and
compared to the sales and profit outcomes. Both methods
SKUs in the category?
are sound, and both have their place. Often some combina-
tion of the two methods is the most efficient.  Pricing data. A history of average prices for each
SKU in the category is essential. Pricing is almost
Regardless of method, marketing mix modeling can be
always an important variable.
successful only if accurate and highly specific data are avail-
able upon which the modeling can be based. The greatest  Distribution levels. What is the history of distribution
barrier to successful modeling is always a lack of relevant, levels for each product and SKU? What is the
specific, accurate data. So the first step in any modeling quality of that distribution and the average number
effort is designing the data warehouse that will support the of shelf facings per SKU?
modeling. The next step is collecting and cleaning all of
the historical data and entering it into the data warehouse,  Retail depletions. It’s essential to have a clean measure
and then cleaning and entering new data on a continuing of sales to end-users, undistorted by fluctuations
basis. Clean, accurate, highly specific data is absolutely es- in inventories. Factory shipments are worthless for
sential to successful modeling. The data must be specific to modeling purposes, in most instances. Retail take-
individual brands and product lines, not the company as a away (or retail depletions) in dollars and in units
whole. Attempting to model at the corporate (or aggregate) (ounces, pounds, cases, etc.) is the most common
level rarely works because what’s going on in one part of the measure of sales to consumers. The goal is to ac-
company is canceling out or confounding what is going on curately easure sales to ultimate users (the people
elsewhere in the company. Here are some types of data to the marketing efforts are focused upon).
consider when developing the data warehouse:  Advertising measures. Money spent on media advertis-
 Economic data. Employment and unemployment, dis- ing is seldom useful by itself. The media advertis-
cretionary income, inflation rates, gross domestic ing must be translated into television GRP (gross
product, interest rates, energy costs, etc. An rating point) equivalents, or some other common
understanding of the effects of general economic “currency.” That is, the print advertising, the radio
variables is vital to building sound models. advertising, the online advertising, etc., must all be
converted into common units of measure (typically,

 Decision Analyst
 television GRP equivalents). The money spent by very expensive, so it’s generally worthwhile to try to
specific media type (adjusted for comparative effec- include sales force variables within the models.
tiveness) is another way of weighting media inputs
 Service effects. If services are an important part of the
as variables. All of this is apt to prove worthless,
customer’s experience in buying and/or using a
however, if copy-testing scores are not included for
product, then this variable must be measured and
each of the ads. A media plan of 100 GRPs per
incorporated into the models. For example, if a
week might have no effect if a weak commercial
new product must be installed by a service techni-
is run, but might have great effect if a terrific
cian, then the interaction between customer and
commercial is aired. Likewise, the exact media
technician can be a major variable, and must be
schedule is important, and the length of time
tracked with some type of customer satisfaction
each commercial is on the air must be considered
survey.
because of wearout effects.
Depending on the industry and product category, other
 Consumer promotion. Consumer (or end-user) promo-
variables might come into play as well. Every company and
tions come in many forms, but the primary charac-
every brand are unique, and identifying all of the relevant
teristic of these promotions (as compared to adver-
variables, figuring out how to measure them, and getting
tising) is the immediacy of the effects. Promotions
those variables into the data warehouse are the most dif-
are designed to have powerful, short-term effects
ficult parts of establishing a successful modeling program.
on sales. Temporary price reductions, cents-off
Most importantly, the data warehouse must be carefully
coupons, and buy one/get one free are examples of
maintained over time and constantly updated, because mar-
common consumer promotions. These promotions
keting modeling cannot be a one-time thing. The models
must be understood, measured, and incorporated
must be calibrated and reweighted on a regular basis, at
into the models. If not fully comprehended, the
least once a year. Many companies hire one or more full-
promotion effects could easily overwhelm the
time employees devoted to tracking down relevant data,
modeling effort.
cleaning it, coding it, and getting it into the data warehouse.
 Trade promotion. These promotions usually take the Often the analytical firm guiding the modeling will place
form of discounts or allowances given to the trade employees on-site to help ensure that the data warehouse is
in order to stage in-store promotions of some type properly maintained.
(temporary price reductions, end-aisle displays, in-
Remember, that the modeling must be specific to an
store signage, local advertising, and so on). Trade
individual brand (or narrow line of business), because what
promotions must be fully understood and included
works for one brand or one company might not work for
within the models because of the sales fluctua-
the next brand or the next company. As a company learns
tions they cause. When the manufacturer offers
what drives its individual brands, commonalities are often
one dollar off the price of each case for 30 days (a
found that make it easier, and less costly, to build market-
typical trade promotion), the retailer is very likely
ing models for its other brands.
to take actions to increase sales of that brand.
Here are some additional rules of thumb to guide the
 Sales force effects. Every company and industry are
modeling work:
different, but the nature and strength of a compa-
ny’s sales force (and how it is organized, managed,  Beware of threshold effects. Often a marketing input
and compensated) can create variables for the (print advertising, for instance) never reaches a
marketing models. Sales organizations tend to be measurable threshold and, therefore, does not
show up as important in the models. But if the

Marketing Mix Modeling 

 print advertising’s budget had been slightly greater, brands will help ensure the success of the modeling
perhaps it would have shown up as a meaningful effort, and will encourage the acceptance, dissemi-
variable (i.e., it would have reached the threshold of nation, and use of the results.
effectiveness).
Who should do the modeling work? Some large companies
 Be sensitive to “lagged” effects. Some marketing have internal modeling departments, but most companies
inputs have immediate effects, while the effects of will outsource the modeling and analytical work. The
other inputs are “lagged” (that is, occur over time modelers, ideally, should have an in-depth understanding
or occur at a later point in time). For example, of marketing and marketing research, so that they really
media advertising tends to have short-term effects understand the complexities of the marketing variables
on sales, as well as long-term effects on sales. they are trying to simulate. True, the model builders need
statistical and mathematical skills, but without the market-
 Keep it simple (at least in the beginning). Focus the
ing knowledge and marketing research experience, the
modeling efforts on a limited number of major
modeling effort is not likely to be very successful.
marketing variables (the big budget items). Don’t
clutter up the models with a large number of Lastly, the issue of cost and ROME (return on modeling
trivial variables that complicate and confound the effort) must be considered. To set up and operate a
modeling work. Once the major variables are truly comprehensive marketing mix modeling program can cost
understood, then smaller variables can be explored. hundreds of thousands of dollars a year, or even millions
per year for a large, multibrand company. Does your
 Be realistic. It may take several years of diligent effort
organization have the stomach for that kind of ongoing
before the marketing mix modeling begins to
investment? Will your company really use the results? Will
pay off. There are no instant cures or short-term
senior management heed the findings? Will the learning be
solutions. It is hard work, trial and error, and a
disseminated throughout the organization to improve stra-
long-term search for marketing truth.
tegic planning? Every company must ask itself these hard
 Seek top management commitment. Involve questions, but if the answers are positive, and the company
the senior leadership of the corporation in the is willing to pursue objective, scientific truth about its
modeling effort, especially at the initial planning marketing efforts, then marketing mix modeling can lead
stages. Their understanding of the nitty-gritty to sustained, long-term sales growth and to improved
details of the industry, the company, and the profitability.

About the Author

Jerry W. Thomas is the President/CEO of Decision Analyst. The author may be reached by email at
[email protected] or by phone at 1-800-262-5974 or 1-817-640-6166.

Decision Analyst is a leading international marketing research and analytical consulting firm. The company special-
izes in advertising testing, strategy research, new product ideation, new product research, and advanced modeling
for marketing-decision optimization.

604 Avenue H East • Arlington, TX 76011-3100, USA

Strategic Research  Analytics  Modeling  Optimization 1.817.640.6166 or 1.800. ANALYSIS • www.decisionanalyst.com

 Copyright © 2006 Decision Analyst. All rights reserved.