Neil Borden, professor of advertising at Harvard is said to have coined the term “marketing mix” first using it in his 1953 presidential address to the American Marketing Association.
Marketing sciences and big data provide us with a process known as Media Mix Modeling. These econometric based models intake and process large streams of data across time periods, called time series, then create patterns suitable for analysis. These models have many uses from explaining weather patterns to predicting the stock market. Like a set of mechanics tools, there are many econometric models optimized for specific applications. For marketing, the key application is evaluating previous spends and testing future plans with high levels of certainty.
Driving thousands of inbound calls per day or a dozen diners to a restaurant, marketing creates the demand that drives the traffic that results in sales. Effective marketing plans use a blend of tactics some sophisticated, like geo-fenced mobile digital, some simple, like a guy on the corner spinning a sign. Since multiple tactics play off each other driving production higher than their individual contributions the marketing challenge is always to find the highest production driving mix of tactics for the lowest cost.Some tactics, like direct mail, digital and direct response TV can have pretty quantifiable returns. Other tactics, like radio and
Some tactics, like direct mail, digital and direct response TV can have pretty quantifiable returns. Other tactics, like radio and print are far less measurable because prospects are typically not polite enough to use the unique 800 numbers embedded in the ads to track response.
Where marketing generated traffic is essential to companies, there is zero tolerance for any demand production risk. But, unless plans continuously adapt and iterate, production decay is a certainty caused by creative burnout and changing market factors.Therefore, marketing generally iterates by testing small changes, observing directional results, taking many periods to enact material adjustments.
Therefore, marketing generally iterates by testing small changes, observing directional results, taking many periods to enact material adjustments. The small samples and time lag skews results induced by burnout. Less quantifiable tactics skew results further relying more on ‘directional’ results or focus groups.
Balancing risk and the need to iterate usually means hedging with overspends as high as 20%-50% to offset the tediously slow test methods.
Media Mix Modeling creates two work products of interest:
- illustration of time series impact on past period results
- application of game theory testing scenarios on future periods
The general project outline to implementation is:
- Capture the time series – obtainable direct and indirect data points affecting sales over as many periods as possible. Some examples include:
- Spends by tactic
- Sales by widget and revenue
- Promotions, offers and discounting
- Competitor advertising, offers or other factors
- Acts of GodConsumer spending index
- Consumer spending indexGeography at any granularity
- Geography at any granularityProcess the time series using the selected model.
- Process the time series using the selected model.
- Iterate the model regressively tweaking to fine tune and account for unforeseen permutations.
Over time, use and testing increase model accuracy building trust as a credible and quantifiable resource. This is why these models are transformational in enhancing marketing, balancing risk and reducing overspends.
The basic plot looks like ‘s’ curves plotting impact rising with spend to the point where incremental yields begin to fall off indicating optimum tactic spend. The illustration shows typical production in one period against various spend levels. The key insight is how 86% of sales come from just 57% of the spend but driving 14% more in sales requires 43% more spend. Therefore, the much lower yield must be deliberate when almost doubling spend for such low impact.
Amassing the data necessary and finding a firm to build a working model isn’t hard. Many firms are captive or on hot standby with advertising agencies for obvious reasons.Costs fall into the range of 2-5% of a total marketing budget to build and maintain credible models but can pay back in less than two quarters. At these returns, marketing plans over just a few million dollars of annual spend are derelict without this tool. Further, testing and pre-staging ‘rainy-day’ mitigation scenarios can make the difference in meeting objectives.
Costs fall into the range of 2-5% of a total marketing budget to build and maintain credible models but can pay back in less than two quarters. At these returns, marketing plans over just a few million dollars of annual spend are derelict without this tool. Further, testing and pre-staging rainy-day mitigation scenarios can make the difference in meeting objectives. The work product must have these attributes:
The work product must have these attributes:
- One month or less interval between period end and model output
- Open source model and process available for peer review
- Views are broken out by tactic illustrating contribution
- Cumulative and tactical lags documented
- Seasonality and periodic data points documented
- Pricing, discounting and offer inputs
- Mix optimization recommendations
- Capability to apply forward-looking game theory testing
Build it and They Won’t Necessarily Come
Media Mix Modeling does not displace experience nor will it replace traditional media planning. Do not expect Media Mix Modeling to be an expert witness in interdepartmental planning meetings making the case for new tactics or creative approaches.
This is especially true for heavy demand based businesses. Marketing plans, over time, develop a mass creating comfort that X dollars spent in particular tactics will create Y downstream sales. Unfortunately, the cost of certainty begins to rise as tactical entropy and subtle market changes require offsets. In risk-averse organizations, the offsets usually mean multiplying X by a factor even if the resulting Y is not proportional, which it never is.
Over time, as modeling induced plans incrementally improve and changes feathered in with successful results, organizational confidence will grow in plans backed up by the model. The benefits of evaluation, insights and forward planning far outweigh the costs of Media Mix Modeling in just a few quarters.
© 2017 Robert P. Smith II SOME RIGHTS RESERVED