Marketing mix models: Myths vs reality

Opinion
MMMs’ rise in prominence has been accompanied by some misleading criticisms. Allow me to dispel some of these and work out how to build better models.
Marketing mix models (MMMs) are having a moment in the sun — and rightly — for their ability to help marketers do proper attribution, right-size their ad budget, allocate resources effectively and quantify marketing’s contribution to business growth.
But with the renewed interest draws criticism; some warranted and constructive, some mis-leading and unhelpful.
Let’s start with some of the myths floating around and talk about how to build MMMs that are reliable and useful.
Myth 1: MMMs cannot measure long-term brand-building impacts and its reporting window is only [insert small number]
Reality: All MMMs are not the same. Their capabilities are shaped by the data used in their building and the skills of the analysts leading the process.
Models can be built to quantify impacts over long periods of time, provided long periods of data are provided for their construction. Analysts should not impose on a model an artificial short time horizon; this would limit model accuracy and utility.
The long-term impact of brand communications (and other aspects of brand-building) are an important factor in driving good decisions and should be reflected in MMM design.
Myth 2: MMMs focus on media and does not quantify creative’s contribution
Reality: A good MMM design would take into account both media and creative — and, indeed, other factors as well that would impact business outcomes.
Marketers should look for model designs that cover paid, owned, earned and baseline sales drivers as well, including creative.
A more holistic model, provided accuracy is maintained, is more useful at driving better outcomes than one that is more narrowly focused on media.
Myth 3: MMMs optimise for revenue, not profit
Reality: If an MMM is to feed an optimisation process, it should create outputs that compares an ad cost to the financial benefit created so a true return on investment can be calculated.
In our work, we sometimes use contribution margin; the difference between revenue and variable cost. Or we calculate customer lifetime value if we are building models to support customer acquisition.
The metric varies on the decision to be made. But an MMM need not be restricted to considering only revenue.
Myth 4: MMMs over summarise paid media data, resulting in ROI metrics that hide important underlying variation
Reality: A well-designed development process is looking for variation in the causal drivers of business outcomes. We want to quantify cause and effect as accurately as possible and in ways that are meaningful to buyers of media.
When modelling paid search, for example, it is helpful to disaggregate data into branded and unbranded keywords, as the effect of paid search on audiences using these two is so different. Grouping data by session structure is another way to capture important dynamics.
Again, there is no reason why an MMM need to over-summarise when the data offers other options and the outcomes are so varied.
6 steps to better MMMs
How can we build MMMs that are reliable and useful? Follow these criteria.
• Make sure they are built around C-suite goals — incremental business outcomes such as sales or new customer acquisition.
• Take a holistic approach to the data used. Cover paid, owned and earned media, creative and baseline factors that should be controlled for in order to properly quantify cause and effect.
• Only accept implementation models that are highly accurate. A model should explain at least 90% of outcome variation over the calibration period if it is to be used for decision-making.
• Keep actionability in mind when building. Recognise the decisions the model will support and build data inputs and model outputs accordingly.
• Demand high standards of testing in field. An acceptable model is capable of predicting outcomes with a reasonable level of accuracy, proving the changes in actions recommended are effective.
• Rate models in part on the leverage they create. How much more business can a model help generate than the status quo?
MMMs deserve their place in the sun as a valued part of the marketer’s analytics toolkit. Valid criticisms should always be welcomed by analysts, but outright myths should not be allowed to take up anyone’s time.
David Beaton is co-founder of Navigation ME, part of Crater Lake & Company