MOVING BEYOND TRADITIONAL MARKETING MIX ANALYTICS
Executive Summary: Marketing Mix Models are fundamental to answering pressing marketing and sales questions, but they need to be spun up faster, more often, and they have to be able to deal with the complexity of today’s promotional ecosystems.
Marketing mix analytics are a lot like baseball, everyone can swing a bat but few can do it well. The complexity and importance of impact analytics have grown significantly over the last year. Once a year analyses of high-level promotion groups and isolated ad-hoc efforts to assess individual promotions miss the insights that a comprehensive “on demand” capability can provide. So what does it take to move beyond traditional marketing mix analysis and set yourself up with a core analytic asset that can be used to answer a host of business critical questions?
DATA ACQUISITION AND INTEGRATION
It has been said many times many ways but analytics will always live and die by the data they are fed, and, as importantly, how quickly new data can be added. Sales, customer attributes, promotions, and spend data can be messy let alone hard to get. Having said that, with established processes and sharing between data providers and data consumers, a painful once a year fire drill turns into a routine monthly activity.
Getting good data regularly is one thing, turning it into something that can be fed directly into analytics is another. Oddly enough, the data scientists who can work miracles with complex machine learning algorithms are terrible at data integration. They want “analytic-ready” files that are already integrated for their purposes. For example, giving a data scientist 40 separate tables of raw unorganized promotion data will inevitably lead to disaster. To be successful, they need one standardized table that includes all vetted promotion data. And, as an added bonus, this table now has all the information necessary for longitudinal or sequence analyses.
MACHINE LEARNING
Given enough time and money, linear regression can be used to answer complex marketing mix questions. But no one has that much time or that much money, and you are still left with the issues of collinearity and non-linearity. The practical application of machine learning to marketing mix modeling has progressed significantly over the last 3 years. What used to take months can now be done in hours with better results. The effect that promotions have on each other can also be factored into individual promotion assessments. Machine learning can handle the complexity of today’s marketing and sales environments. It is not unusual for Sentier to be asked to analyze 40-60 individual promotion/campaign combinations in one analysis.
Machine learning allows the answers to flow from the data with minor tweaks rather than from a heavy set of upfront assumptions and some statistics. Does anyone really believe that the effect of one TV commercial lasts 12 months or that digital efforts accounted for all of the new to brand scripts? Yet, this is the accepted wisdom. With machine learning, hundreds of thousands of iterations can be run to determine how promotions interact with each other and what their individual impact on sales really is.
RAPID AND FREQUENT DELIVERY
The ship has sailed on the year-long, multi-million dollar analytics project that builds the capability but does not deliver the answers. Comprehensive marketing mix analyses need to be able to be run in 4 weeks or less (including data acquisition and integration) since the results are required on a routine basis. This is the only way to answer the many questions about the effectiveness of marketing and sales efforts as promotion mixes and business conditions change. The inputs also need to be well understood and shared with stakeholders in order to be believed. Executives need answers in business time and are increasingly analytically savvy. If you can’t deliver when they need it, and can’t point back to where the answers came from, you are already out of the game.
WHY IS A STRONG MARKETING MIX CAPABILITY SO IMPORTANT?
Marketing mix modeling has a reputation of being a once a year herculean effort performed to provide a rubber stamp on the annual marketing budget plan. Given the data and modeling limitations of the past, this is understandable. However, with the right data, machine learning approaches, and rapid delivery, marketing mix becomes a key commercial asset that is updated as often as needed. Why is this important? Because the information generated from the marketing mix model can be used to answer so many other pressing questions. What would happen if I increased my marketing spend on consumer digital by 20%? How is the CRM campaign that I launched 2 months ago doing? If a disruption event interferes with one or more of our promotional tactics, how do we recover? By ensuring your marketing mix capability is at the next level, you can be confident that you are ready to support the business on their terms and on their schedules as conditions change.
