Uncovering the Realistic Capabilities of Marketing Mix Modeling

It’s 2017: you’re explaining to your CMO, CFO, and other department leads about the vast opportunities that Marketing Mix Modeling (MMM) can unearth for your brand. You run your audience through various scenarios and examples that showcase these opportunities in detail. The promise of a ‘new way forward’ – one that will enable your organization to more accurately make strategic, tactical decisions relating to budgets, brand health, and more, all of which are rooted in data – hangs in the air.

Fast-forward to today: your audiences of long-ago still quote the promises of MMM, positioning it as the solve for all your marketing troubles. Although your salesman duties were legitimate, it seems that MMM’s presented capabilities seem to have outshone its limitations.

Are your MMM metrics consistently misunderstood across your organization? Does it seem like you’re spending majority of your time defending the model’s results and caveating the interpretations instead of utilizing them?

It’s important for your organization to recognize the realistic capabilities and uses of MMM when interpreting results and including them in the marketing allocation decision process. A first step here would be complete and transparent understanding of what kind of model is being used, what exactly is being estimated, and how. Ultimately, the legitimacy of any model is based on the economic theory that forms the basis for the causal relationships being estimated. Measuring the complex marketing landscape of today is not easy! Factors such as short and long-term effects, interaction effects between marketing channels, and issues relating to aggregated data, will lead to over or under estimation of marketing effects if not properly addressed. These measurement errors can misguide recommendations on budgeting allocations and optimizations, which can in turn lead to a serious loss of credibility – especially in the CFOs office. Thus, a first step in measuring your MMM results would be a solid understanding of your analytics approach; only then can you effectively defend and sell-in the model findings.

Furthermore, MMM should be considered a fact-based starting point to a marketing planning process – not a prescriptive, final answer. More importantly, MMM alone is not your answer; it should be used in conjunction with other quantitative and qualitative data as well as marketer judgement to inform decisions.

Outlined below are four primary use cases of MMM, and their associated caveats.

Use Case #1: Understand your business drivers & outcomes

• MMM is a valuable tool to help you understand business drivers & outcomes such as what tactics drove what percentage of sales, but there are some important nuances that need to be kept in mind in order to correctly interpret those results. Specifically:
  • Be careful when comparing percentage contributions over time as the basis (absolute volume) may have changed.
  • A low contribution from a control variable doesn’t mean it is not important, but it could mean the variable hasn’t changed much. For example, if the number of sales representatives at your company has stayed consistent year over year, and your model shows that these reps contribute only 2% of sales, it doesn’t mean that sales reps aren’t important an important driver of sales, but rather that there hasn’t been much variation.
  • Because marketing is just one of many factors – inclusive of distribution, pricing, product features, seasonality, competitive activity – that impacts and drives sales, overall business results can decline even with marketing contribution and ROI improving.

Use Case #2: Plan and optimize allocation of spend

• As a spend optimization tool, MMM can give a business the inside look into over-performing and under-performing channels so that smarter, more strategic budgeting decisions can be made in the future. However, there are certain limitations in the model that should be called out when using it for this purpose. Specifically:
  • Due to historical limitations, MMM should not be used for spend shifts larger than ~25%. As a of way of combating this, many marketers should opt to use MMM in conjunction with in-market testing to ease their way into these larger budget shifts.
  • Evaluate assumptions that may be changing in the planning period versus the MMM modeling period (e.g. unit costs or execution) prior to making spend decisions (such as allocating spend across channels or setting total investment levels).
  • Marginal ROI (MROI) is dependent on overall spend levels. All other things equal, the more you spend behind a tactic, the lower your MROI. Maximizing MROI in isolation should not be the goal because you can maximize simply by spending less. Instead, marketers should strive to maximize overall economic value (i.e. the absolute dollar return with MROI over 1 or over 1+ the hurdle rate).

Use Case #3: Track marketing performance

• Once implemented, MMM is able to provide teams with a consistent tracking measurement of marketing performance year-over-year. When using MMM for this purpose, remember that:
  • Multiple factors impact marketing ROI: Spend, Unit Cost, Execution, Channel Choice, Mix, Flighting and Business Economics, to name a few. Thus, it’s important to decompose drivers of change when interpreting period over period changes in ROI and marginal ROI and contribution. Without this exercise, you have no idea why something changed.

Use Case #4: Build a test and learn plan

• MMM can be extremely valuable to help your business make tactical “test & learn” decisions that help your organization move forward, at a small scale, with in-market optimizations. However, there remain caveats around what these results can and can’t inform. Specifically:
  • Since MMM models are based on historical data, they can’t provide insight on channels or levels not historically experienced.
  • Test and learns can provide new experience data for MMM models to read, thus allowing the model to understand how certain tactics and channels that may not have been utilized in the past can impact performance.
  • Test and learn results can be difficult to read without considering control variables, such as seasonality, competitive, pricing, etc. MMM can be a tool to read test results and help control for these factors when interpreting results.

These use-cases and caveats show that MMM can be an extremely valuable tool to help you measure, understand and optimize your marketing dollars, but only if it is used in the right way.

This piece was also published in MarTech Advisor on 6/11/19, here: http://bit.ly/2IDZt2X