During the campaign wrap-up report delivered by her marketing agency, Laura was awash in data, graphs and metrics on the various channels and tactics her team deployed during the campaign. She struggled to see how the campaign take-aways could inform the larger media-planning approach or multichannel marketing strategy being developed for next year. Did the TV ad campaign move more SKUs? Did investment in display or paid search drive more sales?
Increased spending and more complex consumer journeys increase the pressure to measure the true impact of marketing. For Laura, it was time to kick off a more formal marketing measurement project to better understand the impact of various tactics and channels.
“Advanced attribution and marketing mix modeling promise greater fidelity of spend analysis and optimization, but bring their own cost and additional complexity,” noted Martin Kihn, research vice president, Gartner for Marketers.
Attribution and marketing mix modeling explained
Attribution examines multiple marketing exposures along a user’s path to purchase and is often referred to as multitouch attribution. Attribution uses individual-level data rather than aggregated data and as a result is usually confined to digital marketing channels, particularly display advertising, paid search, affiliate referrals and email. The challenge is how to use individual user data to estimate the impact of an event on a desired goal (e.g., how many clicks registered for the test drive.)
Marketing mix modeling uses aggregate data, not user-level data, and explores the impact of a wider range of channels and factors, including non-digital media such as broadcast television and radio, as well as retail promotions and coupons. Marketing mix modeling has more academic rigor, uses better-established analytical methods and is more widely used than attribution.
Both of these techniques are highly complex, time- and resource-intensive undertakings, but the benefits can be significant. According to interviews with practitioners, these methods can yield 20% to 30% improvement in the efficiency of marketing spending, primarily by optimizing media.
Use a five-step process to know how to start and what to expect with attribution and marketing mix modeling.
- Define the goal
Begin with the end in mind and set up key performance indicators (KPIs). These success metrics must be available, accurate and in a form that can be used for analysis. - Build the team
There’s no getting around the fact that measurement projects require trained analysts. While skilled data scientists with knowledge of the business can build marketing mix models, attribution models are often more complex and require additional skills such as facility with big data systems. - Prep and collect data
Data preparation is generally the most frustrating and time-consuming phase of measurement projects. Identify, inspect, clean, transform, transport and maintain important data sources. Reconciling taxonomies and naming conventions, assessing accuracy and usefulness, and obtaining access and permissions will likely challenging. Ad server logs that register an “impression” often do not indicate whether impressions were even viewable. - Perform modeling
Modeling includes data exploration and visualization, training, testing, trial and error. An experienced internal advanced analytics team, outside providers specializing in marketing measurement or other expert guidance is recommended. Attribution models include basic methods like linear and time-decay, as well as algorithmic methods such as logistic regression and random forests. Marketing mix modelling approaches are typically based on regression, structured equations and Bayesian hierarchies. - Use the results
Turning findings into useful recommendations is a major challenge. An organization’s ability to use the output of marketing mix and attribution models separates success from failure. Prioritize recommendations and push the operations team to incorporate approved changes into day-to-day operations. Don’t forget to assign someone at project kickoff to capture findings and maintain a detailed log of changes.
