Predictive Modeling for Marketing Mix and Attribution in the Beverage Industry

There is general acceptance in the alcoholic beverage industry that linking marketing spend to consumer purchase is very difficult, if not impossible – especially for smaller producers.

We don’t agree.


How It Came to This

The alcoholic beverage industry is complicated by regulation and relationships. For example, in the US, there is a three-tier system that separates production, distribution, and consumer sales. This means that for a producer, consumer sales data is passed back through multiple channels. The data is often incomplete and/or inaccurate. And the specific terms of pricing at different layers of distribution are typically opaque. Add in cross-border shipments and you add another layer of complexity.

Measuring the impact of marketing is further complicated by the retail channel. Bars and restaurants (known as “on-trade”) can drive sales in liquor and grocery stores (“off-trade”).

Connecting the dots between marketing effort and consumer purchase is obviously difficult – especially compared to online product marketing where there is a clear path from advertisement to conversion point.


There Is an Answer

It’s not as easy as tracking ecommerce sales attribution, but we believe there is a way forward for the beverage industry. There is enough meaningful data that can model cause and effect — if you look in the right places and prepare to adjust your attribution model with experiments.

All the data needed to answer this should be available to beverage companies:

  • Paid marketing and advertising efforts, impressions, or equivalent
  • Earned media (PR, press, events) efforts, impressions, or equivalent
  • Depletion data from distributors
  • Website analytics
  • Search trend data around brand, competitors and product category

There are five phases to the analysis:

  • Collect the data from the different sources.
  • Apply a time value to the different events within the data sets to the level of known accuracy (i.e., hour, day, or month).
  • Aggregate and reshape the data against the time values so all events can be seen on the same timeline.
  • Analyze the relationships and dependencies between the different events/variables 
  • Create and tune a predictive model

This needn’t take the most advanced and expensive software; in fact, some of the best software is open source. Success is based upon matching analysis with commercial knowledge and instinct.

Hunches grounded from years in the beverage industry coupled with analysis and new predictive modeling can answer the question: “Where do I spend my next marketing dollar?”

Want to learn more about how we can help your organization? Drop us a line at