Lessons Learned through Modeling Channel Attribution

At Alight Analytics, one of our goals is to help our clients understand what the optimal marketing mix is through answering the question “Where should I spend my marketing budget?” In studying these problems and working through the data I’ve learned a lot; a few lessons are shared below.

1. Channels (TV, paid/organic search, direct mail, etc) do not behave the same way between campaigns. Channel value is often correlated with demographic sets. For example, think about how young and old people interact with social networks. Young people spend more time on social networks and are more likely to respond to an add there than perhaps a direct mail piece.  Further demographic correlation is found when moving beyond medium (social) and into the source (Facebook, Twitter, etc). Assessing a channel value over campaigns directed at different demographic sectors will produce less than optimal model that hides much of the information about channel values. Direct mail may be worth twice the cost of printing and sending a catalog in a campaign directed at middle aged people, but not worth the price of a stamp when sent to middle to upper-class 20-somethings; or you may find that the direct mail piece still produces value, however a greater ROI could have been attained by allocating more resources to social versus direct mail.

2. Maintaining consistency across the inter-channel spend provides very little to model. Some products are advertised on an annual basis; these provide an extreme example. The campaigns might spend 90% of their advertising budget over a two month period. This type of campaign might make it very easy to determine the overall impact of marketing by enabling one to measure the increase in response (sales, inquiries, etc) during and after the marketing campaign. (I say might because the effect of other factors such as seasonality may be unknown.) However, if the marketing mix remains the same across the campaign time and locations, understanding the effect of any one channel is nearly impossible.  The best method to overcome this obstacle is testing. Similar to A/B testing of a website, we can A/B test marketing mixes in selected areas or over time periods. (Don’t forget to spend time designing your experiment up front! A badly designed test can produce flat wrong results.)

3. Seasonality cannot be overlooked. You may be thinking “well that is obvious” but what I learned here is that the response to media can change throughout the year independently of the products utility. Some products are largely season-less like nuts and bolts or replacement light bulbs however; the marketing channels promoting these products are not. Think about the amount of time people (in colder parts of the world) spend inside during the winter. When people are in their homes out-of-home marketing’s value will decrease while online channels can skyrocket. Miss Muffet will sit on her tuffet and browse all day while snowfall incites a “no-way” to outdoor play. (I should have been a poet.)

All of these obstacles can be overcome but must be considered when trying to determine a channel’s value. The first step in overcoming these obstacles is good campaign planning and recording. The goal, duration, and target demographic need to be maintained for use in the model. Following that, online channels must be tagged properly, and a lot of thought needs to be applied to the reach and effectiveness of offline channels and those metrics need to be stored at as specific a level as possible. When storing metrics about effectiveness (impressions, visits, directly attributable conversions, etc) don’t forget to also store the cost. When good data is being collected and stored in a reliable fashion, marketing mix tests should be developed and run.

The lead up time to having a useful marketing mix model can be long and requires some work to set up tracking however, the return from running an optimized mix has the potential to be huge.

Leave a Reply