The Triangulated Attribution Model

triangulated attribution

This is a practical, conceptually sound attribution framework we have been developing with Mr. Tommi Salenius who is the Head of E-commerce in a fast-growing international e-commerce company.

The point is this: >90% of e-commerce companies (just an educated guess based on our experience) solely rely on analytics to attribute sales to channels and campaigns. Analytics can be based on conversion pixels in ad platforms or analytics tools like Google Analytics or Shopify Analytics.

There are multiple situations where analytics falls short in terms of correctly attributing sales, most of which deal with missing datapoints. These problems are well-known among analytics professionals.

So, we need something else in addition to analytics. Applying the scientific principle of triangulation (i.e., finding multiple sources of data or using multiple methods to arrive to the same conclusions), we now ask, “What other data can be used?”.

The first additional data is self-reported attribution. After a conversion takes place, we ask customers what channels/campaigns influenced their purchase. Tommi and I have observed cases where customer responses almost completely differ from analytics reports, in cases like influencer marketing where there is a major drop rate of analytics data, yet influencers had a major effect on the purchase decision for customers. So, self-reported conversion can tell us about this. It’s also common sense: let’s ask consumers what they think affected their behavior.

Because self-reported data can suffer from reliability issues, such as poor recall resulting in omitted or erratic information, we could do with additional data. Hence, our third data source is MMM, otherwise known as marketing mix modeling. This is an old approach based on econometrics: we have some observed data (marketing spend per channel/campaign and sales) organized in individual time series. MMM aims to find a connection between the marketing spend and sales (btw, for this, we recommend the SaaS tool Cassandra by Gabriele Franco & team). Based on the model, we can estimate how much each channel or campaign influences sales, and we can also use the model to calculate how a budget should be split between the channels or campaigns to maximize sales. MMM requires a lot of high-quality data the model always involves an error rate of some degree. So, it’s not without its issues either.

No single source of attribution is perfect. They all have shortcomings. Even their combination – the triangulated attribution model – is not perfect. But it’s much better than relying on one source alone. In the practical sense, if 2/3 attribution approaches refer to the same channels or campaigns being important (or not), we have much stronger grounds to believe this story than when we have evidence only from one attribution approach.