How I created a custom attribution model for NA-KD and improved their ROI (Facebook Case Study)

I have outlined in one of my posts how to create a custom attribution model using your own data and not relying on third party tools, where the attribution model is a black box.

When I was working at Facebook as Marketing Science Partner, attribution was one of the most pressing issues for our clients, and I actually published the first case study globally for attribution at Facebook.

The problem

My client, NA-KD, a Swedish fashion ecommerce company and one of the fastest growing companies in Europe, was increasing and diversifying its marketing budget to enhance its growth. The problem was that because they were advertising on multiple channels (Facebook, Google, affiliates, email etc) each channel would report its own figures which made reporting for the whole marketing mix confusing and allocating budget very hard.

Moreover, as a fashion company, they had a lot of ads and content that generated millions of impressions, especially on visual mediums like Instagram and, also, collaborated with influencers. These activities usually are not reported on tools like Google Analytics – which they were using – with a standard last-touch attribution model in place.

The solution

I worked with NA-KD’s analytics team, used data from multiple sources including Google Analytics and Facebook and created buckets of marketing activities (e.g. prospecting, retargeting) and assigned weights to them by analyzing conversion paths.

I then designed a new data-driven attribution model based on Shapley Value to reflect the contribution of each channel/activity more accurately.

Christos Visvardis image-1024x481 How I created a custom attribution model for NA-KD and improved their ROI (Facebook Case Study) — An outline of the Shapley Value Model for Data Driven Attribution. Each color represents a different channel in the conversion path.

The model worked well but had one gap: the fact that the data we were using was mainly click data. In order to incorporate impressions to our model, I ran a series of conversion lift tests to identify the incremental impact of impressions and Facebook/Instagram (which was the main channel relying on impressions).

Christos Visvardis Facebook-Conversion-Lift-1024x505 How I created a custom attribution model for NA-KD and improved their ROI (Facebook Case Study) — How a conversion lift test on Facebook works

The results

As you can see in the published case study, the results were impressive: using the new attribution model, NA-KD saw an 8X incremental return on ad spend for new customers, 5X incremental return on ad spend for all customers, and was able to measure the true impact of Facebook ads which was heavily undervalued by using the old last-touch model.

Moreover, my work didn’t stop there: I designed a plan to run lift tests on a regular basis, in order to calibrate the model every few months and make sure that decision-making is optimized.