We focus on the campaigns sent via the email channel. We augment data by including more variables that we see from our side, such as the percentage of the customer segments the email campaign is reaching – for example 20% mass customers, 50% advance customers, 30% premier customers – and what times the campaign is going out. It might be that 20% of emails are sent in the morning and 80% in the evening.
We assess and quantify all the correlations between the variables we or the client can control, including campaign volume, product and segment split and the KPIs. By looking at these correlations, we can then predict the efficiency of a future campaign by combining these variables and the correlations found in the historical data.
Who came up with it and where did the idea come from to create it?
The idea came from our head of data science Henrik Nordmark and me. It originated from ongoing research and development that we carry out as part of our knowledge transfer partnership with the University of Essex. It was then developed by our teammate Diyora Makhmudova, who named it Babochka – ‘butterfly’ in Russian.
It is the first step in our long-term mission to create predictive attribution modelling that will determine the optimum marketing mix of direct mail, email, SMS and other channels.
“We believe an algorithm is almost always better than the human eye at recognising patterns in data”
Almost since joining Profusion, I’ve been working very closely with our in-house marketing effectiveness team for a major retail bank, which is using the Fernando report to assess the impact of its marketing communications.
The first thing Henrik and I noticed is that they are essentially trying to identify these correlations with the human eye, by looking at campaign KPIs and trying to see what went wrong or what went well. As we believe an algorithm is almost always better than the human eye at recognising patterns in data, we decided it would be very useful for them and other marketing effectiveness teams.