A Mathematical Approach To Attribution

July 19, 2019

The digital marketing space is no stranger to jargon and buzzwords. There’s a new hot topic every couple of months. There was “single source of truth”, “holistic customer journey”, “end to end visibility”, and other words like transparency, full service and full funnel are seemingly thrown about as filler words.

We’ve all heard of attribution, but what the hell is it? What does it actually even mean for me and my business or clients?

A Brief Overview of Attribution

In my own words, attribution is the process of assigning (or attributing) value to a particular channel, touchpoint or object depending on how much it contributed to the end outcome.

Basketball is a 5 on 5 game. Lebron James currently leads this NBA season in assists, and averages 25.7 points per game. But his contribution to the team is more than just his points. His 10.9 assists per game means that at least 22 additional points come from him, but what about the other 4 players on the court? His stats change depending on who’s playing and who he’s playing against, and a whole host of other factors both on and off the court.

It’s never easy to determine how much each player contributes, nor is it easy to predict how a player will perform, but when it comes to digital media, there’s a model that comes pretty damn close to solving the conundrum.

Markov Chains

If you’ve never heard of Markov Chains before, most people haven't, I only learned about it through my actuaries degree at university.

There’s a thesis’ worth of reading to go into depth about Markov Chains and what they are, but in summary, Markov Chains are a state dependent model, where each channel or state that someone moves through is assigned a value based on its probability of contributing to the outcome we want.

So in digital, a simple example would be user flow between different channels.

Example Markov Chai

Let’s pretend State 1 is “YouTube”, State 2 is “Facebook” and State 3 is “Organic Search”, and the interaction between the states are how our users are navigating through the different channels. Some users may interact only once with Facebook, some may interact 3 times with YouTube and then go on to organic search. There are only 2 possible outcomes at the end; a conversion, and a non conversion.

So with enough data, we will end up with a long array of data, of how each user navigates through our channels and if they convert or not. It will look something like this;

  • User 1: Facebook > Organic Search > YouTube > Organic Search > Conversion
  • User 2: YouTube > Organic Search > Organic Search > Conversion
  • User 3: Organic Search > YouTube > Organic Search > Conversion

If you looked at that, you could see that FB and YouTube contributed to every single conversion, but if you reported from Google Analytics you’d see that only organic search drives conversions, but you know that’s not true.

Markov Chains now uses all of the above data and calculates the probability that each channel contributes to a final conversion that takes into account the sequence of events. The equation for how this works is below.

Don’t be shy.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.