Attribution Modelling: Why you’re doing it wrong

Medals

Once upon a time, I loved my Nokia 3210. I had never owned anything like it. Being able to text people? Instant connectivity on the move? Snake?! I got so used to (and dependent on) my precious little best friend that I felt that nothing could improve on it. It was everything I’d ever need from a phone. Fast forward a few years, and I’ve an iPhone in my hand. Enough said.

The same sort of thinking can often lead even smart marketers into a world of myopia, where what’s ‘worked’ before will always be enough, or at least will do just fine for now. One of these areas is attribution modelling.

A big problem with traditional attribution models such as last touch, first touch, linear, position based, and time decay is that they rely on a predetermined, subjective weighting being placed on a touchpoint. Additionally, these models often only consider one consumer path at a time, without any consideration for the performance of key variables in any of the other paths. Mostly the only paths that are considered are those that led to a conversion, without considering those that didn’t lead to a conversion. This is where short-sightedness kicks in.

Consider an example. 100 users interact with 3 advertisements (A, B & C) in various ways, but only one user converts.  That user’s path to conversion looks like this: A > B > C > Conversion.  Let’s apply some traditional models.

First Touch: A (100%), B (0%), C (0%)

Last Touch: A (0%), B (0%), C (100%)

Linear: A (33.3%), B (33.3%), C (33.3%)

Time Decay: A (10%), B (30%), C (60%)

A quick look at these and C is the clear winner, with A performing quite well, right?  As an astute marketer your instinct may be to redirect budget from B, to C and A proportionately.  But, you’re not seeing the whole picture.  You’re not considering the paths that weren’t successful.

Now suppose I told you that in the other 99 user’s journeys A featured in 99 of them, C was in 50 of them, whilst B was only in 5.  Do you still think C and A performed best?  Would you still move budget from B?

A and C were involved in so many users’ journeys not leading to conversion. Their hit rate is really low compared to B, but without considering all paths, you wouldn’t know this.

Tip: when considering platforms and providers of attribution solutions, find out what their methodology looks like. Do they use an algorithm that can consider millions of these data points at a time, and apply a data driven performance index to each touchpoint? Or are they reliant on subjective weighting and blinkered analysis? In other words, are they an iPhone of attribution modelling or a poor, old, Nokia? If they’re the latter, you need an upgrade.

Like what you see? Why not sign up for our newsletter to receive updates straight to your inbox!

Get the Newsletter