Guide to Attribution Modelling
As a full-service digital agency we always have a variety of marketing channels to plan around, which always brings up the question of how to properly assign costs and revenue in a multichannel journey.
So we put together this guide to attribution modelling, with a few helpful tips on how to best utilise the wealth of user data that sit in your web analytics and mold an attribution model around your unique customers which I also presented at the MeasureFest analytics conference:
Step 1: Understanding user behaviour
How do users tend to browse your site? Depending on whether the answer is frequently, sparsely, regularly or something else, creating plots of visits by time lag, session count, duration, pages/session and studying any emerging patterns can give a more precise idea of which base model is suitable.
Step 2: Identifying Conversion trends
Much as we marketers wish to, not all traffic results in a sale. And the traffic that does is most likely different to regular site visits. A study of conversion rates against significant traffic signals from step 1 can reveal which visits are most likely to convert and which are wasting marketing budget.
Segmenting converting and non-converting traffic can also uncover user clusters who react differently to marketing activities and should be treated uniquely in attribution crediting and campaign planning.
Step 3: Life in a MultiChannel World
It is almost impossible to only be touched by a single marketing channel when interacting with an online brand. Even doing a simple online search exposes users to two channels, whether they realise or not. So how do your Marketing Channels behave? Clickstream data can show whether they work in unison or independently, have common sequences and if they act as Converters or Assists.
The basic type of Attribution is channel-based and therefore it is important to up-credit the Assist channels such as Email or Display and those which promote conversion-focused activity as in previous findings while dis-crediting those that make little effort to get the sale of a mind-already-made-up kind of user.
Step 4: Putting it all together
We won’t lie and say to you that Attribution Marketing is easy. It is a continuous process of re-examining, re-adjusting and re-defining a model in order to make it respond to the users, products and site as realistically as possible.
Studying the above signals and diving in the reach web analytics data can teach us a lot that is invaluable to a data-driven organisation, but human behaviour cannot be described by a model, a fair reason that deters most marketers from employing an Attribution model in their work.
But how about a mix of Attribution models instead? Business and seasonal cycles evolve, but in the short term they repeat. A divide-and-conquer approach to Attribution rather than a one-to-rule-all with a more focused intent can help guide budget decisions during sales periods, the holidays or other key events without trying to fit your complex marketing requirements around the nonexistent “median user”.
We Harvesters find it difficult to live with ourselves using Last Click and when there is a standard industry workflow of Collect – Transform – Analyse – Apply – Rinse & Repeat for web analytics that can be very efficiently used in Attribution Model to justify channel investments we don’t see an excuse to not do it.