This series of blog posts is a written version of a presentation I gave, in one form or another, at Hero Conf 2017, Superweek 2018, SMX Munich 2018 and DA Summit Greece 2018. As the deck and talk was slightly modified each time, the deck included here is a consolidated version of those talks, as is this written description. It became quite a lengthy description so splitting into three parts.
Given the dates of those talks and the dates of these blog posts, quite a period of time has passed and hopefully my memory of what I said is still fairly accurate. During that time, LeapThree (branding on the slides) was acquired by Ayima (in mid 2018) and I moved on from Ayima in mid 2020.
Origin of this Presentation (slides 3 to 10)
An Exchange on Twitter
The trigger for this presentation was a tweet announcing a blog post on attribution from @PPCKirk. Attribution is a word that always gets my attention so I had a read and disagreed with an example used within the post, sharing this disagreement on twitter (politely). I also mentioned as an aside that the logic described for Google Analytics was wrong.
Further explanation was required and it was amusing/depressing to learn I was the first person to point this out to Kirk. Google Analytics is used by so many companies and people around the world and nearly every user doesn’t know key definitions that can really impact the data they are looking at (understandably as there are so many definitions). As we were talking attribution, he found my old deck on attribution and kindly gave positive feedback. This positive feedback was extended to pushing me to pitch a updated talk on attribution to Hero Conf for later that year. This deck was the result.
I provided Kirk with my typical disclaimer on attribution and repeating it here for you – I dislike & disagree with attribution. This is not a talk/blog post on how to do attribution. It is a talk/blog post on why you should not focus on finding the right attribution model for you, instead investing your time/money on a different approach.
A Final Caveat
The other caveat is that I frustratingly haven’t done a heap of practical work on attribution models or alternatives. I am coming at this from a theoretical viewpoint, not a practical one. That was true at the time of these talks and is still true now.
Defining the Business Problem (slides 11 – 14)
An early request from me in client workshops to kick off engagements is “What do you need to know”. Common questions related to marketing include:
- How is my performing?
- How should I split revenue between my marketing sources?
- What is the right attribution model for me?
These questions frustrate me as they are the wrong questions. They focus on past behaviour and carry the assumption that attribution is the only approach to use, if only the right attribution model could be identified.
I wish for questions focused on the future instead, as that is what really matters. Questions like “How do I optimise the allocation of future marketing spend?”. A blunter version of that question is:
How can I best spend my marketing budget to make the most money?
People need to think in terms of how they can get the best results from marketing and not to put their time into trying to pick an attribution tool and/or model.
Understanding Attribution Models via an Analogy (slides 15 – 26)
Last Touch Attribution (slide 17/18)
I used to work for Logan Tod (since acquired by PwC) and the founder Matthew Tod had a simple analogy/story to explain Last Touch attribution (note this was 10+ years ago). A famous English football (soccer) striker Gary Lineker was asked how he was so good & successful. The answer was he just stood up near the goals and knocked it in when his teammates passed the ball to him, taking all the credit for these goals. Implicit in that is the teammates actually do the hard work but get none of the credit.
For Matthew, this was the perfect analogy for Last Touch attribution, that the traffic source where the sale or conversion happened receives all the credit for that conversion. All previous sessions and the traffic sources that did the hard work, driving those sessions to the website, receive zero credit.
The same setting, that of a goal being scored in a football match, can be used to illustrate the logic of other common attribution models as seen in the next few slides.
First Touch Attribution (slide 19/20)
A midfielder initially gets the ball, passes to the winger who crosses it to the striker (note that as an Australian, football/soccer is not my native sport so apologies for any dodgy descriptions of play) who, as previous, scores the goal. If the midfielder received all the credit for the goal, as the player who first touched the ball, this would be equivalent to First Touch attribution.
Note that in both first and last touch attribution models, a single touch point or marketing campaign receives all the credit for each conversion.
Weighted Attribution (slide 21/22)
But there were three players involved and they all contributed to the goal, therefore it seems reasonable they should all get partial credit. How that credit is apportioned though, well that is the challenging part. You could assign credit based on how near the players were to the original player or to the player who scored the goal. You could take into account the time between the player touching the ball and the goal being scored.
Each player will have a bias in how they think credit for goals should be assigned, based on the position they play. Strikers would argue that it should be weighted towards the players at the end of the passage of play. Defenders would disagree, believing the reverse, that players who started the play should get more credit.
It can be the same for marketing channel owners who disagree on the calculation within a weighted attribution model. They want to maximise the credit their channel receives and define the calculation accordingly.
Data Driven Attribution (slide 23/24/25)
What if that all sounds too subjective? We are in a modern world now with Big Data, Machine Learning and Artificial Intelligence. Can’t the machines take over and do all the thinking for us? Yes, in fact they can and that is the point of Data Driven attribution.
Imagine that football passage of play was repeated multiple times, to see if the goal was scored every time. And then repeated more times with different players involved, to see how that impacted the situation. In fact, what if the whole game was replayed, thousands of times, with different combinations of players and in different weather conditions on different grounds. Then you could see that more goals were scored when certain players were involved and that those players should get more credit for goals.
The machines can tell you exactly how much each player contributed to all the goals scored and games won, based on all the data collected. Statistics (and this amazing football game simulator) does all the work and gives you the exact right answer. And this is what Data Driven attribution tools do for you. They take all the data on all the touchpoints for all the users & sessions, working out the contribution of each marketing channel to the conversions that happen.
Choosing the Right Attribution Model (slide 26)
So, back to that common question of which is the right attribution model for you.
Hopefully, everyone can agree that giving all the credit for a goal to a single player, ignoring all the other players on the pitch, is wrong. So, you shouldn’t rely on last touch or first touch attribution model.
The subjective nature of Weighted attribution models, where you can adjust the logic to get the results you wanted in the first place, should automatically disqualify this approach. In addition, having a single set of logic that is applied for all conversions just can’t be applicable for all conversions.
Does that mean Data Driven attribution, with its use of machine learning and statistics, is the answer?
Return for part two where I answer that question (spoiler: Data Driven attribution is flawed too) and part three where I provide an alternative approach to optimising your marketing efforts.
