Revealing The Number 1 Unsolved Mystery Of Attribution
The marketing attribution dilemma that we face
In today’s world, an attribution software-as-a-service (SAAS) easily costs USD 5000 — USD 15000 per month. PER MONTH, mind you. And with good reasons too.
If I must choose, accurate attribution is THE one metric that triumphs above all other metrics. It tells you where your dollars are coming from, which translates directly to where should you focus your resources on.
But there is a catch. Attribution is not as straightforward as that. There are countless ways to do it and no one can firmly say one way is better than the other.
In fact, most companies do not even know what they are doing when it comes to attribution. Should I use rule-based or data-driven attribution? Should I use last-touch or first-touch attribution? Which marketing channel should I prioritize logically?
These questions have not been answered satisfactorily and holistically for decades. Hence, creating an attribution dilemma.
The attribution dilemma is arguably one of the most sought-after mysteries in marketing.
But before I go any further click on the link to read my Ultimate Guide to Attribution for Beginners first. If not, you might be spending the next few minutes trying to decipher some ancient Egyptian language.
Okay.
Going back to the dilemma, a football game makes the perfect analogy. When a team scores a goal, who is the one that contributes the most to the goal? Can we even single out a person to reap all the rewards from the combined efforts of the 11 people on the field?
The logical answer is “Of course not!”. They each played a crucial role in securing that goal.
Then the next question is, why is it that strikers are usually the ones that earn the most? A quick Google search for the highest paid football athletes shows you an entire list of strikers: Ronaldo, Messi, and Neymar. Does this mean that strikers are more important than the other 10 players?
We can dig deeper by asking ourselves: How do football clubs pay each individual player? Is it fair that clubs pay more to strikers than to goalkeepers? And let us not forget about the coaches. Although they were not physically involved in getting that goal, they played a HUGE part in it. How do we compensate coaches then?
Let me try to put it into better context but do keep in mind that I am no football expert.
A Football Game Analogy
In this game, Goalkeeper A passes the ball to B, who passes it to C, D and lastly E who scores for the team.
In football, is it possible that we skip B and A passes the ball directly to C? Probably. But what are the chances? Who knows? Is it also possible that we skip E, where D is the one that scores? Again, it is possible. But that is not what happened here. Our job is to not guess or estimate.
We want to break down the goal scientifically and mathematically. We want to know exactly how much each player contributed to the score. And that, ladies and gentlemen is precisely what attribution is.
In fact, football clubs have solved this huge dilemma more than 50 years ago.
The answer is the little-known Shapley Value.
So… What’s Shapley Value?
To give a brief context, the Shapley Value was devised by Nobel Laureate Lloyd Shapley way back in 1953. It is derived from a concept of cooperative game theory which has massive applications in attribution. This ingenious model fundamentally breaks down the marginal contribution of each player.
The modest Shapley Value looks something like this:
The formal definition as per Wikipedia:
Yes, I know, it seems a little intimidating, to say the least.
In marketing, Shapley Value helps us understand how each individual touchpoint or marketing channel contributes to the final sale.
Now let us put it into a marketing context:
Instead of scoring a goal, now we want to understand how each ‘player’ contributed to the customer’s conversion.
Our first touchpoint is pay-per-click (PPC). PPC leads to FB ads, followed by affiliates marketing, retargeting ads, and finally paid content on a famous blog before the customer buys our products or services.
Now there are multiple rule-based methods to attribute these 5 touchpoints. We have the first-touch, last-touch, linear, decay, position-based attributions. If they sound like some alien language to you, please read this right now! Otherwise there is really no point for you to proceed any further.
But Shapley Value goes beyond simple rule-based attribution. Now we are entering the data-driven approach to attribution.
We apply Shapley Value to identify exactly how we should attribute conversions to each touchpoint. It gives the fairest distribution of contributions. And this is by far a more sophisticated and holistic attribution technique compared to rule-based attribution.
We do not really have to go through the maths involved in Shapley Value. What we do need to know is the logic behind it. At its core, Shapley Value analyses multiple combinations of all our touchpoints and breaks down individually how much each of these touchpoints contribute to the sale. The keyword here is “Individually”.
Let me give you a real-life scenario. You have been visiting the cinema every day for the past week. And you always go with your best friend, Therese. Both of you will do the usual — walk into the cinema, buy popcorn, drinks, and pizza. You ALWAYS have a good time.
So, your happiness score is 10/10.
This week, the pizza is sold out. You and Therese ended up watching the movie without it. You were not as happy as you would be with pizza, so your happiness score is only 9.
Now we know the marginal contribution to your happiness of Pizza is 1 happy point.
The following week, Therese needs to attend her grandfather’s birthday party. So, you invited Adam to join you instead because watching a movie alone is simply not an option!
Turns out, watching a movie with Adam is less enjoyable. He does not laugh at the same jokes as you. Also, Adam talks a lot throughout the movie. Hence watching a movie with Adam only gives you a final happiness score of 7.5. This is even though everything else is equal, you enjoyed your usual popcorn, Coke, Hawaiian pizza, and your movie.
Now Shapely Value will tell you that the marginal contribution of Therese to your happiness is 2.5 (10–7.5).
But Shapley Value does not stop here. It iterates further.
Does Therese really bring you so much Joy? Or is it the fact that Adam was too boring?
If say you eventually see 1000 movies and each time you bring a different friend along. At the end of each movie, you wrote down your happiness score.
Now instead of just Adam, you have 1000 more variations. You sat down and put all these happiness scores into an Excel spreadsheet and compute the average of the marginal contribution that each of your friends bring.
You will then be left with a WAY more accurate attribution of happiness contributed by Therese.
And that, ladies, and gentlemen, is exactly what Shapley Value brings to the table.
Let us try to understand this in the previous marketing example.
One possible conversion combination is PPC — FB — Affiliates — Conversion. The conversion rate turns out to be 2%.
Another possible combination is PPC — FB — Affiliates — Retargeting ads — Conversion. The conversion rate improved by 0.7%.
From this we can mathematically calculate the marginal contribution of “Retargeting ads”, which is 0.7%.
But Shapley Value does not stop here. It iterates through ALL the different combinations and calculates the average of each touchpoint.
A few more possible combinations:
(Note that the original pathway is PPC — FB — Affiliates — Retargeting ads — Paid Content)
- PPC — FB — Retargeting ads — Paid Content. (Affiliates removed)
- PPC — Affiliates — Retargeting ads — Paid Content. (FB removed)
- PPC — Affiliates — FB — Retargeting ads — Paid Content. (FB and Affiliates swapped positions)
- PPC — Affiliates — FB — Paid Content — Retargeting ads. (FB, Affiliates, Paid Content, Retargeting ads swapped positions)
One thing to note is that Shapley Value does not only consider different combinations, but also the order of which the touchpoints occur (permutation).
Thoughts
Let’s recap.
The benefits that Shapley Value offer:
1. It fairly distributes the contribution for touchpoints of unequal contributions to the final conversion.
- In the football dilemma, it attributes the contribution of all players fairly, even though they are each equipped with a different skill set.
2. It is the only attribution model with a solid foundation. It covers the axioms of efficiency, symmetry, dummy, and additivity. This contrasts with the linear behaviour of rule-based attribution.
One major disadvantage is the sheer volume of computational effort required. In fact, most real-life scenario would require certain compromises to the original model. These compromises reduce the computation time but increase the variance of the value. In short, you will need to sacrifice quality of data for practicality, at least for now.
We at BeamDance firmly believe that Shapley Value triumphs all the rule-based attribution by leaps and bounds. It is not merely just another way of attribution; it is the only way.
Our team is working round the clock to put together the finishing touches to our very own attribution tool, BeamDance made accessible to the masses. BeamDance provides both rule-based and data-driven attribution.
More importantly, it is affordable. The status quo only grants access to attribution to companies with deep pockets or enormous marketing dollars. A SME will not be able to pay US$10,000 per month for an attribution tool, period.
If you are serious about understanding your customers’ journey fully, you can use the tool for FREE once we roll it out. Just reserve your spot as a free trial user for this revolutionary attribution tool.
Reserve a spot with us now and start understanding your customers’ journeys like never before.
See you at the product launch!
About the Author
As a data scientist, Admond Lee is on a mission to make data science accessible to everyone. He is helping companies and digital marketing agencies track and achieve marketing ROI with actionable insights through innovative attribution and data-driven approach.
His story and data science work have been featured by various publications, including KDnuggets, Medium, Tech in Asia, AI Time Journal and business magazines. Besides, he has been invited to speak at various workshops and meetups.
With his expertise in advanced social analytics and machine learning, Admond aims to bridge the gaps between digital marketing and data science.
Check out his website if you want to understand more about Admond’s story, data science services, and how he can help you in marketing space using data science.
You can connect with him on LinkedIn, Medium, Twitter, and Facebook.
