the nuts and bolts behind google’s data-driven attribution modelling
Updated: Jul 23, 2019
By now, everyone in the PPC industry will have heard about Data-Driven Attribution (DDA) modelling. Perhaps you saw Google boast about the new opportunities they’re offering to identify undervalued keywords. Maybe it was an attempt to improve efficiency by adjusting the distribution of your advertising spend due to a re-attribution of conversions. Or maybe a client asked why they can’t access Google’s latest shiny analysis tools. But despite DDA being around for a relatively long time, there are still very few people who actually understand the mathematical principles behind how it works.
So dust off your old maths textbooks and sharpen your pencils, you’re about to get a peek into the mechanics behind this powerful attribution tool…
The Shapley Value is used in Game Theory, specifically designed to understand the contribution each player makes in co-operative games where everyone works towards a common goal. It assesses each player’s contribution, considering every possible variation of the player combinations including the order in which they join the ‘game’, and determining how they affect the co-operative total score on average.
Let’s consider an example.
Ailsa, Bethany and Charlotte are all writing poems, if they individually write for an hour, Ailsa can write six, Bethany can write 11 and Charlotte can write eight. However, if they work together in various combinations, they will creatively inspire each other to write more. The following combinations therefore would produce these quantities:
However, when all are working together, how many of these 29 poems can be attributed to each writer? This is where the Shapley Value comes in.
Considering the permutations of the three writers, we can see six arrangements exist: ABC, ACB, BAC, BCA, CAB and CBA. To calculate the contribution made by Ailsa to a given permutation, we consider the number of poems produced by the group as she joins it (the number of poems produced by the group consisting of only the letters up to and including the A). Subtracting this from the number of poems the group produced before she joined it (the group formed by considering all the letters before the A). This gives us the increase in the number of poems written that adding Ailsa to the group caused. Simply adding these values for each permutation and dividing by six, it’s possible to work out the total number of permutations for the three .
Here we see the sum is 6+6+9+5+9+5 = 40, and dividing by 6 gives 6.67. So we attribute 6.67 poems to Ailsa. By the same method, Bethany is attributed 12.67 poems and Charlotte 9.67. In total, all 29 poems are accounted for and they are attributed unequally based on who brings the most value to the group.
so how does this affect my conversions?
In the same way the poems were based on the merit of each writer within the group, the DDA model attributes conversions to different touchpoints based on merit, or more precisely, the impact each touchpoint had on a conversion happening.
If you switch to a DDA Model you might find that a touchpoint drives very few conversions on its own, but as part of a multi-touchpoint strategy, it is responsible for a lot of the conversions in that pathway.
why does dda modelling matter to my clients?
Clients might point to low converting campaigns and start asking questions on performance and spend. It’s difficult to explain the user journey and the invisible influence of multi-touchpoint campaigns without any hard data to back up your claims. With DDA Modelling, you are equipped to prove the impact of your campaigns behind the scenes and as part of your wider advertising strategy. It’s a necessity to understand which moments matter. Using DDA Modelling, we can determine how much credit to assign each click within the user journey making it a key tool and one worth utilising as part of a multi-touchpoint strategy.
Google’s model opens a new realm, enabling you to prove the impact of campaigns through analysis of performance and spend, in turn, pushing clients towards the desired goal of better results driven and backed by hard data. If you would like to find out more about data-driven attribution modelling and how it can help your business, contact the equation team.