Just type away and hit enter


Why everybody’s doing attribution analysis wrong hero image (1)

Why everybody’s doing attribution analysis wrong

***Updated June 2016***

Ahead of his talk about maximising ad spend ROI at next month’s Benchmark Search Conference, our Head of Paid Search, Dave Karellen, examines the flaws in the existing attribution models for conversion and offers solutions for analysing the data available more accurately

Attribution evolution

John Wannamaker’s oft-quoted “Half the money I spend on advertising is wasted; the trouble is, I don’t know which half” at first glance seems a strange concept in today’s world of digital analytics, where every penny is scrutinised. However, while we are now spoilt with more data than most know what to do with; our eyes are opening more and more to the limitations of measurement in the digital world.

John Wanamaker quote

A long way from the perfect attribution model

The major restrictions are around cross-device attribution and also any offline influencers on a path to conversion. Google realises this as one of the biggest flaws in its system and is ploughing a lot of resource into bridging the gaps we can see in a user’s journey. They have very recently started providing cross-device attribution reports to top agencies that give more data on the uplift provided by mobile traffic. Understanding these factors is crucial, so that you don’t undervalue certain devices, or underestimate offline activity and word of mouth. The big issue though, is that even taking these problems aside, there is no unified solution on how to correctly assign value to each channel step in a path to conversion even if it takes place solely online and on a single device.

Google has already gone some way to helping marketers look beyond the default ‘last click’ model. There are now 7 standard models available in Analytics, and one extra custom model. While this is definitely a move in the right direction, we are still a long way from a perfect model.

Googles standard attribution models

Google Analytics’ 7 conversion attribution models

The fact that Google seemingly can’t decide exactly which is best, and so offers all these different models speaks volumes. Google will happily admit that each has pros and cons, and so the onus is on the marketer to decide which one most adequately meets their business objectives. But does a business itself really lend itself more to a position based model as opposed to a linear model? Often you’ll find a marketer’s decision is down to personal preference rather than based on business objectives.

The 8th option, the custom model, started off quite basic with only minor changes from the standard models allowed. It has since progressed to allow for additional weighting to be given to certain channels or if the channel has driven certain engagement metrics. This is significant in that it’s a step away from the de facto dogma that the only factor in assigning attribution is where it sits in the path to conversion. A major flaw is that the custom model does not allow you to differentiate between a 2nd or 4th interaction in a particularly long path to conversion.

At Click Consult, we’ve built a tool that is able to counteract this oversight and works outside of Google Analytics. This allows us to assign attribution value to each step with complete flexibility. One example that we see works well is a mix between the Time Decay and Position Based model wherein the first and last clicks get the most attribution value, and then this progressively scales down as you get towards the middle. Here’s how we imagine the logo would look if such a model were already available in Google Analytics alongside the 7 pictured above:

click example attribution model

We should stop focusing on the fundamentally flawed ‘position in conversion path’ models and find a novel way of addressing the situation

 Uncovering the true path to conversion

One point that sticks out like a sore thumb is that there seems to be an awful lot of preoccupation with where a channel sits in a path to conversion. While definitely a key factor, is it really the be all and end all of attribution modelling?

There are the beginnings of movements within the industry to suggest we should stop focusing on the fundamentally flawed ‘position in conversion path’ models and find a novel way of addressing the situation. Google is working on a feature that analyses similar paths to determine the uplift given when a certain channel is or isn’t present. The example below shows how we can assume that the presence of the social channel has resulted in an extra percentage point in conversion rate compared to when it isn’t part of the path.

click example attribution model

While the possibility of this kind of attribution analysis has been on the radar of marketers for some time, it would require a lot of data to be able to have statistically significant information on which to base decisions. This would only become exponentially more complex if a large number of different channels were used, and to drill down to individual campaigns or even keywords would be near impossible. So while this seems like it might be the Holy Grail of attribution, in practice we should be treating this with a fair degree of scepticism.

What we really need is a way to look at channel performance beyond its position on a path to conversion AND that can be used for sites that don’t necessarily have over a million visitors a day. One solution that we’ve come up with Click is to look outside the box in terms of just considering paths that lead to a conversion. Instead we also examine paths that did not lead to a conversion, and where users dropped off. By considering a ratio of when Channel A led to another interaction, compared to when it was the last interaction in a chain, it is possible to see how effective a given channel was at contributing to retaining interest.

To use a football analogy, to what extent can you really blame a midfielder if he consistently passes to a striker, but the striker always fails to score? What we are effectively suggesting is a point system whereby more attribution value is assigned on what is the marketing equivalent of a high pass completion rate. This method prevents all the middle interactions being treated as the same.


A factor in our attribution model is to assign value to a touchpoint based on its ability to successfully ‘pass’ to other touchpoints. Image source: http://www.mjfolio.com

This goes much further beyond the user engagement metrics you can currently use to add more value in Google Analytics, which currently are only ‘time on site’ and ‘page depth’. Obviously, this shouldn’t be the only factor in allocating attribution value. There should still be a large reward for the channels that are stimulating interest in the first sense, and those which are ultimately able to close the conversion. The main benefit of this is that it is scalable to very long and complex paths for larger data sets. It can also easily be layered with a position model, with some channels naturally being valued higher than others; more attribution value can also be applied if the ‘pass’ did actually result in a conversion too.

What we are effectively suggesting is a point system whereby more attribution value is assigned on what is the marketing equivalent of a high pass completion rate

A fresh perspective on attribution strategy

Of course, this doesn’t solve the issues with cross-device attribution and offline activity. We really have to wait for Google to help out here with true multi-connected measurement for the multi-connected world. What we feel is the most important consideration for marketers wanting to provide an attribution solution today, is that we mustn’t be entrenched in the established ‘position in path’ is the sole factor in attribution. The part a channel plays in influencing a user’s journey continues is clearly another important piece of the puzzle that until now has been largely overlooked.

While the ultimate goal of attribution is to be able to accurately assign varying targets on each channel, it is important to note that attribution analysis can often provide secondary insights that are just as critical to overall performance. If a channel is failing to either convert or stimulate further interactions, then yes, we should be assigning less attribution value. But moreover, we should be looking at why this is the case. The content and ad copy may vary greatly between this channel and others. If so, then it is important to address this disparity to ensure there is a seamless experience between channels.

The attribution riddle is far from solved, but it is clear that the ideal attribution model is one that looks far beyond a channel’s position in the conversion path, and it seems that a channel’s ability to stimulate further interactions should be an important factor moving forward.

Our one-day Benchmark Search Conference on 12th July 2016 at Manchester’s Bridgewater Hall, featuring industry experts from Google, Microsoft, Late Rooms and AO.com. 

Last year’s event was awesome, but we’re confident this year will be even better. It’s free to attend, so sign-up today!

Share this:

View all posts
View all Resources