title
 

Attribution modelling – what’s it all about?

, Posted by Nick Comber in Nick's posts, Web Analytics

Nick Comber, Senior Technical Analytics Executive at Mezzo Labs, explores attribution modelling: what it is, why it’s important, and how to utilise it to inform your marketing campaigns.

What is attribution?

One of the fundamental facets of online marketing is understanding what is successful and what is not. If you run a marketing campaign and don’t measure performance, it’s hard to know where to look for improvements. 

This is where attribution comes in:  simply put, attribution is the practice of assigning credit for a conversion to one or more channel or marketing campaign, allowing us to understand what is driving the most revenue. But attribution isn’t just transactional – in Google Analytics, for example, you can attribute on any kind of conversion or configured on-site goal.

Before entering the world of analytics, I spent a number of years working within the affiliate marketing industry, where attribution is a hot topic as different attribution models directly relate to how much a publisher earns.

For example, a customer considering a purchase on a retail website might, prior to transacting, perform a quick search for a voucher code relating to the purchase. If so, the voucher code publisher would earn commission if the retailer is attributing based on last click, but not if the retailer was using a different attribution model.  So different attribution models can be truly game-changing where certain businesses are concerned.

Attribution Models

There are 5 main attribution models, all of which are compatible with Google Analytics.

  1. Last Interaction

What is it?

  • The most straightforward attribution method available,
  • Attributes conversions to the channel from which the user clicked most recently prior to converting.

Pros

  • Easy to understand. No need to consider weighting / user behaviour – you know straight away that these are the channels that refer a user directly before conversion.
  • Ideal for processes with a short lead time between first visit and sale, as the last interaction is most likely to lead to a user converting.

 Cons

  • Looks at conversions in a very monoline way. Only recognizes and attributes based on last interaction prior to purchase.
  • Difficult to attribute Brand activities, or those activities not primarily designed to drive sales.

 

  1. First Interaction

 What is it?

  • Diametrical opposite of the last interaction model; attributes 100% of conversions to the first channel from which the user clicked to visit the site
  • Ignores subsequent channels visited before converting.

Pros

  • Easy to understand, involves no complicated calculation, simple to analyse.
  • Better suited to general awareness/brand activity rather than sales activity per se.

Cons

  • No true ROI picture for your marketing campaigns. It’s very common for customers to be exposed to a series of campaigns after their first visit, of which one or more will eventually encourage them to go ahead and convert – first interaction attribution ignores all of these further efforts, which may in fact be more likely to drive the customer to convert.

 

  1. Linear

 What is it?

  • Multi-touch attribution model; gives every touch-point on route to conversion an equal amount of credit.

Pros

  • Every channel that “contributed” towards a customer converting is now credited for doing so – there is no risk of important parts of a customer’s conversion journey being ignored.
  • This is the simplest multi-touch attribution model, it should still be easy to understand and work with the data it returns.

Cons

  • Since a customer is influenced more by some channels than others during a purchase journey, depending on several factors, rewarding everything equally probably isn’t going to give you the most accurate picture of how successful each of your channels are. It is likely to over-represent some channels whilst under-representing others.

 

  1. Time Decay

 What is it?

  • Slightly more advanced model than linear in that it tries to get around the issue of some touchpoints being more influential upon customers than others by attributing a proportionately high percentage of a conversion to the most recent channel a customer visited, with a sequentially lower amount attributed for touchpoints that occurred further in the past.

Pros

  • Still a multi-attribution model and so all touchpoints on the way to a customer converting will be considered, with nothing being ignored completely.In theory, touchpoints that are closer to the time a user converts will be more influential in instigating the conversion itself, so this should produce data that is closer to an accurate picture of attribution than the linear model is able to.

Cons

  • It may not necessarily be accurate to assume that the last touchpoint has the greatest influence over the customer converting. Going back to my voucher code example – a user that gets all the way to the checkout page of a site before searching for a voucher code is probably going to convert whether they find one or not, so it can’t be fair that we attribute more of the conversion to the voucher code site than any other channel.

 

  1. Position-based

 What is it?

  • A typical position-based attribution model will assign the most credit to both the first and last touchpoint before conversion, as these are both widely regarded to be the most influential in driving a customer to convert, before evening assigned a smaller amount of credit amongst everything in between. A typical implementation of position-based attribution will assign 40% credit to both first and last touchpoint, with the remaining 20% spread across everything in between.

Pros

  • Closer still to the ideal of “true” attribution:First touchpoint is important in creating initial brand interest for a customer, with last touchpoint a crucial factor in driving the actual conversion. Everything else is just a contributory factor and should be accordingly rewarded less.

 Cons

  • Some of those “middle” touchpoints can actually be very important in driving a conversion.
  • Position based attribution can get very confusing very quickly, and it will take some analysis to decipher solid reasoning behind the data you see.

 

Google Analytics Model Comparison Tool

So, which one of these attribution models to use? The answer, quite simply, is that you should choose an attribution model that best represents/rewards channels that drive traffic/conversions on your site in the best and fairest way.

Those using Google Analytics you employ GA’s Model Comparison Tool to help make decisions. This will enable you to easily compare and contrast your conversion data across all of Google’s preset models, which will go some way to help you understand what is producing the most relevant and informative data.

Summary 

Today’s conversion attribution is far more complex than in years gone by – the idea of attributing everything to last click no matter what is outdated and in many cases no longer relevant. Whilst some still attribute and decision given the sheer amount of different attribution models you are now able to easily implement.