Like adding a potted plant to your bedroom, let’s heat things up in here by talking about something really sexy… Attribution Models, oh baby. Let’s get started.
It’s all about what we do. In ‘Human Behaviour’, the opening track to Bjork’s Debut Album, she beautifully harmonises that:
If you ever get close to a human
And human behaviour,
Be ready, be ready to get confused.
Bjork’s right. Our actions are incredibly strange and elusive even to ourselves. I use the word ‘our’ as I assume you’re one of those peculiar humans like me. If not, I’d like to welcome our new alien overlords.
Behaviour has been studied for over a thousand years. Classically approached from boffins often quoted (but never read) like Aristotle or Colonel Sanders himself; Sigmund Freud. They observed then critiqued our illogical twitches to find the universal truth that binds us all. But we’re not here for that.
We want to know what made you buy that caterpillar draught excluder…
Digitally, we can track where you came from before you reached the product, but does that paint an exact picture of the deciding action of why you purchased? Arguably, I would say no. And, you don’t have to study your customer’s physiological state to figure this out. (God forbid)
There are a plethora of attribution models that you can use for a more accurate framework to how goals are achieved from start to finish. Even better for us, ‘self-proclaimed Data Analysts’ (aka graduates with degrees from former polytechnics *cough*) is that we can also compare these models against each other and we can even get machines to do it for us. So we can spend more time doing the things we like, like spending time with friends and family.
Actually, I will personally fight the machines before I have to interact with people.
What are Attribution Models?
An Attribution Model is simply about dividing credit or attributing merit between touch points towards what you set as a goal, KPI or a conversion. There’s not much else to say about that. That is exactly what it is; it’s about adding more value to the journey. The most popular Attribution Model is the last click, last touch or last interaction method, which is deeply flawed.
What’s Wrong with the Last Click or Single Source Attribution?
Traditionally, the last click attribution method is used. For instance:
- You are an online retailer that sells Blu-rays & DVDs. (VHS on the side)
- Six customers this week bought six copies of Kieslowski’s The Dekalog. (It’s the most excellent TV-series ever made, why wouldn’t you buy it?)
- They each came from a different source: organic, pay-per-click, a different domain, social media, email and direct.
- You would then award whichever source 100% of the credit towards that conversion.
Is this right? Should the last man to cross the finish line be the one to receive all the credit? All your marketing sources should be a relay race team, with your customers being the baton that is passed onto each racer. Each is receiving equal credit for their participation.
However, it’s even more complicated than my relay race analogy. If the customer journey was a relay race, then it would be absolute chaos. Some racers wouldn’t turn up, others would decide to go off track then return after 30 days. Some would take part in another race and some racers may trip and fall. I’ll stop with all these metaphors (although I was trying to think of a way to fit in performance-enhancing drugs and how that was like PPC or paid social).
Where was I? Oh yeah, it all goes back to Bjork, i.e. human behaviour.
We’re very hard to understand as our behaviour is not black and white. Hence why our action cannot be categorised so easily. Hence why we need various models of measurement. Hence why I’m writing this blog. Hence why I never use the word ‘hence’ as I tend to use it at the start of each sentence.
What are the other Attribution Models?
Like most things in Digital Marketing, some clever bod has already done the thinking-work. If you’re using Google Analytics, there are many of models available to you.
Here I’ll explain in simple terms what each model means, using famous 90’s footballers as examples. (Just kidding)
You can find Google’s official explanation here. But, if you’re keen to read my less boring explanations then follow me.
Briefly explained above, this where you give 100% of the credit to the previous channel interacted with for a goal completion. Most folks will be reporting on this method. It’s as dull as Weetabix, but successful and straightforward. If you don’t have a consideration phase in your buyer journey, then this is probably the most correct measurement for you.
Last non-direct click doesn’t take direct as an answer, and attributes the credit to the last known channel.
In Google Analytics, direct means it doesn’t know where the user has come from. This is a constant problem for most marketers, an upturn in direct traffic in that month’s report means your shoulders will be sore from shrugging. I’d suggest if you have a significant amount of direct traffic then I’d look at assessing where this could potentially come from.
For instance, you could have multiple campaigns that don’t have proper tracking within the links, and this could be via email or in an ad. In the meantime, last direct click can save you. It’s not entirely accurate but it’s useful for plotting what resources work in your journey right now.
The Last Adword Click
This model is all about giving all your goal-scoring prowess to all the adverts. Not much to explain here. Last Adwords Click is deaf to anything that’s not an ad and will give credit to Adwords. Obviously, not the one to use if you don’t have any ads.
The First Interaction Model
We all remember our first time – in the buyer’s journey, of course. [Note to Editor: Add winking emoji.] Wherever your customer first popped up, that’s where you give the attribution to. This is a fascinating way of thinking. As you’ll be able to see to scale where users first know about your brand or site. Seeing how many channels they go through before validating a purchase or contact. If the process is pretty long, then you’ll need to work on your first impressions to tighten that first meet.
A fair and equal model, Linear spreads the credit equally over every channel before reaching a goal. This is one of the more accurate models and highlights every step in the journey. This is kind of hard to explain when it comes to reporting and I’d say only use if you’re committed to deep analysis.
Time Decay Model
This is a good one for goals with a quick turn around. It attributes all the credit to closest to the time when the conversion was made.
Position Based Model
Position based modelling give a specific set of credit to the first and last click, then shares the rest of the credit with everything in between.
Reporting and Comparing Attribution Models
All these attribution models are available to you right now. You can see how they work without affecting any of the data within GA. It’s well worth looking at these and expanding your mind when it comes to goal conversions. As you’ll see, it’s not as straight-forward as you think.
You can compare each model against each other in the model comparison tool. Some of the many reasons to do this are to:
- See if there’s any under-nourished channels or campaigns within your goal path that appears to be a solid touchpoint. It may help to identify areas to improve.
- Properly value and prioritise sources/channels and its effects on neighbouring channels.
- Spot better ways of future reporting of conversion paths.
- Build a full proof projective customer journey.
- Align your own values with the proper channels.
Creating a Bespoke Attribution Model
Within Google Analytics, you can create your very own attribution model. Instead of using the standard template you can truly create a framework that properly fits your business. However, this is extremely complicated and not for the faint-hearted as there are so many variables to consider it might become overwhelming.
And, that’s why we make machines think for us.
Using Machine Learning to Create Attribution Models
Despite standard Attribution Models being very logical, they’re not very intuitive. Within Google Analytics, there’s no real way of using these models for statistical prediction or forecasting. This is where machine learning can come into play.
This is a bit of a tease, as I’ll be writing a big piece on how to use machine learning to not only create a fully customisable model but also a predictive framework.