Blog

A Quick Guide to Google Analytics Attribution Modelling

,

For as long as I can remember people have talked about attribution modelling with digital channels. With the blend of channels getting increasingly complex (before we even think about offline) it’s becoming more and more important for marketers (especially those in more strategic positions) to understand the role each channel in the plays in their marketing mix, so their budget can be assigned effectively, and ROI can be increased.

Google Analytics (GA) has now introduced attribution modelling as part of its standard features. This is yet another fantastic development for the tool, but as with a lot of these more complex reports in GA, it’s easy to spend a lot of time looking at it and not really gaining any insight into the data.

Here’s a quick start guide to give you an indication of the kind of things you can do with this tool, by no means is this a complete guide, but it’s a starting point. Not many of us have a full time data analyst on hand, but the following should give you some easy to follow ways to make the most of the new features.

The attribution reports are tucked under the Conversions tab on the main GA menu. Once you’ve opened that, click Attribution, and ‘Model Comparison tool’.

1

By default the tool shows you the last interaction model (see below). This is the model that GA runs as standard.

2

So to look at data you’ve not seen much of before, you can click where it says ‘Last Interaction’ and change the attribution model.

3

The list of models is pretty comprehensive, but what do they all mean, and how do you use them?

Attribution Models in Google Analytics

  • Last Interaction – This is the standard GA model. Good for ‘at a glance’ performance comparisons, and for short conversion/purchase cycles i.e. downloading a guide, or a low value ecommerce purchase.
  • Last non direct interaction – This is a useful tool to help identify mid – late channels in the pipeline. Direct often closes down a lot of conversions that have been generated by your other marketing channels so on the last interaction model it takes value away from those other channels, even though they have often done more of the actual ‘selling’.
  • First Interaction – You can use first interaction to the value of conversions to the click that starts the journey to conversion. This is useful to help measure the success of brand awareness campaigns, for example, display, and cold email activity.
  • Linear – The Linear model gives equal weight to each touch point. This is great for measuring activity over time, i.e. for long sales cycles/decision processes. It can help to assign fair value to nurturing programmes with lots of touch points.
  • Time decay – Time decay attribution works by reducing the value of a click, the further it is (in terms of time) away from the actual conversion. This is great for short term campaigns (especially if you’re supporting TV campaigns, or conferences for example), and sales or time limited offers.
  • Position – As standard, this gives more weight to the first, and last interaction in the conversion journey, arguably the most important parts. This is useful to help identify the channels that are opening and closing the journey for you.

So what can I really get out of this?

Here are a couple of examples of how I’ve used the tool already. It’s by no means a definitive list (in fact it’s barely even a list) but it gives an idea of the potential this tool offers.

Work out the real value of paid channels for budget allocation.

This has to be the most valuable use I’ve found so far. In the example below you can see a comparison between last interaction, and first interaction models for PPC for one of our clients:

4

You can see that on a last interaction model we saw 1905 conversions (sales in this case). Looking at the first interaction model however, we can see that paid search has attributed 2251 conversions – revealing that paid search is starting more journeys than it’s finishing.

This means the last interaction model doesn’t give us the full picture in terms of the value of PPC. This could mean we are deleting keywords that look unsuccessful on a last interaction model, but are actually starting lots of journeys to sale.

At a keyword level, the amount of detail goes through the roof, allowing you to understand in exactly if/how each individual keyword adds value to your campaign.

Affiliate last clicks

On the flip side, it can also devalue some activity. I won’t be the only person who has ever had a discussion about whether affiliates add value, or whether they just swoop in for the last click (and the commission).

Attribution modelling goes someway to clarifying the situation. You can use the first interaction (or a customised position based) model to devalue that last click and see whether your affiliate channel actually does anything other than last interactions. This is useful to help make sure your commission models are optimised.

Measuring Start-up or new product brand awareness

Another useful measure is to look at the first click model when running product/service launch campaigns. This can help you identify where your customers for these new products or services are finding out about you first, showing you where to increase activity in the future.

This is particularly relevant to start-up business, who don’t always have the luxury of lots of past website data to work with.

Limitations

As with anything, attribution modelling has limitations. The main one being the shear amount of data it produces (even at a basic level), which can make it difficult to manage and gain insight.
Each model has strengths and weaknesses too. You’ll need to experiment with the models that work best for you. This is going to be highly dependent on the type of business you are and the type of customers you deal with. A deep understanding of the customer journey will help work out where the data can add value for you.

The data available is fantastic, and it can and should be used to help you to optimise your marketing mix. To make the most of this though, I would suggest that you take small steps into the data, make small tweaks and understand the impacts. Consider the data a tool for optimisation, rather than Attribution modelling being a silver bullet which will revolutionise your entire marketing effort overnight.

More on this subject