Accurately attributing sales to your marketing activities is one of the most important – and difficult – jobs of a Digital Marketer. There are so many paths a customer could take to reach your purchase page, and not all of them are easy to measure.
Even if you just focus on the digital channels (because they’re easy to measure), the raw numbers don’t always tell the whole story. Sometimes you have to dig a little deeper to get to the truth.
Take one of our clients for example – a high-end, one-brand, designer furniture retailer. As you might expect, their website isn’t selling products one buys on impulse. Instead, conversion paths are extremely long, with over half of all traffic coming from Paths with 6 steps or more.
Now, these guys had a pretty young account and it was growing rapidly. To make sure they invested their money in the right places, they needed us to assess the optimal spend for different channels. What we found, was that the performance reported on Non-Brand Text Ads (AKA “Generic Search”) seemed extremely poor at face value.
That seemed a bit strange to us, so we assessed the conversion paths and tested different attribution models in the Analytics Attribution Modelling tool.
Mapping the Attribution Model
The current “Last Non-Direct Click” model suggested that Generic Search had a ROAS of just 0.85, while Google Shopping had a ROAS of 2.87. Which makes it look as though Generic Search is doing really poorly.
However, when we looked into the overall Revenue contribution for Shopping and Generic, we found that the number of paths involving Generic queries that resulted in Conversions, and those of Shopping, were actually already proportional to their current spend.
Shopping had spent €13K, and was involved in paths totaling €72K in revenue. Generic search had spent just €6K, and had €29K revenue. Both of these are a ratio of about 5:1, though they do not represent ROAS.
The role of Generic Search
The message was clear that Generics did contribute significantly to conversions, but according to the current Attribution model, this wasn’t coming through.
So what is the role of Generic Search for this client?
We divided the paths into three groups – Generic as introducer (the first click in the path), Generic as a converter (the last click), and “Helpers” – anywhere in-between.
As it turned out, 78% of Conversion paths involved Generic Search as the first click. That meant, our current attribution Model was only representing between 12-22% of conversions.
So, of course, the ROAS looked poor because that is not the role of Generic Search for this client. Also, the revenue share of the “Converters” is significantly lower than its share of conversions. This suggests that where Generics is the last click, this is because they bought something relatively cheap, which means not only did the current Attribution Model misrepresent the volume of conversions that are due to Generic Search, but also the buying behavior and the ABV of this Channel.
Find the model that works for you
So, what should we do with this information?
Google Analytics has a useful tool for testing out different Attribution models and the results they give. Though the data is somewhat limited, it helps to give you an idea of how different models affect different channels.
One model which shows your Brand Campaigns in a better light will affect Channels which typically appear earlier in the journey more negatively, and vice-versa. All in all, it depends on the impetus of the Company and their aims.
In our example, we still consider the Converting click to be the most important, but we want to recognize the significance of the introducer. You can make your own models based on your own needs, so we started testing a 30%-10%-60% split.
It is also important to note, that a more representative Attribution Model, balancing the needs and purposes of your different channels, allows for much more than distributing budget and assigning revenue to channels. It also allows you to make more informed decisions concerning bids, adding KeyWords, negatives, Location targeting, Scheduling, audiences and more.
As a final note, for small accounts, there may not always be enough data to make informed decisions, and Attribution models that share a conversion over multiple paths can help with this. Though it may seem odd for a KeyWord, location, time, or audience to have 2.6 conversions, if that number would have been 0 from another model, there would be no data here on which to judge it.