Quantifying the Bias Across Different Attribution Models| No Comments
As I mentioned in my first post on the importance of attribution, last-click is the most common attribution model among BloomReach customers. Online businesses tend to use because it’s the easiest model to use that doesn’t “double count” across channels. It’s the default attribution method available in most analytics software – implementing something other than last click is a lot of work. Last click also appears to do a reasonable job of attributing across channels. Most of our customers, however, recognize that last click is not necessarily the best attribution model. They understand that it is biased in favor of channels that tend to appear later in the buy cycle, such as coupon sites that often attract customers right before they buy and would have likely bought anyway. The question is how much? Since there wasn’t a lot of research on this point, we decided to look into it ourselves.
To do this, we looked at clickstream data across a selection of sites to calculate the percentage of time a channel was the first, middle or last click for clicks that led to conversions. We looked at the following channels: paid non-navigational search, organic non-navigational search, paid navigational search, organic navigational search and everything else (“referral”). As I’ve discussed in another blog post on the importance of analyzing navigational (or branded) and non-navigational traffic, it’s critical to split paid and organic into non-navigational and navigational because these are fundamentally different channels. Here’s what we saw:
What this shows is that referral and navigational paid and organic search traffic is more likely to be last click than non-navigational search traffic (both organic and paid). Last click attribution will, thus, favor these channels over non-navigational traffic. If online businesses were to move from last-click attribution to a methodology that attributed value to first and/or middle clicks, then they would attribute more value to non-navigational search channels than they are doing now. Our research showed that for many businesses this might increase the value of these channels by 10-15%. However, we also found that this number varied a lot by businesses. Some businesses would see a greater than 20% improvement in the value of these channels while for others the change was smaller.
Other research seems to corroborate this finding. For example,Havas research shows that on average, its multichannel attribution model re-allocates 50% of conversion credit to early funnel searches. If we had broken the “referral” channel in our analysis out into more distinct channels, it is likely that analysis would show greater diversity in the value allocated to different channels.
It’s clear from this analysis that it’s critical for online businesses to understand how their attribution model is biasing their current attribution of value. Otherwise, the business could be significantly under-investing in some channels and over-investing in others.
This brings us to one of the most important question for most marketers: How much should I invest in a particular marketing channel? The answer, as most would agree, is that the level of investment should be based on the value contributed by that channel. This means that a marketer should not only understand how the value of a channel is affected by their attribution methodology, but also consider what value a channel contributes even if it is not covered by their attribution methodology. In the next post, we focus on a particular channel to show how one might better understand its total value using this approach.
Our white paper, “Making Online Attribution Work: Three Steps to Better Business Decisions”, discusses the findings covered in this post in more detail. Click here to download a copy.