The problem of last click attribution versus "other model" attribution has been one of the most discussed and misunderstood issues in web marketing in the past year.
To be fair, attribution issues are mostly well understood but the solutions to the problem are not. There are infinite different ways to build an attribution model, yet nobody seems to know exactly how to build such a model for any given business, and how to validate that the model in question provides more value than the old last click model.
How many times have you seen any attribution expert provide a how-to guide to building an attribution model? (Answer: never)
Few marketers have had the opportunity to actually deploy and experiment with a proper attribution management technology. And while some have indeed had the opportunity to deploy such technology, the vast majority of them are still uncertain as to how one should go about building an attribution model that fits their needs and that is also theoretically defensible, whose superiority to the last click model is at once measurable and verifiable with acceptable levels of confidence.
- Should ad impressions be weighted the same as clicks? No, but what exactly are their relative weights is a difficult question.
- Are clicks from search worth the same as clicks from Facebook? No -- again, how does one go about assigning relative weights to these?
- What about time -- is a click that happened 12 days ago still worth anything? Yes it is, but the question is how much more than one that happened 25 days ago?
There are so many important variables to consider when building an attribution model, and frankly, for the average marketer, there is significant temptation of building a model that fits whatever we would like to show. In other words, nothing prevents someone who makes more money by selling display ads from building an attribution model that makes display ad views much more important than search clicks, for example.
The danger with letting marketers build models is to end up in a world where agencies or marketers can explain away their poor results by tweaking the attribution model. We need some kind of standard, and whether we like it or not, we may have just been given something like a standard by our old friend Google.
Google to the Rescue?
Last week, Google took a long awaited step in the direction of clarifying and standardizing the attribution problem by announcing Multi-Channel Funnels, a potential new Google Analytics feature which is in private pilot at the moment.
Multi-Channel Funnels and Assists are designed to show what happened before the last click (e.g., which channels generated assists in the conversion, enabling marketers who would like to optimize their efforts to see further than they ever did before into the influence of display on search). This is arguably the most important improvement in years for Google Analytics, and definitely the most exciting from where I stand.
There are five main sets of reports available, each of them with the usual Google Analytics dimensions, filters, drilldowns, and sorting capabilities.
- Overview: This is a simple snapshot of the total number of conversions, as well as those who has assists.
- Path Length: This report shows how many conversions required 1, 2, 3, 4 or more visits before a conversion happened.
- Time Lag: This report simply indicates how many conversions required 0, 1, 2, 3 or more days from the first channel interaction until a conversion occurs.
- Top Conversion Paths: This report highlights the routes that are most frequently observed before conversions occur. This report can also tell you how channels work together to generate conversions, and in what order.
- Assisted Conversions: This one shows which specific channels campaigns or keywords assisted and how.
Like most of you, I haven't yet been granted access to this new functionality, but from what I gather after reading Google's official statements and documentation, plus watching their videos, the models are designed to highlight the complexity of the online sales funnel, but don't require any serious intellectual gymnastics prior to getting really useful data. This is key to successfully moving beyond the last click attribution model and finally giving every channel at least some the credit they deserve.
If you have any experience with this new functionality, please feel free to let us know in the comments below.