Hang on to your seats, data lovers. Google Analytics has announced a new way for examining the multiple touch points customers make on their way to conversion on your website. The feature is live for all Analytics Premium users worldwide.
Two years ago, Google Analytics introduced us to multi-channel funnel reports. These reports – available to all users – allow you to see which multiple paths visitors followed to get to your website. It counted all the visits and sources that it took a visitor to convert a goal on your site.
Earlier this year they introduced the the Customer Journey to Online Purchase tool that helped site owners get a better picture of various models for attributing touch points. This tool worked with the existing multi-channel funnels to view purchase paths.
But all of these tools required web analysts to apply values to each touch point based on various attribution models. Some models suggest the first touch point or the last touch point should be weighted higher (first-touch or last-touch attribution). Others suggest that the path of all channels be weighted equally (linear model). Either way, it required an analyst manually applying values.
Lastly, it is difficult to determine how much one particular channel worked. In combinations, a channel's contribution to making a user convert might change by any number of factors.
Google Analytics Premium has now taken attribution modeling to the next level. Premium users can now enjoy the Attribution Model Comparison Tool. Essentially automating the process, the new Data-Driven Attribution in GA Premium automatically sets the weights based on data in the world. And it’s smart!
Instead of leaving the assigning of weighted values to the analysts, the Data-Driven Attribution uses an algorithm to do it for you. It determines how individual ad interactions or other touch points across channels interact with each other. Taking samples of data from the previous 90 days, the Data-Driven Attribution model compares one set of users to another set of users and figures out the probability of each group converting.
For example, say you have a segment of visitors who converted after interacting with social media, display ads, and organic search. Perhaps that segment's conversion rate was 3 percent. Google Analytics Data-Driven Attribution will compare that segment to another segment that interacted via social media and search but not display. If that segment converts at a 2 percent rate, the attribution model can reasonably predict that how much weight the display ad channel contributed. The incremental increase in probability of converting helps feed the algorithm and becomes the basis for attribution.
Additionally, the Data-Driven Attribution model also compares non-conversion visitor groups. By running the model against both converter and non-converter provides a more accurate comparison. The attribution model looks at "similar visitors" based on various channels, ad exposures, ad placements ad types, and repetition.
While the weights across channels are automatically algorithmically applied, Analytics Premium users are able to tweak them, if they choose.
Rules for applying manual weights can be applied. After the model does its thing, manual weighting rules can apply. For example, if your administrator feels direct visitors do not have any bearing on conversion, a rule could be applied removing all weight from direct visitors.
In addition to its incredible simplicity, the algorithm is transparent. The Model Explorer Tool, another new feature, will allow you to see exactly how each channel is weighted and why. By clicking around the report the tool creates, it will help you explore how the model will actually score various channels.
Additionally, you can drill-down into secondary dimensions – similar to any other Google Analytics report – to see which combinations work best. This allows you to view how certain creative ads perform against various placements. While you many know display ads, as a channel, is weighted high, by drilling down in, you may discover sidebar skyscraper ads outperform wider horizontal banners. The data will tell you which yields the best ROI.
A Google spokesperson told Search Engine Watch this is where the model really soars.
For display ads, the tool allows you to understand the difference between click-throughs and impression views. If a user is exposed to numerous banner ads, regardless of whether that user clicks, the exposure is counted in the model.
For channels like YouTube, the data gets more granular. You can determine if full video ads were watched completely or skipped after 15 seconds. All of this data goes into the attribution model to fine-tune your ad spend.
Reporting methodology is geared toward aggregate data based on trends of a customer. It isn't prioritized toward identifying individuals. It's geared toward of users exposed to a particular channel or ad strategy. When all is said and done, it truly only speaks to the probability of how much a channel is helping visitors convert.
The model self-adjusts itself as it goes. It uses a rolling 90-day history and adjusts, if necessary, on a weekly basis.
If your new video campaign goes viral, it will account for those additional click-throughs, especially if they lead to conversions. When the viral spike is over, it will adjust accordingly.
No attribution model will ever be able to tell you exactly what any given user will do on any given visit. Attribution models that require values and weights to be applied manually are prone to personal bias and statistical inaccuracy. While this data-driven model isn't perfect, it is certainly one of the more useful.
At the end of the day, by which model are you making your decisions?