Derived metrics – what is this sorcery you speak of?
Share this article
Derived metrics are calculated metrics, on steroids. You can now build complex calculations as a metric, or apply a segment to a metric, or combinations thereof, super-powering your analytics.
They’ve been around for a little while now, particularly in Ad Hoc (Discover) but they’ve come to life in Analytics through the new metrics manager (and the segment manager).
So, what can you do with them? There are generally three use cases:
- A simple segmented metric
- A calculated segmented metric
- A much more complex calculated and segmented metric
Let’s start with something simple. Perhaps you want to run some analysis around new visits as a percentage of overall visits. You could certainly do it with a segment, but derived metrics gives you the building blocks to do far more powerful things – later.
A simple segmented metric.
Using the metrics builder add the Visits metric as you would normally. Then, from the segment list, drag the New Visitors segment over (or another segment that you’ve previously created).
This new metric – New Visitors – can be applied to any report, trended over time, and even added into other custom metrics. For example, when you apply it to a page report you can add the metric against your other metrics.
A calculated segmented metric.
If you want Percentage of Overall Visits as a single metric, you can simply modify this calculation by dragging in another metric underneath it. Here we’ve dragged in Visits so our formula now is New Visits / Visits.
Don’t forget to change the format to percent and the decimal places etc. There are plenty of features available, including a preview graph to show the outcome of the calculation.
So that’s a pretty straightforward example too. And yes, you could have also achieved the same through segmentation by using comparison and would get the same result. But now for something more complex…
Using Metrics to find dogs and diamonds.
Strap in, it gets a little crazy from here.
At the Adobe Summit, you might remember Bret Gundersen introduced Derived Metrics and showed a few examples. We also featured a couple in our London Summit recap.
One of the more complex metrics he introduced was Weighted Bounce. This metric pushes the ‘interesting’ traffic to the top and bottom of the report. You can then sort or reverse sort this metric on the pages report to find ‘dogs and diamonds‘ as Bret put it. Be sure to include the real bounce rate in the report as well. The formula was as follows:
=(mean(bounce rate)*(1-(pageviews/maxv(pageviews)))+(bounce rate*(pageviews/maxv(pageviews)))
Using Derived Metrics to create this was a bit challenging with all of the nested containers etc. My top tip for complex calculations is to split them apart into components and build them as separate metrics. Then combine them into a new metric by dragging all the component metric on. The added benefit here is that you can validate each component to make sure you didn’t mess it up (like I did initially – thanks Bret).
So the end result of the new Weighted Bounce metric is shown in the summary below:
Click on the thumbnail image for a full-screen view of the actual underlying structure of the metric containers.
And now for something nuts.
We often talk about engagement metrics and scoring, it’s a been a hot topic for quite a few of our posts. With Derived Metrics, we can now create an Engagement Metric as a single metric within Adobe Analytics.
Within a single metric, we can calculate engagement based on different factors such as:
- Click Depth: % of visits with a visit depth of more than X
- Feedback % of visits who fed back during a session, such as email, comments, click-to-call etc
- Recency: % of visits that return in less than X days
- Interaction: % of visits who complete an action during a visit
- Duration: % of visits who were on the site for more than X minutes
- Loyalty: % of visits that are repeat visits
We end up with a rather crazy derived metric with segments applied across each sub-metric.
We can then review engagement by different dimensions over time, such as channel, or referring domain, or by different segments.
Don’t forget that the engagement metric is designed to quickly call our attention to domains, pages, campaigns, geographies, demographics and other factors that are somehow attracting a larger number of visitors who satisfy a pre-determined set of criteria.
This is just a sample of what can be done with derived metrics. There are many, many functions within the metrics manager that can be applied as well, such as z-scores, t-scores, percentiles, quartiles, standard deviations, all manner of regressions (another favourite of mine for another post) and all sort of other sorcery.
Derived metric sorcery? Nope, just Adobe wizardry.