It is much easier to look after analytics on one or a few web or digital properties. However, even simple things could be getting difficult when there are a few tens or up to hundreds of digital properties to be monitored. Perhaps, the simplest is to ensure the tracking. Tracking code failure is fatal in analytics, as no data is collected and nothing can be done.
The basic idea could be monitoring the daily page views and when it drops to zero or close to zero, then there is a problem with the tracking code. Then I further think that a sudden rise in traffic may also indicate some problems or potentials that require attention. So it comes to outliner detection.
The Intelligence Events in Google Analytics and Anomaly Detection in Adobe Analytics provide such functionality but personally do not prefer to use it. One of the major issues is there are too many digital properties to be monitored, but both Google and Adobe solutions could only reveal issues for one digital property at a time.
So I came up with something on my own. The basic idea is simple, using the standard score of the point of measure compared to a period in the past. In day-to-day detection, I picked the last 28 days for the comparison as it roughly represents a month. 90 days and 365 days are also good picks for using a quarter and year as the baseline.
The formula itself is easy, just search for the standard score or go directly to the standard score page on wiki. It can be calculated using simple average() and stdev.s() function in Excel. When the standard score is above 3 or below -3, it is an outliner and requires attention. However, to use it as a tool for early problem detection, the standard score should be calculated and monitored daily, where getting the data is a time-consuming task, especially if you have a few tens of websites to monitor.
Like enough that most analytics platforms have API to grab data and there are many existing tools or Excel add-ons built around that. Using those tools or creating your own to automatically download data and calculate the standard scores could greatly improve the tracking stability and react if any problem is identified.
The following is what I get in monitoring outliners for a list of digital properties based on 28-day standard scores. It highlights outliners in green or red so I can react promptly.
Leave a Reply