Long-time blog readers are likely aware that I’m not prone to writing about individual technologies or product features unless I have the opportunity to break the news about something new and cool (or not, as the case is from time to time.) But once and awhile a single feature comes along that in my mind is so compelling and cool I need to bend my own rules; Google Analytics new “Intelligence” offering is exactly that feature.
Just in case you’ve been living under a rock for the past month and haven’t already heard about “Intelligence” have a quick watch of the following video pulled from the Google Analytics blog:
While Google is certainly not the first vendor to apply some level of statistical and mathematical rigor to web analytics data, an honor that would likely go to Technology Leaders for their Dynamic Alert product or Yahoo for their use of confidence intervals when exposing demographic data in Yahoo Web Analytics, in my humble opinion Google has done the best possible job making statistical analysis of web analytics data accessible, useful, and valuable.
Some things I really like:
- An approachable way to determine confidence intervals via their “Alert Sensitivity” slider. While the implementation doesn’t necessarily impart the level of detail some folks would like, the slider mitigates the prevalent concern that “people won’t understand confidence intervals.”
- Great visual cues for alerts, especially when statistically relevant changes are not obvious based on traffic patterns. Sometimes traffic patterns just look like hills and valleys, even when something important is happening — for example, the next figure shows two alerts at the lowest threshold setting on September 16th that, upon exploration, turned out to be great news (that I might have missed otherwise.)
- Good visual cues regarding the statistical relevance of the insight being communicated. This is tough since Google is trying to present moderately complex information regarding the underlying calculations and how much emphasis you should be putting on the insight. By showing a relative scale for “significance” I think Google has more or less nailed it.
- Google Analytics finally starts communicating about web analytics data in terms of “expectations” instead of absolutes. All of us (present company included) have a tendency to get wrapped up in whole numbers, hard counts, and complete data sets. But we also know that Internet-based data collection just isn’t that accurate, and so any push to get us to start thinking in terms of predicted ranges and estimates is a step in the right direction. For example, I love knowing that on a given day Google Analytics “expects” between 311 and 388 people to come to my site from the UK!
- Lots more, including the ability to pivot the views and look from a “metric-centric” and “dimension-centric” perspective, the ability to aggregate on day, week, and month, and the ability to add your own custom alerts based on changes in traffic patterns. Perhaps ironically this last functionality (“Custom Alerts”) is how we’ve all historically thought about “Intelligence” in reporting, and while useful seems somewhat weak compared to Google’s stats-based implementation.
While awesome in it’s first instantiation there are some obvious things that the Great GOOG could improve in the feature. Some ideas include:
- More dimensions and metrics, although I believe both Nick and Avinash have commented that they are already working on adding intelligence to other data collected.
- Some way to expose confidence intervals and p-values would be useful (perhaps as a mouse-over) so that the increasing number of analysts with experience in statistics could have that data in their back pocket when they went to present results.
- Email alerts for the automatically generated insights, for example when “Intelligence” determines that five or more alerts have been generated it would be cool to get an email/SMS/Tweet/Wave notification.
- The ability to generate alerts against defined segments, so that I could see the same analysis for different audiences that I’m tracking.
Mostly ticky-tack stuff, but again I’m pretty damn impressed with their freshman effort. I suppose I shouldn’t be surprised since evangelist Avinash has been talking about the need for statistics in web analytics for an awfully long time, but given that so many in our industry have balked at bringing more mathematical rigor to our work (including said evangelist, oh well) it’s encouraging to see Google move in this direction.
What do you think? Are you using “Intelligence”? Is it helping you make better decisions? Do you like the implementation as much as I do? I’d love to hear your thoughts and comments.