Have you ever been compared with someone else? If you are from a country like India, then answer has to be yes. More often than not it's our parents who say something like, look at your elder cousin Ashutosh, how he has made a career of himself, you are on internet god knows doing what? Or may be your teachers saying, my students of batch 2000 were so much better, look at them where they are now, how successful they are. But my question is more rudimentary, can you compare two distinct groups of two different eras, timelines and find out which has been more effective or beneficial?
Let's see this problem in a little more detail, students who graduated in 2000 are having 15 years of experience, whereas students who graduated in 2010, are having 5 years of experience, how can you tell which group has been more successful? Cohort Analysis to your rescue. Cohorts are nothing but groups, more often than not, groups identified by timelines, or groups sharing some common characteristics.
Now let's start collecting data to showcase what I mean,
So, can you make a guess which class is more successful? May be you can, but I can't till the time I actually start looking at this data in little more detail. Let me give you an example:
Now, if you see the chart above, a new picture would emerge, where we are able to create cohorts of two classes and understand how successful they are by base-lining their year of graduation and plotting their performance over time. We can observe that the class of 2010 is probably doing better than 2000 by looking into their success graph. We can obviously do better by accounting for the inflation in the same period to come to a lot more accurate insights, but this data can actually help us understand the trend in a much better way.
How to use Cohort Analysis in Web Analytics?
Now that you know the concept behind Cohort Analysis, let's see how can we utilize it in our world of web analytics. We will again try to create groups, or cohorts on the basis of common traits and try to gauge their performance/effectiveness over time.
One of the examples can be social media:
This kind of analysis might come in extremely handy to draw your marketing plans, release campaigns and contact your customers.
Cohort Analysis on the basis of Acquisition Dates
GA provides very basic form of cohort analysis on the base of acquisition dates. For example see the chart below:
Table above shows the cohorts on the basis of acquisition time frame, and it's weekly performance. In the first column we are having acquisition date, on weekly basis, for 4 weeks of August. Next 4 columns show the data about the average order value on week 1, 2, 3 and 4. You can see that for the customers acquired in the week of Aug 2 - Aug 8th, week 1 revenue per order was $200, on week 2 it dipped to $170, so on and so forth.
With this analysis you can probably see the effect of some campaigns, may be there was a campaign launched in the weeks of Aug 2 and Aug 23, therefore high week 1 revenue, and then at the same time you can see when exactly my revenue numbers started showing a dip, in week 3. It seems like that in week 3 I suffer a killer blow to revenue per order and if somehow I can engage my users more in this time frame, my books will show better revenue in subsequent weeks.
In next posts we will talk about how to create cohort analysis on GA, and how to read the GA cohort charts.
