In my last post I explained Latency Analy­sis, which is a canned analy­sis tech­nique in Data Work­bench. This post cov­ers a lesser-known analy­sis tech­nique called Cohort Analy­sis.

Cohort analy­sis entails seg­ment­ing a group of cus­tomers who share a com­mon char­ac­ter­is­tic over a cer­tain period of time. Cohort analy­sis becomes more infor­ma­tive when you can start to com­pare two or more groups over time.

Now that you have a basic idea of how cohorts work let’s walk through a basic cohort exam­ple so you can under­stand the value of look­ing at dif­fer­ent cus­tomer groups over time.

In this exam­ple we will look at cus­tomers com­ing to your web­site from dif­fer­ent refer­rers on Novem­ber 1st, and the num­ber of orders they have over time after that visit. This will help to iden­tify the con­ver­sion effec­tive­ness of dif­fer­ent refer­rers, and how long a refer­rer needs for the vis­i­tor to con­vert. Below I will walk you through each step to show you how to cre­ate this analy­sis in Data Work­bench, one of the capa­bil­i­ties within Adobe Ana­lyt­ics Pre­mium.

number1  number2

Steps to Cre­ate a Cohort Analysis

Step 1: First, cre­ate a cus­tomer seg­ment.  Based on the exam­ple I gave above we will look at the dif­fer­ent refer­rers that came to your site on Novem­ber 1st.  So we will open a Vis­i­tor level seg­ment in Data Workbench.


Step 2: To ensure that the Vis­i­tor has both the con­di­tions of com­ing from a spe­cific refer­rer on a cer­tain day, you need to first cre­ate a visit level seg­ment, with the selec­tions in the refer­rer table and the time table (day/week/month) and then cre­ate a Vis­i­tor level seg­ment based on it.  Remem­ber, the Vis­i­tor seg­ment dimen­sion is exclu­sive and the order in which you cre­ate it mat­ters.  So if we cre­ate a Google seg­ment first, the cus­tomers in that seg­ment can’t be in any other sub­se­quent seg­ment, when using the Vis­i­tor seg­ment visu­al­iza­tion. (There are other approaches if needed for analy­sis, but that would be the sub­ject of another post.)


After you cre­ate a seg­ment, remem­ber to rename it so you know which vis­i­tors are in that seg­ment.  You will need to repeat this step for all of the seg­ments that you want to create.

Step 3: Once you have cre­ated all three vis­i­tor seg­ments you will save the seg­ment as a dimension.


The dimen­sion will be saved in your Data Work­bench user folder.  Once the dimen­sion is saved you will be able to find that dimen­sion in table menu struc­ture and use it in var­i­ous other visu­al­iza­tions and analy­sis tech­niques within Data Workbench.

Step 4: Now that the dimen­sion is cre­ated, open a graph visu­al­iza­tion based on the time seg­ment that you selected above.  For our exam­ple, select the day graph visualization.


Next, change the dis­play from bars to lines so you can see the met­rics over time.


Step 5: Once the graph is changed to be a line graph, change the series in the graph to the dimen­sion you cre­ated for the refer­rer segment.



Open the dimen­sion out­side this graph so you will be able to under­stand which line rep­re­sents each segment.


Step 6: Now we are able to see what these spe­cific groups are doing, over time, after the time and event (refer­rer selec­tion) cri­te­ria.  Since the Google seg­ment is much larger than the other two in this sce­nario zoom into the Yahoo and MSN seg­ments.  To zoom in on a graph, hold down the left and right but­tons on your mouse and drag your mouse up or down.  Now that we are zoomed in on Yahoo and MSN, add the met­ric of orders.  To do this, right click on the Vis­its met­ric and select “add met­rics”, and then select “orders”.  Here is what the graph will look like after you have done the above selections.


You can see in the above graph there is a spike (on the day you selected) when cre­at­ing this vis­i­tor seg­ment.  The days fol­low­ing the spike are what those cus­tomers from each seg­ment are doing.  You can see that there are dif­fer­ent spikes through­out the 30 days after Novem­ber 1st.

When we dig deeper into this exam­ple we see that the MSN vis­i­tor seg­ment has a bump about 4 days after Novem­ber 1st, for orders, then doesn’t really spike until about 30 days later.  The Yahoo cus­tomers have a cou­ple of small lifts 6 and 12 days after Novem­ber 1st, but they have a larger lift days later (Nov. 19th).


This is the power of cohort analy­sis: you are now able to see which refer­rer is best at con­vert­ing cus­tomers from a cer­tain point in time. If your web­site changes on a reg­u­lar basis, using cohort is a must because one change to your site can impact dif­fer­ent cus­tomer seg­ments greatly. This also allows you to com­pare dif­fer­ent KPIs based on when the web­site changed. Select­ing the date the site changed and the user group you want to analyze.

As evi­denced from the above steps, this is a more com­plex and time con­sum­ing analy­sis tech­nique than Latency analy­sis, but it is able to pro­vide some addi­tional analy­sis points.  Once you get the hang of doing Cohort analy­sis you will find it to be another great tool to add to your Data Work­bench tool belt, and will be able to apply it to many dif­fer­ent dimen­sions of analy­sis.  If you have any ques­tions or com­ments on how to do a cer­tain analy­sis please feel free to leave them below or reach out to me on Twit­ter at @adaste.