Cohort Analysis in Google Analytics
If you were to ask me what I consider to be the most important analysis a company should constantly be looking at, it’s a cohort analysis. For those of you unfamiliar with what a cohort is, it’s a group of people who share a common characteristic over time.
What does this even mean? Let’s dive into an example to illustrate first:
Cohorts: an example
Imagine right down the street, there’s a coffee shop called Hipsterville brews. For the past three months, you’ve visited this coffee shop everyday. The coffee is great, and you always leave happy (and caffeinated). You’re a loyal customer, so you come back day after day.
Suddenly, there’s an ownership change. The new owner is cheap, so he replaces the artisanal coffee with Folgers. People don’t like it, and therefore don’t come back.
In this scenario, you could create a cohort of people who 1.) visit the Hipsterville coffee shop for the first time and 2.) come back and complete an action (purchase coffee) on subsequent days. Below is a chart to illustrate:
In this chart, the Y axis (vertical) is based on the people that first visited the coffee shop. The X axis (horizontal) is how often they came back and purchased more coffee. If you look at the first row, you’ll see that 21% of the new customers that placed their first order (on Mar 26th) came back on day 10 and purchased coffee again.
A cohort analysis would quickly tell you that new ownership is having trouble getting first-time customers to come back and make repeat purchases.
Cohorts + Google Analytics
Google recently released the ability to perform a cohort analysis in Google Analytics (Beta). I was pleasantly surprised by this addition, however the current implementation is mediocre and almost useless. I’ll first explain the functionality, but also point out the current deficiencies.
I have no problem with how the data is displayed. I personally find the grid view is the most interesting. My problem with the current GA setup is based on two options below:
- Cohort Type
Here’s a zoomed in view of the problem spots:
Problem #1: Acquisition Date
The acquisition date shown above is based on when the user visited the website for the first time.
There’s no ability to specify what this acquisition event is. For example, let’s say I fire a Google Analytics event (Goal #1) when a new user registers for a trial account. I’d much rather prefer to bucket people based on when they started a trial instead of when they first visited. Basing cohorts on the first session seems way too vague.
My rule of thumb is to base cohorts on high-intent actions that someone completes (signing up, making a purchase, starting a subscription). Visiting your website for the first time is not a high-intent action, especially for websites with tons of traffic.
Problem #2: Metric
My second problem is the metric that is being measured. Here’s a few options I could choose:
- Goal completions per user
- Sessions per user
- Session duration
What’s the common denominator here? Once again, these metrics are very high-level and abstract. If I created a cohort measuring session duration over time, how would that help me improve the business? It’s interesting, but not actionable. The current implementation leaves much to be desired.
What I’d like to see instead:
Cohorts are insightful when they are based on explicit actions. Let’s go back to the coffee example. What should the cohort look like? What matters to Hipsterville coffee shop?
Answer: “how many new people order, and then come back and place another order on subsequent days/weeks?”
In this example, I’d want the “acquisition event” to be people who placed an order for the first time. The metric would be purchases. I could then see how many people last week visited the coffee shop for the first time, and came back and made another purchase. That’s much more insightful.
Another question to answer would be: “How many people purchased the Venti-gross-terrible coffee and came back and purchased it again? If I viewed this in a cohort, I may learn that this coffee is indeed gross, and should be removed.
Cohorts are a fantastic way to measure how sticky/sustainable your product is over time. Just don’t expect the current implementation of GA cohorts to give you much insight.