Cohort Analysis
Use cohorts in your customer analytics, there is no “one size fits for all”.
Cohort analysis is a branch of behavioral analytics in which clients are not viewed as a single set but are divided for analysis into groups called cohorts. The idea is to analyze the behavior of groups of people united based on some common behavioral patterns operating over a certain period, or in a certain location.
Cohort analysis allows companies to see patterns more clearly at different stages of the customer lifecycle, rather than parsing them across all customers blindly without regard to the natural cycle the customer is in. By observing behavioral patterns across cohorts, companies can better tailor their business to those specific customer groups.
It should be noted that cohort analysis is sometimes seen as part of a more general statistical technique called cohort study. Cohort analysis is used in business intelligence and Big Data, while cohort studies are used in medicine, epidemiology, psychology, and sociology.
Cohort analysis consists of the following steps:
- Define metric. The analysis is about selecting a meaningful indicator to evaluate the situation and optimize performance: for example, to increase sales or reduce customer churn. To do this, it is necessary to define the indicator to be evaluated, for example: customer churn rate, number of purchases, CLV, etc.
- Form cohorts. You need to decide what metrics to group customers by, i.e. what is the starting point for forming a cohort. One of the most popular options for cohorting is the first action a customer takes when they engage with the company, such as registering on a website, making a purchase, or downloading an app.
- Compare cohorts based on metrics. The analysis consists of discovering differences between cohorts and explaining the patterns of customer behavior specific to a particular cohort.
Two types of metrics can be distinguished in cohort analysis:
- Actionable metrics are metrics that link repeatable actions to observable outcomes (e.g., user registration followed by a purchase). These types of metrics help to identify the real situation, make decisions, and improve the business.
- Vanity metrics are metrics that make a company look good, but don't help understand the big picture and are useless when it comes to finding solutions to improve the business (e.g., number of likes on social media).
Let's take a simple example of cohort analysis: studying the effectiveness of subscribing to a promotional newsletter. Let's say there are three ways to subscribe to an online store's newsletter: a pop-up window on the store's website, a link in an article on a third-party website, and a sweepstakes on one of the social networks to sign up for. In February, 1,000 people signed up through the window on the website, the sweepstakes attracted 700 subscribers, and the partner's blog attracted 150. These three groups make up the cohorts.
Let's analyze which group subscribes to the newsletter longer. To do this, we need data on the percentage of emails opened over several months.
| Cohort | Number of subscribers | March | April | May | June | July |
|---|---|---|---|---|---|---|
| Website | 1000 | 14% | 9% | 7% | 4% | 3% |
| Sweepstake | 700 | 5% | 2% | 0% | 0% | 0% |
| Partner's blog | 150 | 24% | 19% | 15% | 14% | 12% |
| Total | 1850 | 11% | 7% | 5% | 3% | 3% |
As you can see from the table, the most loyal readers of the newsletter are those from the partner's blog. They read the newsletter the longest. The contest on social networks brought almost nothing - they stopped reading the newsletter immediately after the sweepstakes.
From this we can conclude that the effectiveness of the sweepstakes on social media is low. It makes more sense to focus resources on advertising through partner websites. The chart shows that this conclusion is almost impossible if the sample is analyzed as a whole rather than by cohort, as the data is averaged after aggregation, hiding significant differences between cohorts.
