Cohorts are groups of people who share a common characteristic. They're typically organised by time (usually months) in which they started using your product. So all of the users who started to use your app/SaaS/startup in January 2018 would belong to one “cohort”, or simply a group.
I feel the actual word ‘cohort’ is more widely used in the USA and relates to academics intakes. A bit like “the class of 2018” or something like that. It's kinda a buzzword I suppose, but means something very simple conceptually; bucketing, grouping, class etc
Cohort analysis is most often associated when looking at retention of users but it is also used to understand customer lifecycle and why certain customers have different behaviors. More on that later.
In cohort analysis, you can compare two or more cohorts and see where there might be differences between them relating to retention rates, lifetime value (LTV), etc. This information is actionable and will help show what may need to be adjusted for future iterations of your startup.
In this section, we take a look at three types of cohorts - behavioural cohorts, acquisition cohorts, and retention cohorts. Let's understand what differentiates them.
Behavioural cohorts are formed when a user takes action. For example, if you create a campaign that asks users to purchase something and sign up for the newsletter in order of most recent, then every time they take this action, it will update their respective cohort. The date of the first conversion determines what month he or she falls into because there was more than one occurrence.
A behavioural cohort should include:
Behavioural cohorts are an excellent choice if your aim is to understand how consumers interact with your product.
Acquisition cohorts are formed on the day someone becomes your customer (or otherwise joins). The date they join does not matter as long as they joined during the same year/month, which creates a clear distinction between them.
Retention cohorts are grouped by when they left and how long they were with the company. If someone leaves on their own accord, you can define the cohort as those who have remained for a fixed period of time or reached your minimum customer lifetime value (LCV). A retention cohort is best if your aim is to understand why people leave
Your cohort analytics goes beyond mere segmentation by differentiating your customers based on certain actions or traits they exhibit within a specific time period.
Imagine you didn't have cohort analysis, you would just have a broad aggregate view about your users. Without observing the cohort level data, you will be unable to observe how well consumers use your product or service over time. If you merely wanted to increase the overall size of your client base, you could add costly new customers. However building a sustainable company is dependent on having repeat clients, who are considerably less expensive to keep engaged than recruiting new ones. A cohort analysis gives you a better understanding of your customers' lifecycle and behavioural patterns.
Typically, cohort analysis looks at questions like “We changed XYZ, lets see if the month on month change to ABC is improving”.
For example, XYZ = targeting ABC = retention.
“We changed our targeting. Before this change, users would typically stay with our app for 2 months. However, users that joined AFTER the change, typically stay with our app for 5 months. With no other changes being made, perhaps this is evidence that we are targeting a more motivated set of users who love what value we provide”.
In this fictional example, if you just looked at a 7 month average, that number would not be meaningful. Typically, lots and lots of changes are happening all the time, so it's important to be able to isolate not just the time but also the user segments and their particular behaviours. Slicing your data into cohorts (across time, user segmentations and behaviours) allow you to do that.
Put another way, cohort analysis allows you to drill deeper into individual level usage data and then map that across time. Aggregate data only tells you broadly a trend, which might actually be super misleading.
A cohort analysis reveals a lot of information, which is important to keep an eye on as your company grows:
Retention Rate - a metric that reveals the percentage of people who are still active in your business (monthly or yearly)
Exit Rate - a metric that shows how many people have left since they first joined.
Churn Rate - a metric to calculate the number of customers lost over time.
Lifetime Value per User - this is calculated by multiplying monthly revenue for each user with their average tenure and dividing it by total churn rate, which will show you how much money you can make off someone on an ongoing basis before they leave.
Customer Acquisition Cost - the amount of money a company invests in order to get a new customer. To arrive at a monthly marketing budget, divide the number of new consumers the platform has acquired by the total marketing dollars spent (subtracting any marketing spend on retaining existing customers)
LTV: CAC - the total lifetime value of a customer divided by the acquisition cost. You may get this ratio by dividing the total value your customers are worth by the total money you spent on them. LTV: CAC ratios may vary per company, but an LTV: CAC of at least three times is a reasonable starting point.
Cohort Analysis should become one of the primary ways that businesses measure their success because they provide deep insight into underlying trends and help with long-term planning. Cumulative data, such as Total number of users, is great to have. However you may not be making any progress on your specific value creates per cohorts.
To wrap up this article, let me use an uuuuber super example to hammer home the point; Lets say in month 1, you had 20% conversion rate from freemium to paid, across 100 users. In month 2, 3 and 4, that number went to 15, 10 and then 5. Overall, your total number of users would be UP, and so would be paid users; 20+15+10+5 = 50 paid users. 100 x 4 = 400 users. If were mind numbingly naive, you would say “I had 20 paid users, now I have 50 paid users. Things are going well”.
Anyway, don't be this person! Get your data set up so you can slice and dice as many cohorts as you want, ideally within seconds or minutes, and ANYONE in your organisation can do this too, to drill into whether changes and experiments are moving the needle or not and if you are getting closer and close to PMF.
Until next time!