Cohorts

How to use cohorts in building financial models

Cohorts are subsets of users, customers, or other entities that share common characteristics. Grouping data into cohorts allows us to find patterns, trends, and insights that might otherwise be hidden in the overall dataset.

Types of cohorts

The most common types of cohorts in financial modeling include:

  • Acquisition cohorts: Group users based on when they first signed up or became a customer (e.g. all users that first made a purchase in January 2024).
  • Behavioral cohorts: Group users based on specific behaviors, such as making a purchase, subscribing to a service, or interacting with a particular feature (e.g. all users that have made a purchase in the last 60 days).
  • Time-based cohorts: Group users or customers by their interaction within certain time periods (e.g., week one, month two, etc.). Particularly powerful for tracking user behavior and performance metrics across different time periods.

Often we will combine approachs, creating cohorts based on user behavior or acquisition time period and using time-based cohorts to see the trends over time for those subsets. Here's a few examples of what we could use that approach to:

  1. Track customer retention. How long do users stay with active? If you acquire 100 customers in January 2024, how many are still active after three months, six months, and one year?
  2. Measure user engagement. How often do users log in? What is their frequency of use in months 1, 2, 3, etc? What types of engagement actions do they take?
  3. Evaluate marketing campaigns. How do users perform after being acquired through a particular campaign? Are there differences in the characteristics orperformance of the January 2024 cohort compared to the April 2024 cohort?
  4. Analyze product performance. How does user adoption and engagement change over time for new features?
  5. Track revenue growth over time. Do customers change their spending behavior over time? When does the total revenues pay back the acquisition cost? What is the LTV of certain types of users?
  6. Analyze churn and retention. Are there certain months where customers churn more? Do customers acquired in certain months churn differently?
  7. Understand feature adoption. Are there certain features that users adopt at certain times in their customer lifecycle? Are there usage trends that could help change onboarding strategies?

More on how cohorts are used in modeling revenues at Mastering Revenue Models.

Forecasting using cohorts

Of course, all of these examples are focused on analyzing historical data. We can also use cohorts for forecasting. The typical approach is to use time-based cohorts where you list the periods in which users or customers were first acquired, and then forecast their performance - retention, revenues, etc. - going forward.

The key is to identify the cohorts that are most revelant to your business and can help highlight important narratives or make key strategic and tactical decisions. Segment the data into relevant customer sets, user groups, product behavioral characteristics, or other aspects that are important to understand or may have differential changes in performance in the future. Setup your data sources so you can slice the data into subsets, analyze their past performance, use that to create a baseline forecast to use for future cohorts, and then forecast the size and performance of future cohorts.

How to use

Cohorts are built into many Foresight financial model templates, including:

  1. Standard Financial Model
  2. Ecommerce Forecasting Tool
  3. Venture Capital Model, Rolling Funds
  4. Fund of Funds

The structure is straightforward, the hard part often comes down to creating the standard forecast curve to then apply to each cohort, which is typically forecasted to grow each period. For example, you could use historical data to calculate the average revenue retention per month for 1 user over the first 12-24 months from date of acquisition, and then use that standard curve to apply to each cohort. Each cohort can be a different size - the number of users forecasted to acquire in each period - and then you can use the historical curve to forecast the future performance.

Cohorts can be difficult to build in spreadsheets, here's an example of how to build cohorts in Causal, a web app built for spreadsheet modeling, that also shows how cohorts are typically laid out: