Cohorts modeling

Unlike standard time-series forecasting, cohort models let you isolate trends across specific dimensions like start date, customer tier, or region. This makes it easier to answer questions like:

• How do self-serve customers from Q1 2024 behave by month 6?
• What’s the cumulative revenue impact of our March enterprise cohort?
• Where is growth accelerating or tapering across segments?

here are a few characteristics that make cohort modeling a bit more specialized:

• Data source configuration: Cohorts are typically made up of many individual entities—like deals from a CRM or users from a database. To fully leverage this data in Runway, some setup is required.
• Granularity: Instead of tracking individual rows or events, cohort models group data into time-based cohorts and analyze patterns across those cohorts.
• Relative timing: Most modeling in Runway is calendar-based (month-by-month). With cohort modeling, assumptions are often made on a relative timeline—for example, 3 months after a cohort’s start.

This guide walks through how to build a flexible, forward-looking cohort model using age-based assumptions. While the example shown is for demo purposes, you can adapt the same concepts to fit your own forecasting needs. You can also explore the full model in this Runway example.

We’ll walk through how to:

• Import raw CRM-style data into a Source data database
• Structure your cohorts using drivers and dimensions in a Cohorts database
• Set up and apply cohort-based assumptions in an Assumptions database
• Roll everything up into a Consolidated cohort overview to see the full picture

Along the way, you’ll set up a flexible foundation for forecasting cohort behavior—whether you’re modeling usage-based revenue, product adoption, or any outcome that plays out over time. You’ll be able to see trends at a high level, and drill down into the individual cohorts that drive those patterns.

Let’s get started.

​

Step 1: Bringing in CRM data into a source data database

If you’re importing already-cohorted data from external sources like Snowflake or BigQuery, you won’t need to segment by dimensions like “customer name.”

To configure a new source data database:

​

Drivers

1. Select the column that indicates the month when values for a given deal should start applying.

2. Check the columns containing numbers or dates you plan to use in your model.

   We recommend enabling “Most Recent Value” — this ensures values continue to apply in future months beyond the start month. Read more about filling sparse data here.

​

Segment by and Other columns

1. Segment by the most detailed dimension you expect to drill into:
   • If that’s customer name, make sure to:
     • Add any supporting metadata in the Other columns section (e.g. sales owner, contract type).
     • Add the cohort start date as an Other column — this helps with rollups in the next step.
     • If your cohort start dates vary within a month, you’ll see an option to group timestamps by month (or another timeframe).
   • If you’re working with pre-grouped cohorts (e.g. without individual deal visibility or with too many deals to break out), segment by the cohort start month:
     • You can group assorted cohort dates by month or other units if needed.
     • If you want to split cohorts further by other dimensions (like product line, region, or contract type), select those as additional Segment by options.
2. Use Other columns to bring in any extra metadata you want available in the model.
3. Once your configuration looks good, press Confirm.

For more detail on database configuration see article.

​

Step 2: Configuring formulas on source data

The next step is making sure your cohort data is modeled correctly across time — especially for numbers that extend into future months.

1. Apply forecast formulas to match actuals where needed.

   If your cohort data includes values extending into the future (e.g. pipeline data), copy the actuals formula into the forecast for your driver columns.

   This ensures that the “Most Recent Value” is carried forward even beyond the Last Close.

2. Use Timeseries view to manage formulas.

   Switch your database view under Customize > View as > Timeseries to easily manage formulas across all drivers and trace how values flow from one driver to the next.

   → Learn more about Timeseries view

​

Step 3: Setting up cohorts database

In this step, you’ll group your source data by cohort so you can model at the cohort level and track trends over time.

Steps:

1. Create a new database

   This will show one segment per cohort. A name like Cohort Model works well.

2. Configure the source

   Point the cohort database to the one holding your individual customer or deal-level data.

   If you’re importing already-cohorted data directly from a data warehouse (e.g. via Snowflake or BigQuery), select the appropriate integration query as your source instead of referencing another Runway database.

3. Drivers

   • Bring in the cohort start month as a driver — this allows for time-based modeling.
   • Include any drivers you want to aggregate and model at the cohort level, such as: Seats, Contract value, Revenue.

4. Segment by

   • Select the dimension that defines cohort start (e.g. Contract Start).
   • Add any other dimensions you’d like to separate cohorts by (e.g. Geo, Product Line, Contract Type).

Hit Confirm and review the segments. If anything looks off, you may need to adjust your Segment By or Driver selections.

Now that each cohort is set up as its own segment, the next step is to add columns that enable cohort-based modeling.

1. Add a driver for cohort age

   This driver calculates the number of months since each cohort started. It allows you to apply the right assumptions in the right month based on cohort age.

2. Add drivers for key cohort metrics

   If there are parts of your model (like churn rate, expansion revenue, etc.) that aren’t available in the source data:

   • Add placeholder drivers now so that each cohort/segment has everything needed for forecasting. You can leave formulas blank for now — you’ll connect them to assumptions in the next step.
   • Optional: If you want more granularity or plan to drill into these metrics, you can create them in the source data database and roll them up via the cohort database configuration. Just keep in mind that approach may be less ideal for high-level forecasting.

Next up: setting up an assumptions database — where you’ll define the rules that govern how each cohort behaves over time.

​

Step 4: Setting up cohort assumptions database

In this example, we’ll create a database to hold the assumptions that drive your cohort model — specifically, Seat growth rate as a function of both Cohort age and Customer tier.

Here’s how to set it up:

1. Create a new database and segment it by:

   • Cohort Age (as a dimension)
   • Tier (e.g. contract tier, customer segment)

2. Add rows for each combination

   For each unique combination of age and tier, add a row. In this example, we’ve added 12 months of assumptions for each tier.

3. Add a driver column for the growth rate

   This is where you’ll enter assumption for the expected seat growth for each cohort over time.

4. Add a numerical version of cohort age

   To apply these assumptions based on a cohort’s calculated age, add an extra driver column:

   • Name it Cohort Age - Number to distinguish it from the dimension.
   • This enables formulas to correctly match the cohort age from your main cohort model.
   This duplication is a known limitation — Runway plans to improve formula functionality so that dimension values can be referenced directly in the future.

5. Optional: Calculate assumptions dynamically

   While the growth rates in this example are hardcoded and static across months, you can:

   • Change formulas over time.
   • Reference other data sources in your model to make these assumptions dynamic and responsive.

Each of these five steps is labeled in the annotated screenshot below:

​

Step 5: Applying your assumptions in cohort forecasts

With both your Cohorts database and Assumptions database set up, you can now connect them to forecast key metrics — like growth in Seats — based on each cohort’s age and tier.

In your Cohorts database, use the cohort’s age and the dynamic filtering capability of This Segment to fetch and apply the right assumption values for each month.

​

Step 6: Aggregating your cohorts into a consolidated overview

The final step is to create a consolidated overview — a high-level rollup of all cohorts into the segments that matter most for your business. This view makes it easy to analyze trends at a glance while preserving the ability to drill into specific cohorts and underlying customer data.

Example: Summarizing by tier

1. Create a new database
2. Set its source to your Cohort Model database
3. Segment by: Tier (or another dimension you care about)
4. Pull in the drivers you want to aggregate (e.g. Seats)
5. Press Confirm

You’ll now see one row per tier — a clear overview of performance by segment. From this view, you can drill into:

• The individual cohorts that contribute to each tier
• The underlying customer-level data from your source database

The video below shows how to configure the Consolidated Overview and how to drill in from the top-level view down to specific assumptions, cohorts, and customers.

Cohort modeling is famously nuanced. If you’d like help applying these ideas to your own setup, don’t hesitate to reach out to your CXM — they’re here to help.

Happy modeling!