This article explains the end-to-end process for creating a churn prediction model in Pendo Predict. It outlines each phase at a high level so you can plan your setup, align on definitions, and activate predictions with confidence.
Phase 1: Assess your setup
Before you begin, confirm that your data and systems meet the minimum requirements. You can’t build a model without these in place.
| Prerequisite | Details |
|---|---|
| Data volume | At least one year of usage data, or enough history to capture 200 churned accounts and 500 renewed accounts, whichever is greater. |
| Outcome tracking | Churn and renewal events are tracked in your CRM with a date field for each outcome. |
| Account ID linkage | Accounts in your CRM map to accounts in Pendo using a shared identifier or metadata field. |
Phase 2: Connect your data sources
Identify the data you want to include in your model, then connect the sources that store it. Most churn models use a combination of the following data types:
| Data category | What it covers | Typical source | Required |
|---|---|---|---|
| Account information | Company attributes, segments, contract details | CRM (for example, Salesforce Account object) | Yes |
| Renewal and transaction data | Renewal outcomes, churn events, close dates | CRM (for example, Salesforce Opportunity object) | Yes |
| Product usage data | Feature adoption, session frequency, engagement depth | Pendo | Yes |
| Support data | Ticket volume, severity, resolution trends | CRM or support tool | No |
| Account engagement | Emails, meetings, and other customer interactions | CRM or engagement tool | No |
Include the sources that are relevant to your business. You can also connect additional signals from a data warehouse if needed.
Connect your sources
- Go to predict.pendo.io and sign in.
- If a data source is owned by another colleague, invite them from Profile > Invite users.
- Follow the setup guide for each source in the Predict integrations documentation.
Phase 3: Define your business metrics
Define what the model should predict before configuring anything in the product. These definitions determine how your model labels historical data and scores future accounts.
Churn criteria
Define the conditions that represent a churned account in your CRM, along with the date field that marks when churn occurred.
Example definition: Account has a renewal opportunity with a status of Closed Lost
Example date field: Opportunity.Close_Date
Renewal criteria
Define what a successful renewal looks like, along with the associated date field.
Example definition: Account has a renewal opportunity with a status of Closed Won
Example date field: Opportunity.Close_Date
Audience
Define which accounts the model should score. This is typically accounts with an upcoming renewal.
Example definition: Account has an open renewal opportunity that isn’t Closed Won or Closed Lost, with a close date in the next 90 days
Prediction window
Choose how far ahead the model should predict, such as three, six, or 12 months.
You need two to three times the prediction window in historical data. For example, a six-month prediction window requires 12 to 18 months of history.
Segmentation (optional)
If different groups of customers behave differently, define segments before building the model. You can scope a model to a segment or create separate models for each segment.
Examples: Enterprise and SMB, annual and monthly subscriptions, or product tiers
Phase 4: Build a decision base
The decision base is the dataset used to train your model. It combines your data sources, applies transformations, and labels each record based on your business definitions.
Building a decision base includes three parts:
- Select data objects. Choose relevant tables and objects, such as Account and Opportunity data from your CRM and usage data from Pendo.
- Transform and aggregate data. Join objects, aggregate time-based data (for example, usage over the last 90 days), and structure records so each represents a single scoring unit.
- Define the metric. Apply your churn criteria, renewal criteria, and audience definitions to label each record.
Phase 5: Create a correlation analysis
Create an analysis to identify which factors in your data are most related to churn.
Review data hygiene
After the analysis runs, review the data quality report. This report highlights fields that might reduce model accuracy.
Common flags include:
- Bias likelihood fields. Fields that may skew results.
- High blank rate fields. Fields with missing data.
- High variance fields. Fields with extreme value ranges.
Resolve or exclude flagged fields before continuing.
Review predictors
Review the predictors identified as correlated with churn. Use the recommendations to decide which factors to include in your model.
Phase 6: Train a model
Train the model using your selected predictors, then evaluate its performance.
Choose predictors
Select the fields to include in the model based on correlation strength and data quality.
Evaluate the model
After training completes, review performance metrics:
- Accuracy. Percentage of churned accounts correctly identified. Aim for 70 percent or higher.
- Score distribution. How accounts are grouped into risk tiers.
- Personas. Common patterns associated with each risk tier.
If performance is low, revisit your data quality and metric definitions before retraining.
Review the simulation
Preview how the model scores your current audience. Validate that the results align with your expectations before activation.
Phase 7: Enable the Predict CRM widget
Set up the Predict CRM widget so account owners can view scores and recommended actions in their workflow.
Configure the AI agent
Prepare the agent that generates recommended actions:
- Upload playbooks that reflect your retention strategies.
- Provide instructions for how the agent should structure and present action plans.
Deploy the browser extension
Work with your IT team to install the Pendo browser extension for users who need access to Predict in your CRM.
Phase 8: Activate predictions
Activate your model by setting up automations to distribute scores.
- Send scores to your CRM. Sync prediction outputs to account or opportunity fields.
- Send scores to Pendo. Store scores as account metadata for segmentation, guides, and reporting.