Translating Business Goals into a Modeling Task
Explore how to translate broad business goals into specific predictive modeling tasks by defining target outcomes, prediction timing, and success criteria. Understand the importance of using appropriate data snapshots to avoid data leakage and ensure models align with practical business decisions. This lesson helps you specify modeling problems clearly to enable accurate and responsible machine learning solutions.
“Reduce churn for a subscription app” names a business goal, but it does not yet define a prediction problem. Churn means a customer cancels or does not renew within a specified period. A customer who has not logged in for seven days may need a prompt or onboarding help. A customer who cancels within 90 days may need a different retention decision. Those are different outcomes, so they require different data and different actions.
A complete problem statement fills in seven blanks:
Predict whether [unit] will [target outcome] within [time horizon], using information available on [prediction time], so [action] can occur. Success means improving on [baseline] while managing [error cost].
For example, the eventual statement might predict whether one active customer will cancel in the next 30 days, using information recorded today, so a retention team can select customers for outreach. We will make each part precise before choosing any algorithm.
Define one row and one outcome
The problem statement needs a stable row before it needs a model. The unit of prediction is the real-world item represented by one row. For churn, one row can represent one customer at one recorded date, rather than one customer forever.
The target variable is the outcome we want to predict. In this example, churned_in_30d records whether the customer canceled within 30 days after the recorded date. The prediction time is the moment when the organization makes the prediction. The time horizon is the future period used to judge whether the outcome happened. ...