How do you frame a business problem as a machine learning problem?

Framing a business problem as a machine learning problem means converting a broad business goal into a specific prediction or decision task that can be learned from data and evaluated objectively.

Start with the business objective

First define what the business is trying to improve.

For example:

• reduce customer churn
• detect fraudulent transactions
• predict product demand
• prioritize sales leads
• recommend relevant products

The business goal should come before the choice of algorithm.

Identify the decision the model should support

Ask:

• What decision will be made using the model's output?
• Who or what will use the prediction?
• What action becomes possible because of the prediction?

A prediction is valuable only when it supports a useful action.

Define the prediction unit

Specify what one prediction represents.

For example:

• one customer
• one transaction
• one product
• one loan application
• one day of future demand

This determines what each sample in the dataset should represent.

Define what should be predicted

Translate the business outcome into a measurable target.

For example:

Business goal: Reduce subscription cancellations

ML task: Predict whether a customer will churn within the next 30 days

The exact process of defining the target variable is covered separately.

Define when the prediction is made

The model must use only information that would actually be available at prediction time.

For example, if churn is predicted on January 1, the model should not use customer behavior recorded after January 1.

This helps prevent unrealistic models and data leakage.

Identify the type of ML task

Once the output is clear, the problem can usually be expressed as a task such as:

• classification
• regression
• ranking
• forecasting
• clustering

Define success

Choose both:

• ML metrics, such as accuracy, precision, recall, or prediction error
• Business metrics, such as reduced churn, increased revenue, lower fraud loss, or faster processing

A technically accurate model is not useful if it does not improve the business outcome.

In short, framing a business problem as a machine learning problem means turning a business objective into a clearly defined prediction or decision task with a target, prediction point, usable data, and measurable success criteria.