# Frame a business problem as a machine learning project

Turns a business goal into a well-posed ML problem with target, unit of prediction, success metric, baseline, and a go or no-go check.

- Topic: Data & Analysis
- Framework: Few-Shot
- Level: Standard
- Tags: Analyze, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/frame-a-business-problem-as-a-machine-learning-project

## Prompt

```text
Help me frame a business problem as a machine learning project before I build anything.

**Business goal:** [e.g. "reduce churn among annual subscribers"]
**How a prediction would be used:** [who acts on it, when, and what they do differently]
**Data available:** [sources, history length, rough volume, label availability]
**Constraints:** [latency, interpretability needs, regulation, team skills, budget]

Work through:
1. **Is ML needed?** Could rules, a heuristic, or a simple analysis solve this? Say so if yes.
2. **Prediction target:** the exact label definition, the unit (customer, order, session), the prediction time, and the outcome window. Flag label leakage risks and ambiguous cases.
3. **Problem type:** classification, regression, ranking, forecasting, clustering - and why.
4. **Metrics:** an offline model metric that fits the use (e.g. precision at the top 10% if only 10% can be contacted) and the business metric it should move. Explain the cost of false positives vs. false negatives.
5. **Baseline:** the simple benchmark the model must beat.
6. **Data requirements:** features likely available at prediction time, class balance, and how to split data (time-based if appropriate).
7. **Risks:** feedback loops, fairness or bias concerns, drift, and what happens when the model is wrong.
8. **Proof of value:** how to test impact (pilot, holdout group, A/B test).

Return a one-page project framing doc with these sections, plus a go / no-go checklist and open questions marked [CONFIRM].
```

## What to fill in

- e.g. "reduce churn among annual subscribers"
- who acts on it, when, and what they do differently
- sources, history length, rough volume, label availability
- CONFIRM

Free to copy and use. From ThinkWithPrompt (https://www.thinkwithprompt.com/), a library of 2,663 curated AI prompts.