Why scoping comes first
Most failed ML projects fail before training starts. The team builds an accurate model for the wrong problem. Scoping forces you to write down exactly what you predict, for whom, and how success is judged.
- Name the decision the model informs, not just the output.
- Pick a target variable you can actually measure and label.
- Define a metric tied to business value, plus a guardrail metric.
Framing questions
Ask whether the task is even ML shaped. A clear rule or lookup is cheaper than a model. Then check that labels exist or can be collected affordably, and that a useful prediction arrives in time to act on.
- What does a wrong prediction cost in each direction
- What latency and volume must the system handle
- What baseline are we trying to beat
A scoping sketch
Writing this down early prevents months of work aimed at an unmeasurable goal.
Key idea
Scoping converts a fuzzy business wish into a concrete prediction target, a value aligned metric, and a feasibility check before any model is built.