Clarify Before You Solve

Start With the Decision

An interviewer asks, "How would you analyze customer churn?" A weak response immediately names charts, algorithms, or code. That is risky: churn may not be defined, the time horizon may be unknown, and it may be unclear whether the goal is explanation, prediction, or a retention action.

A useful opening is: "Before choosing an approach, I would clarify the decision this analysis should support, how churn is defined, and what information is available at the time we need to act." This creates direction without delaying the conversation.

A Compact Clarification Framework

Use this mental checklist, not a rigid script:

1. Goal: What decision should this support?
2. Definition: What exactly does the target, metric, or problem mean?
3. Unit: What does one row or entity represent?
4. Time: When is analysis or prediction made, and over what horizon?
5. Data: What inputs are available then?
6. Constraints: Are latency, explainability, capacity, or business rules relevant?
7. Success: How will the team judge the result?

For SQL, "Find our best customers" might mean revenue, profit, retention, or value over a particular period. For an experiment, "Did it work?" needs a primary metric, a meaningful effect size, and a decision context. For fraud modeling, prediction timing, false-positive cost, and review capacity matter before model choice.

Ask, Then Move Forward

Clarification should create momentum. Do not ask every possible question when the prompt already contains enough context. Ask the highest-value two or three questions, then state an assumption:

"If customer value means completed revenue over the last 12 months, I would first confirm the order grain, exclude refunds, then aggregate by customer."

This makes the assumption visible and gives the interviewer a chance to correct it. Refusing to proceed until every detail is known is no stronger than silently making a major assumption.

Weak Versus Strong

Weak: "I would use Random Forest, check feature importance, and make a churn dashboard."

Stronger: "I would first ask whether the goal is to explain churn patterns or prioritize outreach. For a prediction workflow, I would define the scoring date and churn horizon, because that determines which features are valid. Then I would inspect target balance, available actions, and the cost of missed churn before choosing metrics or models."

The stronger answer demonstrates problem framing, leakage awareness, and business judgment. It does not claim an algorithm is already the answer.

Rehearsal Prompts

1. An interviewer asks, "Find our most valuable customers." What three questions would you ask?
2. What assumption could you state if experiment success is not defined?
3. For a no-show model, what timing question protects against leakage?
4. When has an interviewer already provided enough context to begin?
5. Turn "build a fraud model" into a concise problem-framing statement.

Interview Perspective

What this demonstrates: Many interviewers are evaluating whether you can turn an ambiguous request into a tractable decision before producing work. A focused clarification is evidence of judgment, not avoidance.

Next Lesson

Next, make that reasoning easy to follow without narrating every thought.