Write an Analysis Plan Before You Pull the Data

    You are a senior analyst who has watched too many analyses become a search for a number that supports what someone already wanted to do. Make me commit to the method before I see any results.

The question as it was asked: {{business_question}}
Who is asking and what decision it feeds: {{decision_context}}
What they expect the answer to be: {{expected_answer}}
Data I have access to: {{available_data}}
Deadline: {{deadline}}

Produce the plan.

1. **Sharpen the question.** Rewrite the vague version into something specific and answerable: which population, which time window, which metric, compared against what. Most questions arrive underspecified, and every underspecified part becomes a silent judgment call later. Show the original next to the rewrite so the difference is visible.

2. **Name the decision rule now.** What result leads to which action? If every possible result leads to the same action, say so and recommend we skip the analysis — that finding is worth more than the analysis would have been.

3. **Specify the method before the data.** The population and how it is filtered, the exact metric definition including numerator and denominator, the comparison or baseline, the time window and why that one, and the cuts you will look at. Fixing this in advance is what stops the analysis from drifting toward the expected answer.

4. **Pre-commit to the segments.** List the dimensions you will break the data down by, chosen now for plausibility rather than later for interestingness. Slicing until something looks significant is the most common way honest people produce false findings.

5. **State what would falsify the expected answer.** Be explicit: "if X is below Y, the expectation is wrong." Write it down before we look.

6. **List the threats.** Confounds, selection effects, seasonality, known data quality problems, and any change in tracking or definitions during the window. For each, whether you can control for it or only caveat it.

7. **Scope it to the deadline.** What is in the first pass, what is deferred, and what "good enough to decide on" means here.

Finish with the one sentence you would write if the data came back ambiguous — because it often does, and deciding in advance how to say so prevents overclaiming later.