Lesson and practical work
Professional outcomes
- Translate vague language into measurable outcomes and comparison groups.
- Define population, period, metric, benchmark, and decision before calculating.
- Distinguish a diagnostic question from a descriptive report.
- Identify context that must be resolved with the stakeholder.
Need a refresher?A useful analytical question has boundariesView concise foundation
A measurable question normally specifies a population, metric, period, comparison, segmentation, and intended decision. “Customer performance is getting worse” specifies none of them.
“Our customer performance is getting worse”
A commercial lead raises this concern after hearing more complaints. Possible meanings include lower revenue, fewer repeat purchases, higher churn, falling frequency, worsening satisfaction, or slower service. Treating one available proxy as the definition could lead to the wrong analysis.
Convert the concern into a question tree
- Outcome: Has net revenue per active customer declined?
- Behavior: Did frequency, order value, or retention change?
- Experience: Did complaints, delivery time, or cancellations change?
- Segments: Is movement concentrated by cohort, region, channel, or value?
- Decision: Is the team choosing retention effort, service improvement, or pricing review?
A precise question is: “For customers active in both comparable 90-day periods, how did net revenue, purchase frequency, and complaint rate change overall and by value segment?”
What the result means—and does not mean
The rewritten question creates boundaries, but it focuses on retained customers and may hide churn. A complementary cohort analysis is needed for new, retained, and lost customers. Good framing makes exclusions visible instead of pretending one query answers everything.
Analyst's checklist
- Define active customer and confirm it matches business use.
- Use equivalent windows and allow for data latency.
- Separate new, retained, reactivated, and lost customers.
- Confirm complaints can be linked to customers and periods.
- Record which decisions the analysis supports.
Common mistakes and failure modes
- Accepting stakeholder wording as a metric definition.
- Choosing a metric because it is easy to query.
- Mixing cohorts with different exposure periods.
- Producing a broad dashboard for a focused diagnostic decision.
Challenge yourself
Revenue per active customer is down 6%, total customer revenue is up 4%, acquisition is up 20%, and repeat rate is down.
- Which populations should be separated immediately?
- Can total growth coexist with weakening customer quality?
- What decision could this evidence support?
Show hint
Separate growth in customer count from value and retention within cohorts.
Show analytical guidance
Total growth can be acquisition-led while existing-customer performance weakens. The stakeholder needs both an acquisition view and a retention/cohort view before reallocating spend.
How this may appear in an interview
How do you handle a stakeholder request that is too vague to analyze?
What a strong candidate should communicate
Clarify the decision, translate subjective language into candidate metrics, define population and comparison, identify constraints, and confirm the proposed analytical question. Clarification is part of analysis—not a delay before it.
What if stakeholders cannot agree on one definition of an active customer?
Where this skill becomes evidence
Include a question-framing section showing the original request, measurable version, scope decisions, and exclusions. This demonstrates stakeholder thinking that code alone cannot show.
Test your analytical judgment
Answer before opening the guidance. More than one defensible approach may exist.
Rewrite “marketing performance is poor” as a measurable question.
Show guidance
Why might revenue per active customer create survivorship bias?
Show guidance
Create a question tree for declining repeat purchase rate.
Show guidance
Move from reviewing expert reasoning to performing the analysis.
Guided practice can add the synthetic dataset, multi-step assignment, validation checks, solution comparison, project workflow, and progress tracking.
- Work with the data
- Make and defend assumptions
- Validate the result
- Build portfolio evidence