Recap — The Case-Study Framework

# The Case-Study Framework Most DA interviews include a "case study" — an open-ended product question like *"signups dropped 12% last week. How would you investigate?"* or *"the PM wants to redesign onboarding. How would you measure success?"* These questions don't have a right answer. They test how you structure ambiguity. Candidates who freeze fail. Candidates who have a framework win. > **You've already done this once.** Week 8's Case Study A ("Signup Conversion Dropped 8% Last Week") walked you through exactly this kind of investigation — noise check → source → funnel step → segment → external cause. This lesson isn't teaching new analysis; it's teaching how to *narrate* that same reasoning out loud, under time pressure, to an interviewer. ## The 4-step framework ### Step 1 — Clarify the question (60–90 seconds) Never start answering. Ask: - "What's the time period and segment we're talking about?" - "Is this signups overall, or by source?" - "Are we looking for a root cause, or are we trying to decide what to do next?" These questions buy time AND show you understand that the same headline can mean different things. ### Step 2 — Frame the approach (90 seconds) Out loud, sketch your investigation plan: > "I'd approach this in four steps. First, confirm the drop is real — i.e., outside normal week-to-week variation. Second, segment by acquisition source, device, and geo to find where it's concentrated. Third, check whether anything changed — campaigns, product, tracking. Fourth, propose a recommendation." This is the magic moment. You've now showed you can decompose the problem. Even if your specific answers later are imperfect, the structure scores points. ### Step 3 — Walk through, narrating decisions (5–10 min) Go through each step. At each step, narrate why you're choosing one direction over another: - "I'd check overall first because if the drop is in noise range, we don't have a problem." - "I'd look at acquisition source second because the most common cause of sudden drops is campaign changes, not product issues." - "I'd save tracking-bug investigation for last because it's the most expensive to verify and least common." Reasoning > correctness. Most case studies don't have one right path; they have defensible paths and undefensible ones. ### Step 4 — Synthesize & recommend End with a clear recommendation: > "Given the information I have, my best guess is acquisition source X. To confirm, I'd want to see channel-level spend last week vs. baseline. If that's the cause, recommendation is to either restore spend or accept the lower funnel-top and recalibrate forecasts." This closes the loop. Candidates who never reach a recommendation seem indecisive. The recommendation can be conditional ("if X, then Y"). ## The three things you must not do 1. **Jump to an answer in the first 30 seconds.** Showing your hand fast looks decisive but it's actually impatient. 2. **Get stuck inside Step 2 forever.** You'll see candidates spend 8 minutes asking clarifying questions. Cap it at 2 minutes total. 3. **Hedge everything.** "Maybe it's this, but could also be that, hard to say." Have a strongest hypothesis. Be honest about uncertainty AROUND that hypothesis, not instead of it. ## Reusable phrases for the interview - *"Let me think out loud — I'd approach this in [N] steps."* - *"My strongest hypothesis would be X, because [data-driven reason]."* - *"If I had access to the data, the first query I'd write is [specific query]."* - *"My recommendation would be Y, conditional on confirming Z."* These aren't tricks. They're the literal way senior analysts talk. Saying them puts you in the same conversational bucket. ## Takeaway Practice this framework on 5 case-study questions before any interview. Once internalized, you'll handle any open-ended question without freezing. It's not about being smarter — it's about being structured.