DAU is down 14%. No release. No outage. No campaign that ended. The head of product slides a dashboard across the table and says: "Where do you want to start?"
Most candidates start pulling threads at random. They ask about server logs. They speculate about seasonality. They suggest checking social media for negative sentiment. Fifteen minutes later, they've generated a dozen hypotheses but haven't validated a single one. The interviewer has mentally written "lacks structure" on the scorecard.
The data analyst metric investigation interview is the most common analytical scenario across data, product, and marketing analytics roles. According to a 2024 Bain & Company survey, 67% of product and analytics interviews now include a metric diagnosis component. And yet, it's the format where preparation quality varies the most — because candidates confuse "being analytical" with "being structured." They're not the same thing.
Why Metric Investigation Interviews Trip Up Strong Analysts
A data analyst metric investigation interview is a structured scenario where a candidate is presented with an unexplained change in a key business metric and asked to systematically diagnose the cause, prioritise hypotheses, and recommend a course of action — all while thinking aloud.
What makes this format genuinely hard is that the answer space is enormous. A DAU drop could be caused by dozens of things: instrumentation bugs, seasonal patterns, product changes, marketing spend shifts, competitive dynamics, platform policy changes, data pipeline issues. The interviewer isn't looking for the right answer. They're looking for the right process.
Here's what they're actually scoring:
1. Structured hypotheses — Did you frame categories of possible causes before chasing individual threads?
2. Instrumentation first — Did you separate "is this a real behaviour change?" from "is our measurement broken?"
3. Prioritisation logic — Did you rank checks by likelihood and speed-to-validate, not just by what sounds most interesting?
4. Specificity — Did you name the exact data you'd pull for each hypothesis, or just wave your hands?
5. Driving to a recommendation — Did you converge on a likely cause and propose an action, or leave it open-ended?
The typical failure mode: the candidate hears "DAU down 14%" and immediately starts theorising. "Maybe users are churning." "Maybe there's a seasonal dip." "Maybe a competitor launched something." Each hypothesis sounds reasonable in isolation. But without a framework, they're just brainstorming — and brainstorming isn't what gets you hired.
A 2023 LinkedIn Talent Solutions report found that analytical and structured thinking was the single most-cited skill gap in rejected product analytics candidates, ahead of SQL proficiency and statistical knowledge. The metric investigation question is specifically designed to surface that gap.
How to Approach the Metric Drop Scenario
The framework I've seen work best — and the one we've validated through thousands of practice sessions on MORT — follows a three-phase structure: validate, decompose, investigate. In that order.
1. Validate the data before you interpret it (60 seconds).
Your first question should never be "why did DAU drop?" It should be "is the drop real?" Check whether anything changed in the instrumentation, tracking code, or data pipeline. Did a new SDK version roll out? Did a third-party analytics tool change its methodology? Did a data join break? A 14% DAU drop that turns out to be a logging bug isn't a product problem — it's an engineering ticket. Starting here shows the interviewer you won't waste an entire investigation chasing a phantom.
Specifically, say: "I'd pull a comparison of raw event counts against processed DAU to check for pipeline discrepancies, and verify that our tracking SDK version hasn't changed in the past two weeks."
2. Decompose the metric (2-3 minutes).
DAU is a composite number. Break it apart. By platform (iOS vs Android vs web), by geography, by user segment (new vs returning), by acquisition channel. A 14% overall drop might be a 40% drop on iOS and flat everywhere else. That immediately narrows your hypothesis space from "everything" to "something platform-specific."
This is the step most candidates skip. They treat the metric as monolithic. Good analysts treat it as a sum of parts. The moment you say "I'd segment by platform first because that's the fastest way to cut the hypothesis space in half," the interviewer knows you've done this for real.
3. Prioritise hypotheses by likelihood and speed (2-3 minutes).
Once you know where the drop is concentrated, build a ranked hypothesis list. The ranking criteria matter: prioritise by (a) how likely the hypothesis is given the decomposition, and (b) how quickly you can validate or eliminate it.
For example, if the drop is iOS-concentrated:
- High likelihood, fast to check: iOS app store recently changed its tracking policy — pull attribution data pre/post policy change
- High likelihood, moderate effort: App update introduced a friction point — check session funnel completion rates for the latest app version vs previous
- Lower likelihood, fast to check: Seasonal pattern — compare same week prior year
- Lower likelihood, slow to check: Competitive app launch — pull App Store ranking data and review competitor release notes
4. Handle the curveball.
Real interviews layer in complications. The iOS tracking policy change is the standard one — it's designed to test whether you can hold your analytical framework while incorporating new information. A strong response: "That's a significant confound. If the policy change affects how we attribute sessions, the DAU drop might be partially or entirely a measurement artefact rather than a behavioural change. I'd compare our internal session logs against the analytics-reported DAU to isolate the tracking impact."
The harder curveball: the CEO needs a one-line explanation for tomorrow's board meeting. This tests whether you can compress nuance into a decision. Something like: "Most likely cause is the iOS tracking policy change inflating the apparent DAU drop — I'd recommend we report the web and Android DAU trends separately while we requantify the iOS measurement impact." That's not a guess. It's a structured position based on what you've investigated.
5. State your recommendation explicitly.
Never leave the interviewer hanging. End with: "Based on this investigation, my primary hypothesis is [X], my confidence level is [medium/high], and my recommended next step is [specific action]." Candidates who stop at "it could be a few things" fail. Candidates who commit to a position — even with caveats — pass.
What good looks like: structured decomposition, specific data requests, a ranked hypothesis list, and a clear recommendation. What bad looks like: a brainstorm of possibilities, vague references to "checking the data," and an inconclusive ending.
Why Reading About This Is Not Enough
You now have a solid framework. But frameworks on paper and frameworks under pressure are different animals entirely.
The dynamic that makes metric investigation interviews hard to prepare for is the layering. The interviewer doesn't just give you the problem and wait. They add information mid-stream. They tell you the iOS hypothesis doesn't fully explain the numbers. They introduce the board meeting pressure. They ask you to quantify confidence in your answer. The skill being tested isn't analytical knowledge — it's analytical composure.
This is exactly the dynamic we designed MORT's metric investigation scenarios to replicate. When I was building these practice sessions, I discovered something that changed how we structured them: candidates who practised against a static problem set actually regressed on interview performance. They'd memorised a flowchart and froze when the scenario deviated from their script. So we built the AI interviewer to introduce complications adaptively — the tracking policy curveball, the board meeting pressure, a second metric that contradicts the first — calibrated to each candidate's responses. The candidates who practise against that unpredictability three or four times develop genuine investigative instinct, not just memorised steps.
The metric investigation question shares analytical DNA with strategy case interviews, but the pressure profile is closer to incident response scenarios — you're expected to move fast, commit to hypotheses, and adapt when new information lands.
The Real Test
The next time someone puts a dropped metric in front of you, resist the urge to theorise. Decompose first. The analysts who get hired aren't the ones with the most creative hypotheses. They're the ones who systematically eliminate possibilities until only the likely cause remains. That's not intuition — it's discipline. And discipline, unlike insight, can be practised.