10 Data Analyst Metrics Interview Questions and Answers

Short answer: A good metric is measurable, comparable, understandable and hard to game, and above all actionable: it can drive a decision. A KPI is the same kind of metric but tied to a goal and a target value for a period; a KPI is a subset of metrics, not a synonym.

In depth:

1. Actionable, not vanity — a metric should change someone's decision. If the number goes up but nobody knows what to do, it's a vanity metric.
2. Comparability — prefer rates and ratios (conversion, retention %) over raw counts: they compare across periods and segments.
3. Hard to game — ask "how could this be inflated without creating value?". "Clicks" are easy to game; "activated users" are harder.
4. A precise definition — one numerator, one denominator, a fixed window and filters, or two teams will count it differently.

⚠️ Common mistake: calling everything a KPI. KPIs are the 3–5 metrics a team is actually judged on; the rest are diagnostic metrics.

Short answer: A North Star is a single metric that captures the value users get and that a team moves over the long run. A good North Star correlates with revenue but measures delivered value rather than money directly; input metrics are the levers the team controls that ladder up to it.

In depth:

1. Captures value, not activity — e.g. "weekly active teams that sent a message" (Slack), not "signups".
2. Leads revenue — it rises before the money, so it works as a steering wheel.
3. Decomposes into inputs — the North Star is a function of a few controllable levers; teams get goals on those levers.

            NORTH STAR
   (weekly active teams)
        ▲          ▲          ▲
        │          │          │
     new        activation  retention
    teams   →  (aha moment) →   W4
     ▲            ▲             ▲
  traffic     onboarding   engagement

Each lower level is an input owned by a specific team; together they roll up mathematically to the top.

⚠️ Common mistake: picking a North Star equal to revenue or total signups. That is a lagging total you can't act on directly, and it hides whether the product is actually creating value.

Short answer: DAU/WAU/MAU count unique users who perform a defined active action over a day/week/month. For a fixed 30-day period, average DAU divided by period MAU measures how often that period's users were active: 0.2 corresponds to six active days per user on average. There is no universal “good” threshold across products.

-- PostgreSQL: previous 30 complete days; day is a DATE
WITH active_events AS (
  SELECT day, user_id FROM events
  WHERE action = 'core_action'
    AND day >= CURRENT_DATE - 30
    AND day < CURRENT_DATE
), daily AS (
  SELECT day, COUNT(DISTINCT user_id) AS dau
  FROM active_events GROUP BY day
), monthly AS (
  SELECT COUNT(DISTINCT user_id) AS mau FROM active_events
)
SELECT (SELECT COALESCE(SUM(dau), 0) FROM daily)
       / 30.0 / NULLIF(mau, 0) AS stickiness
FROM monthly;

In depth: Use the same activity definition, timezone and period for both numerator and denominator. The query divides by all 30 calendar days, so inactive days correctly contribute zero; AVG(dau) over observed dates would omit them and inflate the result. NULLIF leaves the ratio undefined when there are no active users. Count distinct monthly users separately rather than summing daily counts or using an unsupported DISTINCT window aggregate.

Watch out: A calendar month and a rolling 30-day window are different definitions. Define both the period and the core action before comparing products or cohorts.

Source: PostgreSQL aggregate functions.

Short answer: Retention is the share of an original cohort meeting a specified retention rule. Exact-week retention measures activity in week N; return-on-or-after measures a return in week N or later. Churn depends on the definition of leaving and is not automatically 1 − exact-week retention.

In depth:

1. Define the event and window — for example, first purchase defines the cohort and activity during week four defines W4 retention. A user absent in W4 can return in W5, so absence is not necessarily permanent churn.
2. When complements apply — for one fixed cohort, one window, and mutually exclusive “subscription retained”/“churned” states, the shares sum to 100%. Metrics with other denominators, reactivations, or revenue need separate formulas.
3. Compare equal ages — recent cohorts do not yet have a full W4. An incomplete observation window must not count as zero retention.

Retention by activity in the specified week (%)
Cohort  │ W0   W1   W2   W3   W4
────────┼───────────────────────
January │ 100   48   39   36   35
February│ 100   45   37   34   33
March   │ 100   38   28   24   22

Acquisition-driven growth can hide worsening new cohorts. A stable plateau is useful evidence, but does not alone prove product-market fit: value, usage frequency, segment, and economics matter too.

⚠️ Common mistake: turning W4 retention of 35% into “65% permanently left over four weeks.” Agree on the definition, denominator, and observation window first.

Short answer: A funnel is a sequence of steps toward a goal; at each step some users drop off. Step conversion is the share moving from one step to the next; overall conversion is the share completing the whole path — the product of all step rates. Find the biggest drop-off by the largest relative fall at a step, not by absolute counts.

In depth:

1. Step vs overall — overall = step₁ × step₂ × … Four individually-90% steps still give 0.9⁴ ≈ 66% overall.
2. Where to fix — target the step with the worst step conversion: that's the biggest upside.
3. Fix a window and order — over what timeframe, and is strict step order required; otherwise conversion is ambiguous.

Checkout funnel (step % → next step)
View       ████████████████████ 10,000
  ↓ 40%
Add-to-cart ████████          4,000
  ↓ 75%
Checkout   ██████             3,000
  ↓ 30%  ◄── biggest drop-off
Paid       ██                    900
Overall: 900 / 10,000 = 9%

⚠️ Common mistake: averaging step rates or chasing absolute counts. The biggest relative drop (here 30% at payment) is where the growth is hidden.

Short answer: First I check whether it's a real drop or a data defect, then I segment to localize it, then I separate internal causes (a release, an experiment) from external ones (a holiday, a down partner, seasonality). The goal is to narrow from "everything fell" to "segment X fell because of Y".

In depth:

1. Real or artifact? — did the logger/ETL break, are there duplicates, did the metric definition shift, is the period's data complete.
2. Segment by dimensions — platform, app version, geo, new vs returning, channel. A sharp cliff in one slice = a local cause.
3. Internal vs external — does it line up with a deploy/flag/A-B; external means a holiday, a payment-partner outage, a competitor launch.

Metric −15%
├─ Data intact? ──no──► fix the pipeline (false alarm)
│        └─yes
├─ One segment? ─yes──► iOS v4.2 → release regression
│        └─ all segments
└─ Coincides with a release? ─yes─► roll back
         └─no──► external: payment provider outage

⚠️ Common mistake: jumping to product hypotheses before ruling out broken instrumentation. Half of "crashes" are broken tracking or incomplete data.

Short answer: An average squeezes a heterogeneous audience into one number and hides differences between segments — sometimes to the point of Simpson's paradox, where the overall trend is the opposite of every group's trend. Segmenting by cohort, platform, geo, new vs returning shows who is actually moving the metric.

In depth:

1. The mean masks spread — a flat overall number can hide a rising and a falling segment cancelling out.
2. Simpson's paradox — a shift in segment mix flips the overall result even when each segment improves.
3. Useful slices — new vs returning, platform, geo/language, acquisition channel, plan.

Overall conversion "dropped" though no segment got worse — only the mix shifted. Without the slice the conclusion would be wrong.

⚠️ Common mistake: reporting one averaged number and deciding on it. Always ask "average of what?" and slice into segments before concluding.

Short answer: CAC is the cost to acquire one paying customer; LTV is the total margin profit from a customer over their lifetime. A healthy benchmark is LTV:CAC ≈ 3:1 with a payback period ≤ 12 months (for SaaS). Below 1:1 you lose money on every customer; far above 3:1 you may be under-investing in growth.

In depth:

1. CAC = all sales+marketing spend / new customers in the period (not per lead).
2. LTV = ARPU × gross margin × average lifetime (or ARPU×margin / churn rate). Use margin, not revenue.
3. Payback = CAC / (monthly revenue per customer × margin): how many months until a customer repays their acquisition.
4. The ratio = LTV / CAC: the single health gauge of unit economics.

⚠️ Common mistake: computing LTV from revenue rather than gross margin, and ignoring payback. A pretty 4:1 is useless if payback is 24 months and you lack the cash to bridge it.

Short answer: A vanity metric rises pleasantly but doesn't change decisions (cumulative signups, page views); an actionable one is tied to a decision and controllable (activation, conversion, retention). A guardrail is a protective metric alongside the goal: optimizing one can break another, and the guardrail stops you "winning" by doing harm.

In depth:

1. Vanity only goes up — cumulative counters never fall, so they're useless for decisions. Use rates and fresh cohorts.
2. Actionable = decision + control — it's clear what to do on a rise/fall and which team owns it.
3. Goal + guardrail — move the goal metric while holding the guardrail no worse than a threshold.

⚠️ Common mistake: optimizing the goal metric with no guardrail. You can juice clicks with clickbait and tank trust — the metric rises while the product erodes.

Short answer: A lagging indicator measures an outcome that already happened (revenue, quarterly churn) — accurate but late and hard to steer. A leading indicator measures early behavior that predicts the outcome (activation, week-1 engagement) — you can act on it now. An analyst needs both: leading to steer, lagging to confirm.

In depth:

1. Lagging = result — it reliably closes the books, but you learn it when it's too late to change.
2. Leading = early signal — steerable and predictive, but you must validate the link to the outcome in the data, or you chase noise.
3. Pair lever → outcome — pick a leading metric that statistically predicts the lagging one you care about.

⚠️ Common mistake: steering the product only by lagging metrics like revenue. By the time the quarterly total falls, the cause occurred weeks ago — watch the leading signals.