Product Metrics
Product analytics is not about pretty dashboards but about decisions. Before you plot anything you must answer four questions: what we measure, why the number moved, where along the path users are lost, and whether the business actually makes money. This topic gathers the five skills asked most often in a product-analyst interview, each broken out into its own layer below.
We start with metric design: a good metric is sensitive, attributable, hard to game, and value-aligned. Then root-causing: half an analyst's job opens with "why did X drop 20%", and that needs a structured decision tree, not a scramble of hypotheses. Funnels show exactly where users drop off and which step to fix; retention cohorts answer whether the product has a durable core; unit economics reduce everything to money — is the product profitable, or is it growing at a loss?
Topic map
- Metric design — what makes a metric good, the north-star and metric tree, guardrail and counter-metrics, Goodhart's law, rate versus count, ratio-metric pitfalls.
- Metric root cause — the "why did X drop" framework: artifact versus real, slicing order, decomposition, seasonality, mix shift, and external shocks.
- Funnel analysis — step versus end-to-end conversion, defining and ordering steps, the conversion window, where to fix, and the denominator trap.
- Retention & cohorts — retention curves and the cohort triangle, D1/D7/D30, flattening to a plateau, rolling versus classic, churn and resurrection.
- Unit economics — LTV, CAC, LTV/CAC, payback period, contribution margin, cohort-based LTV, gross versus net.
Common traps
| Mistake | Consequence |
|---|---|
| Optimizing a guardrail as if it were a target | A category error: a constraint metric becomes a goal and the balance breaks |
| Comparing today to yesterday under weekly seasonality | Every Saturday looks like an incident; you must compare to the same weekday |
| Trusting a single-dimension slice under a mix shift | Every segment is flat yet the total falls — the mix moved (Simpson's paradox) |
| Treating the biggest funnel drop-off as the top priority | Rank steps by incremental value, not by the size of the drop |
| Putting revenue instead of margin into LTV | LTV/CAC inflates several-fold; a loss-making channel looks profitable |
Interview relevance
Product cases test not recall but a way of thinking: can you see the decision behind the number? A candidate who, asked "why did the metric drop", first rules out a logging artifact instead of jumping to product hypotheses immediately gets ahead of one who starts guessing.
Typical checks:
- Design a metric set for a feature: one primary, secondaries, guardrails, and the rollout rule.
- Root-cause a drop in order: artifact versus real → segments → decomposition → external versus internal.
- Read a funnel and decide which step yields the most incremental revenue.
- Tell a healthy retention curve (flattening to a plateau) from a business with no core.
- Compute LTV/CAC and payback and say whether the channel pays back.