Business Metrics interview questions
Business Metrics questions a Data analyst / BI can be asked, from first job to senior. Open one to see the strongest answer and why it works.
A VP wants a dashboard of “active users” by Friday. What do you do first?
The strongest answer: Write the definition — who, in what window, what action — and get it nodded, then build; I do not paint a pretty undefined
Why it works: Exactly. A dashboard is a published definition. Pretty undefined is how you republish next week.
- Write who / window / action.
- Get a nod.
- Then build.
- Do not publish an undefined.
A PM asks you for “the conversion rate.” What do you pin down before you write the query?
The strongest answer: The numerator, the denominator and the window — converted to what, out of everyone who arrived or only those who reached the step, counted per user or per session, and within how long
Why it works: Exactly. Those three choices move the number far more than anything in the SQL does.
- Numerator, denominator, window.
- Users or sessions — decide.
- Say who is in the denominator.
- Write the definition down.
Two dashboards disagree on “revenue.” What do you do before you pick a winner?
The strongest answer: Find who owns the definition and what each included — I do not average them to keep the peace
Why it works: Exactly. Disagreement is a definition bug. Averaging and hiding are politics, not analysis.
- Find the owner.
- Diff the inclusions.
- Do not average peace.
- Do not invent a third.
Marketing’s “lead” is not Sales’ “lead.” What do you publish?
The strongest answer: Two named metrics, or one company definition with both teams on the sign-off — I do not silently pick a side in a dashboard title
Why it works: Exactly. Names are contracts. Silent sides and new undefined words are how trust ends.
- Name both, or one signed definition.
- Do not silent-pick.
- Do not average counts.
- A new word still needs a grain.
How do you calculate monthly customer churn, and what has to be stated alongside it?
The strongest answer: Customers who left during the month over the customers at the start of it — and I state how mid-month joiners are treated, what counts as leaving, and that this is logo churn rather than revenue churn
Why it works: Exactly — the denominator, the window and the definition of leaving are what make two churn numbers comparable.
- Name numerator, denominator, window.
- Base it on the start-of-period count.
- Say what counts as leaving.
- Logo churn is not revenue churn.
Your revenue is 3% from finance’s book. What do you not do?
The strongest answer: I do not scale my number to match and I do not declare victory — I diff inclusions (tax, refunds, time, channel) and I write the residual
Why it works: Exactly. A residual is a sentence about inclusions. Silent scales and blame are how you lose both rooms.
- Diff tax, refunds, time, channel.
- Write the residual.
- No silent scale.
- No blame-first.
Finance closes on a 4-5-4 calendar. Your chart uses month-end. What do you publish?
The strongest answer: The calendar I used, and a note that it will not match the book until I join their period table — I do not “almost match” in silence
Why it works: Exactly. Period tables exist. Silent almost-matches are how QBR fights start.
- Name the calendar.
- Join their period table to match the book.
- Do not average calendars.
- Do not mock the close.
Where does a published metric’s definition live?
The strongest answer: Next to the dashboard, versioned, with an owner — not only in a Slack thread from March
Why it works: Exactly. A short, owned, nearby contract. Slack and novels are how drift wins.
- Nearby, versioned, owned.
- Not Slack archaeology.
- Not only a title.
- Not a novel.
A north-star metric keeps getting “improved” in the SQL. What do you lock?
The strongest answer: A versioned definition, a change log, and a dual-run when it must change — silent SQL edits are a new product
Why it works: Exactly. Metrics are APIs. Dual-run changes. Secret board numbers are two books.
- Version the definition.
- Dual-run changes.
- Log who changed what.
- No second board number.
Finance and marketing quote different customer acquisition costs. What do you ask about each one?
The strongest answer: What went into the numerator — paid media only, or salaries, tools and agencies too — which new customers are in the denominator, over what period, and whether it is blended or one channel
Why it works: Right — CAC is a definition before it is a number, and blended against paid-channel is the difference people forget.
- CAC is a definition first.
- State what the numerator includes.
- Count which customers, over which period.
- Blended and per-channel differ.
How do you stop the quarterly “whose revenue?” fight?
The strongest answer: A written bridge: tax, channel, time, refunds — owned by both, updated when either book changes
Why it works: Exactly. A bridge is a product. Winning by volume or veto is how you get two truths.
- Write the bridge.
- Joint ownership.
- Update when either book changes.
- No veto-by-function.
Leadership wants one number to steer a multi-sided marketplace. What do you offer?
The strongest answer: A small tree: north star plus the few inputs that can move it, with owners — I do not pretend one ratio captures both sides without a residual
Why it works: Exactly. A tree with owners. One-ratio dogma and forty-KPI soup are the two failure modes.
- North star + few inputs.
- Owners on each.
- Name the residual.
- No 40-KPI soup.
MRR grew by a small amount last month and the CEO asks whether that is good.
What do you show?
The strongest answer: A bridge from last month’s MRR to this one — new, expansion and reactivation added, contraction and churned MRR taken away — because a small net number can hide a great deal of both
Why it works: Exactly — the components are the story. Net MRR alone cannot tell you whether to hire salespeople or fix retention.
- Decompose the movement.
- New, expansion, reactivation up.
- Contraction and churn down.
- A small net hides both sides.
A board number cannot be traced past a notebook. What do you install?
The strongest answer: Lineage from certified tables to the slide — a notebook can explore; it cannot be the only parent of a board metric
Why it works: Exactly. Exploration versus publication. Screenshots and bans miss the split.
- Certified tables → slide.
- Notebooks explore.
- Screenshots ≠ lineage.
- Do not ban exploration.
A checkout test “wins” on clicks and refunds spike. What was missing?
The strongest answer: A pre-set guardrail — refunds, latency, complaints — that can veto a primary win
Why it works: Exactly. Guardrails are part of the design. Sacred primaries and post-ship footnotes are how you eat the spike.
- Pre-set guardrails.
- They can veto.
- Do not peek until pretty.
- Do not footnote after ship.
Marketing wants you to “give each channel full credit.” What do you say?
The strongest answer: That double-counts the same conversion — I will show last-touch, first-touch, or a named model, not 100% to everyone
Why it works: Exactly. Attribution is a model you can name. Full-credit-to-all is arithmetic theatre.
- Name the attribution model.
- Do not 100% every channel.
- Explainable over flattering.
- Do not refuse to measure.
Who is allowed to change a board metric?
The strongest answer: A named owner, a dual-run, and a note to the people who already used it — not whoever has the warehouse password
Why it works: Exactly. Change control. Passwords, Slack, and surprises are how numbers become politics.
- Named owner.
- Dual-run + notify.
- Not a password.
- Not a surprise rewrite.
Who owns “revenue” as a data product?
The strongest answer: A named product owner with consumers, SLOs, and a backlog — “the warehouse” is not an owner
Why it works: Exactly. Data products have owners. Everyone, first-toucher, and vendors are how it rots.
- Named owner + consumers + SLO.
- A backlog for the table.
- Not “the warehouse.”
- Not a vendor.
A board slide is about to carry a lifetime value figure next to CAC.
What do you insist on before it goes in?
The strongest answer: That it is built on gross margin rather than revenue, that the retention assumption behind it is written on the slide, and that we quote a bounded horizon — value to twenty-four months — not an infinite one
Why it works: Right — LTV is a model, not a measurement. Margin, horizon and the stated assumption are what make it defensible.
- LTV is a model, not a fact.
- Use margin, not revenue.
- Write the retention assumption down.
- Bound the horizon.
After a merger you have two “active user” definitions. How do you close it?
The strongest answer: A written decision, a dual-run, and a sunset — I do not add them together and I do not pick in a hallway
Why it works: Exactly. One paper, dual-run, sunset. Addition, forever-both, and unmapped thirds are how the merger never ends in the data.
- Write the decision.
- Dual-run, then sunset.
- Do not add the two.
- Do not hallway-pick.
A vanity metric has a fan club. How do you retire it?
The strongest answer: Show the decision it never changed, offer the replacement, and sunset with a date — I do not mock the fans in the all-hands
Why it works: Exactly. Replace and date. Mockery and hope make martyrs.
- Show it changed no decision.
- Offer the replacement.
- Sunset with a date.
- Do not mock the fans.