Intermediate everything in Growth & Experimentation, page 2
Set up a client-side button color A/B test
Stable hashing of a persistent ID into buckets, conditional rendering of the variant, exposure plus click event logging.
Attribute a mobile install to a desktop ad
Deterministic matching via a shared login is accurate but needs auth on both ends; probabilistic fingerprinting scales without login but is noisy and privacy-fraught.
Handle interaction effects on a shared page
Combined variants may produce effects neither has alone; use mutual exclusion for likely interactions, orthogonal designs with interaction monitoring otherwise.
Why repeatedly extending a test inflates false positives
Repeatedly checking and extending until significance is p-hacking via optional stopping, which inflates the false-positive rate; fix with fixed sample sizes or sequential…
Architect a configurable, goal-based onboarding flow
Capture the goal, let the backend return a server-driven flow definition mapping goal to steps and content, render generic components on the client.
Increase experiment velocity for non-engineers
Server-driven config, feature flags, and a self-serve UI let non-engineers ship copy or layout variants instantly; add guardrails and metric checks.
Prevent conflicting experiments with layers
Group conflicting experiments into one layer so a user's per-layer bucket maps to at most one of them; orthogonal layers can overlap.
Design a contamination-safe pricing experiment
User-level price tests leak via fairness perception, so use geo holdouts or time-based cohorts where everyone in a unit sees one price.
Run concurrent experiments without interference
Independent non-interacting tests can share traffic through orthogonal layers; interacting ones need mutual exclusion in a shared layer.
Design a referral feature's lifecycle and races
A referral entity with explicit states, a unique constraint on the invited user, and atomic transactions plus idempotency to prevent double credits.
Communicate forecast uncertainty with prediction intervals
A point estimate hides risk; produce a prediction interval via model error, simulation, or scenarios, and state assumptions.
Combine qualitative and quantitative data for hypotheses
Quant reveals what and where, qual reveals why, then triangulate into a falsifiable hypothesis with a metric.
Explain RICE scoring and its Confidence factor
Score equals Reach times Impact times Confidence divided by Effort; Confidence discounts uncertain estimates; ground it in evidence tiers.
Resurrection Campaign
A resurrection campaign is a targeted effort to win back dormant or churned users by re-engaging them with relevant value, often via email or push. It matters because reactivating known users is usually cheaper than acquiring new ones.
Explain the difference between statistical and practical significance
Define statistical vs practical significance; note large samples make tiny effects significant; give a real example.
Build a system to measure viral coefficient and attribute invites to signups
Tests if you can map K=i×c to logged events and resilient pipeline. Good answers define invite_sent, click, signup events with referral tokens; sketch stream joins; and flag cross-device and organic attribution gaps. Red flag: assuming perfect attribution.
Design a referral system: data models, APIs, attribution, self-referral prevention
Tests data modeling with fraud guardrails and idempotent rewards. Cover: Users with nullable referred_by, ReferralEvents state table, async ledger attribution, and device-fingerprint self-referral blocks. Red flag: bare integer credit with no audit trail.
Describe cluster or switchback randomization for network-effect A/B tests.
Tests SUTVA violation, cluster-switchback tradeoffs. Outline: cluster (geo, teams) or switchback (time slices) isolation; cover pipeline changes, 10x inflation, and correlated error. Red flag: user-level randomization with post-hoc fixes or ignoring spillover.
How do you prevent concurrent onboarding and navigation experiments from polluting results?
This tests experiment isolation via layer-based traffic allocation. A strong answer covers hashing users into independent layers with one variant per layer, and assigning each experiment to a distinct layer.
How do you systematically manage and pay down experiment debt?
Tests sustainable velocity through experiment lifecycle hygiene. Strong answers cover isolated experiment directories, TTLs on feature flags, and recurring cleanup sprints. Red flag: banning experiments or treating all experiment code as permanent.
We are hiring for this. Every open role lists the topics its interview covers, so you can prepare for the real thing rather than guessing.
See open roles