Coinbase · Statistics & Data Analysis

Estimate Super Bowl QR ad sign-ups

TrueInterview

October 7, 2026 · 2 min read

On 2025-02-09, CoinFactory aired a 60-second Super Bowl television commercial containing a QR code that pointed to a registration page; anyone who signs up successfully gets a $15 coupon. Your task is to estimate how many sign-ups in the first 48 hours were incremental to the ad, and to put a measure of uncertainty around that number.

Lay out a concrete plan covering the following:

Identification: put forward at least two separate approaches — for instance a high-frequency time-series counterfactual built with synthetic control, a geo-lift/difference-in-differences design using control DMAs, or a calibrated marketing-mix-model attribution over a short horizon. Spell out the identifying assumptions behind each.
Data you would draw on: minute-granularity traffic and sign-up counts from the previous eight comparable Sundays; sessions tagged by QR UTM parameters together with rules for de-duplicating on device and IP; coupon issuance and redemption records (coupon_id, user_id, issued_at, redeemed_at); TV air times and GRPs broken out by DMA; timestamps of press mentions; heuristics for filtering bots; shifts in app store rankings; and logs of site latency and errors.
De-duplication and leakage: deal with scans from multiple devices, the QR URL being reshared through dark social channels, bots, and spillover from press coverage after the game. Explain how you would isolate the organic baseline from the paid lift, and how you would credit sign-ups that arrive late but still fall inside the 48-hour window.
Back-of-the-envelope (compute): assume the logs record 12,000,000 QR scans, that 40% survive de-duplication, that landing-page-to-signup conversion is 22%, that the baseline absent the ad is 50,000 sign-ups per day, and that press coverage lifted the baseline by 8% during the first 24 hours. A competitor ran a comparable QR ad in 8 DMAs making up 12% of our reach, cannibalizing 25% of our QR traffic in those areas. Estimate the incremental sign-ups and give a 90% CI under a sensible variance model; walk through every adjustment step (subtracting baseline, cannibalization, spillover).
Validation: cross-check against coupon redemptions (assume 20% are redeemed within 7 days) and against variation across geographies. Describe how you would reconcile disagreements among the methods and settle on a final estimate.

Overview: This question tests a data scientist's command of causal inference and attribution, high-frequency time series and geo-experiment design, event-level instrumentation and de-duplication, and the quantification of statistical uncertainty.

Coinbase · Statistics & Data Analysis

Estimate Super Bowl QR ad sign-ups

TrueInterview

October 7, 2026 · 2 min read

Lay out a concrete plan covering the following:

Identification: put forward at least two separate approaches — for instance a high-frequency time-series counterfactual built with synthetic control, a geo-lift/difference-in-differences design using control DMAs, or a calibrated marketing-mix-model attribution over a short horizon. Spell out the identifying assumptions behind each.
Data you would draw on: minute-granularity traffic and sign-up counts from the previous eight comparable Sundays; sessions tagged by QR UTM parameters together with rules for de-duplicating on device and IP; coupon issuance and redemption records (coupon_id, user_id, issued_at, redeemed_at); TV air times and GRPs broken out by DMA; timestamps of press mentions; heuristics for filtering bots; shifts in app store rankings; and logs of site latency and errors.
De-duplication and leakage: deal with scans from multiple devices, the QR URL being reshared through dark social channels, bots, and spillover from press coverage after the game. Explain how you would isolate the organic baseline from the paid lift, and how you would credit sign-ups that arrive late but still fall inside the 48-hour window.
Back-of-the-envelope (compute): assume the logs record 12,000,000 QR scans, that 40% survive de-duplication, that landing-page-to-signup conversion is 22%, that the baseline absent the ad is 50,000 sign-ups per day, and that press coverage lifted the baseline by 8% during the first 24 hours. A competitor ran a comparable QR ad in 8 DMAs making up 12% of our reach, cannibalizing 25% of our QR traffic in those areas. Estimate the incremental sign-ups and give a 90% CI under a sensible variance model; walk through every adjustment step (subtracting baseline, cannibalization, spillover).
Validation: cross-check against coupon redemptions (assume 20% are redeemed within 7 days) and against variation across geographies. Describe how you would reconcile disagreements among the methods and settle on a final estimate.