Attribute a mobile install to a desktop ad

WHAT THIS TESTS This evaluates understanding of identity resolution across devices and the modern privacy constraints that shape it. The interviewer wants the two method families, their accuracy and coverage trade-offs, and honest treatment of consent and platform attribution frameworks.

A GOOD ANSWER COVERS The goal is to connect a desktop ad impression to a later mobile install. Deterministic matching relies on a shared, stable identifier across both devices, almost always a logged-in account: the user sees the ad while signed in on desktop, then signs in on the mobile app, and the common account id links the two with high confidence. It is accurate but only covers users who authenticate on both touchpoints, and you must store and join these identities carefully. Probabilistic matching infers a connection from indirect signals, shared IP address, timing proximity, and device or browser characteristics, building a statistical model of how likely the desktop and mobile events belong to the same person. It extends coverage to logged-out users but yields probabilities, not certainty, so it introduces false matches and degrades when many devices share an IP. On privacy: deterministic methods need consent and careful identity handling; probabilistic fingerprinting is increasingly restricted by platforms and regulation. For mobile installs specifically, you typically route through the platform's attribution framework and deferred deep links rather than raw fingerprinting, and you respect consent signals.

COMMON WRONG ANSWERS Claiming probabilistic matching gives exact attribution. Ignoring privacy law, consent, and platform attribution rules. Assuming you can read a stable cross-device identifier with no login. Forgetting that shared IPs and households create false matches. Proposing raw device fingerprinting as if unrestricted.

LIKELY FOLLOW-UPS How do you measure probabilistic match quality? Validate against deterministic ground truth. How do platform attribution frameworks change the design? They mediate install attribution and limit cross-app identifiers. How do you handle consent and opt-outs end to end?

ONE CONCRETE EXAMPLE A user signed into their account clicks a desktop ad, recorded against their account id. Two days later they install the app and sign in; the shared account id deterministically attributes the install to that impression. For a logged-out user, the system instead notes the click came from a given IP and time, and a later install from a matching IP within a short window is attributed probabilistically with a confidence score, flagged as lower certainty and subject to consent and platform rules.