Design anonymous-to-authenticated user journey stitching for ROI

WHAT THIS TESTS: This question evaluates whether you can design a distributed identity resolution pipeline that bridges anonymous and authenticated user states without creating privacy liabilities. The interviewer cares about your grasp of deterministic versus probabilistic matching, your ability to separate identity stitching from the critical login path, and your understanding that anonymous identifiers often qualify as personal data under GDPR when they can be linked to an individual.

A GOOD ANSWER COVERS: First, deterministic stitching using a durable first-party identifier such as a server-set HTTP-only cookie or localStorage token containing a UUID. This ID should persist across sessions with a bounded TTL such as 13 months and be merged into the authenticated user profile via an asynchronous CDP or identity graph at signup. Second, probabilistic fallback using fingerprinting vectors like IP plus user-agent only when consent has been explicitly collected, with clear acknowledgment that this is high-risk under GDPR and increasingly restricted by browsers. Third, consent management architecture where the anonymous ID is treated as personal data if linkable, requiring a consent banner before activation and honoring withdrawal by immediately ceasing collection and scheduling deletion. Fourth, attribution modeling for ROI by joining stitched journey events to downstream revenue in the data warehouse using a first-touch or multi-touch model, ensuring the pipeline handles reconciliation asynchronously so login latency stays under 100 milliseconds. Fifth, data minimization and right-to-deletion workflows where anonymous events are purged or irreversibly hashed if the user never converts within a defined window such as 26 months, and upon account deletion all associated anonymous history is scrubbed via cascading deletes.

COMMON WRONG ANSWERS: A major red flag is proposing third-party cookies or device fingerprinting as the primary mechanism without mentioning consent requirements or ePrivacy directives. Another is suggesting synchronous merging of years of anonymous logs during the login transaction, which would create database hotspots and violate latency SLAs. Candidates who claim anonymous IDs are automatically exempt from GDPR demonstrate a dangerous compliance blind spot that would expose the company to fines up to four percent of global revenue.

LIKELY FOLLOW-UPS: How would you handle shared devices where multiple family members browse before one signs up? What is your strategy if a user clears cookies between sessions? How do you prevent stitched data from leaking into ad-tech vendors without explicit opt-in? What is the cost implication of running identity resolution on a petabyte-scale event stream, and when would you sample?

ONE CONCRETE EXAMPLE: A media company uses a first-party cookie with a 13-month expiry to track anonymous blog reads. When the user subscribes, the login service emits an identity-link event to a stream processor like Apache Flink, which joins the last 90 days of anonymous page views to the new user_id in a CDP. The pipeline writes the stitched profile to a warehouse table with a user_id key, enabling marketing to calculate that 34 percent of paid conversions originated from three or more anonymous technical articles. If the user rejects the analytics cookie, the anonymous ID is never minted and those reads remain unattributed, preserving privacy by design.

Source: Wikipedia: Customer data platform