We count on you! Protecting privacy, enabling analytics

David Barnett26 Jan 2022

Problem

Product teams store personal data such as names and emails to send results or announce new features. When a user leaves and asks to be forgotten, the data may sit in several databases, backups, an object store and a data warehouse. Analysts still need general usage numbers, and giving them raw PII only raises the risk of misuse.

Example from the article: Khan Academy wants to count users per state and hours of usage per state (fictional data in the post), but a user who leaves the site should no longer be identifiable.

Solution in brief

Encrypt every user's PII fields with a key unique to that user and keep the keys in a separate, tightly restricted table. Analysts query the usage columns, which stay in the clear, and never need the keys. Deleting a user's key makes that user's PII unrecoverable in every copy at once.

Questions this case answers
  • How to honor deletion of a person's personal data across many copies of the data without losing the aggregated analytics built on their activity
  • How to keep user-level usage analytics while honoring a request to delete personal data?
  • How to stop analysts from seeing personal data they do not need for their questions?
  • How to erase one user's PII from databases, backups and a warehouse with a single operation?

Tools and methods

  • Per-user encryption of PII fields — makes personal columns unreadable without the user's key.
  • Separate keys table — a single control point with access limited to a few people.
  • Key deletion — forgets a user in every copy.
Continue reading — it's free

Google or a one-time email code. No password.