CONCEPT Cited by 1 source

Causal inference

Definition

Causal inference in production systems refers to estimating the causal effect of an action (sending a notification, showing a recommendation, changing a price) on an outcome — as opposed to mere correlation. A causal model answers "what would happen if we did X?" rather than "what typically accompanies X?"

In messaging/notification systems, causal models predict the incrementality of a send: the additional engagement caused by the notification beyond what would have happened organically. This is more principled than correlation-based relevance scores because it separates the signal the message creates from the signal the user would produce anyway.

Limitations

• Short-horizon causal models estimate effect of a single action over a brief window. They miss cumulative, long-term consequences (fatigue, opt-out).
• Static treatment: Treats each send decision independently, ignoring interaction effects between messages over time.
• Solving this requires moving to hierarchical or sequential decision frameworks that reason over multiple actions.

Seen in

• sources/2026-06-19-netflix-thinking-fast-slow-for-a-personalized-notification-system — Netflix's previous notification system used a causal model predicting single-message incrementality. While effective as a baseline, its short time horizon missed cumulative fatigue, motivating the move to a hierarchical Slow/Fast architecture.