Frequency must be judged by user value, not retention metrics

What it is

The legitimate ceiling on notification frequency should be derived from the value each notification delivers to the user, not from retention metrics — opens, return visits. Retention-derived frequency ends at one notification per minute: send as long as anyone comes back. A frequency budget is the governance device that hangs the ceiling on the user-value side.

Why it happens

The two measurement systems give opposite answers. Retention metrics ask "did anyone come after we sent," do not distinguish "came because useful" from "came because interrupted," and score both as wins; user value asks what the recipient gained, and scores pure recall at zero or below. Hand frequency to the former and optimization pressure pushes toward the tolerance boundary rather than the value boundary — tolerance can hold for a long time between habituation and muting, with metrics continually "confirming" that more was right, until the user silently churns, and the churn is never attributed to frequency. Frequency also accumulates: individually acceptable notifications constitute harassment in aggregate, while per-notification attribution cannot see its own stacking.

Where it stops holding

"Judged by user value" needs an operational definition or it becomes rhetoric. Workable proxies: sampled per-notification usefulness ratings, category-level opt-out rates, and regret sampling ("should not have been interrupted"). Some frequencies are set by external facts — disaster alerts, transaction confirmations — where value-density rules do not apply: those are notices, not recall. The frequency-harassment relation is not linear; timing may matter more than frequency, and a budget fixes the quantity ceiling, not the placement.

Applying it

Give each notification category an independent frequency budget based on user-value indicators (sampled usefulness, opt-out rates) rather than open rates; categories exceeding budget enter content improvement, not continued sending. Set a separate cross-category total cap, because harassment is felt in aggregate — categories each honoring their budgets can still break the user. New categories must declare which existing budget they draw from. Verification: run frequency-reduction experiments tracking user-value indicators alongside retention — value flat or up with retention slightly down means the prior frequency was profiting metrics while overdrawing users; value falling too means the cut hit genuinely useful notifications.

Related

• Same group: P3.06.1 Notifications can manufacture sessions the user never initiated · P3.06.2 Recall unrelated to user goals is attention taking · P3.06.4 The cost of interruption lands on task resumption · P3.06.5 Badges are content-free recall signals · P3.06.6 Permission granularity determines whether users can refuse in part · P3.06.7 Timing more than frequency decides whether notifications read as harassment
• Adjacent: A5.16 Habituation and cue decay · P3.04.3 Platforms policing themselves have a conflict of interest
• Search terms: notification frequency budget · frequency capping · notification governance