Meta's marketing-messages-per-user cap was a policy note when the cap explainer documented its mechanics: one recipient receives a finite, unpublished number of marketing template messages from one business inside a rolling window, sends past the ceiling fail at delivery instead of queueing, and Meta publishes no threshold to plan against. The cadence playbook turned the mechanics into operating levers. What neither post could have was ninety days of delivery history on the other side of the shift. H2 2026 now supplies it, and the patterns in that history are the subject of this post.
Quick answer: three things hold across ninety days of operator delivery logs. Failed marketing deliveries concentrate in the lapsed and dormant cohorts, not the engaged one. Suppression events on your own side of the line rise before Meta-side delivery failures do, which makes your own throttle the earliest readable signal. And re-engagement, not the passage of time, is what restores a capped recipient: windows reset on reads, replies, and inbound messages, so a dormant list left to cool stays capped while a smaller conversational list keeps resetting. Each pattern maps to a control a tenant already owns; the second half of this post walks them.
What ninety days adds to the explainer
A rolling window only shows its shape once recipients have cycled through it more than once. The first weeks after the shift, a delivery failure told an operator almost nothing: any single fail could be the first cap hit, a bad number, or a device-side rejection, and the mix was indistinguishable. After several full window cycles the cohorts separate cleanly, because the cap is per recipient and recipient behavior is stable. The engaged minority resets its window through engagement and rarely appears in the failure tail at all. The lapsed middle hits its ceiling on the second or third campaign of a quarter and appears in the tail exactly once per cycle. The dormant end fails on nearly every send and appears on every campaign. Ninety days in, the failure tail is not noise; it is a cohort map.
That is the frame for the three patterns below. Each one is something an operator can verify in their own delivery log this quarter, without any number Meta has never published.
Pattern one — failed deliveries concentrate in the lapsed cohort
The intuitive fear when the cap shipped was that high volume to good customers would trip it. The logs do not show that. Recipients who open and reply extend their standing, so a campaign sent at a sensible cadence to an engaged cohort lands. The failures concentrate one cohort down: contacts with no engagement in the recent window who still sit in the full campaign audience, because the audience was built from opt-in status rather than engagement. They absorb a full-fan cadence on a lapsed recipient's ceiling, and the ceiling loses.
The read in your own data is a segment comparison, not a threshold hunt. Split last quarter's marketing delivery failures by last-engagement band. If the lapsed and dormant bands carry the majority of the failures while the engaged band lands clean, the cap is doing exactly what the explainer predicted, and the fix is an audience decision, not a cadence tweak: the lapsed band gets a stepped-down cadence and the dormant band graduates out of template sends entirely, as the cadence playbook's cohort structure prescribes. Exclusion at the audience layer spends nothing; a send that was never scheduled cannot fail on Meta's side.
Pattern two — tenant-side suppression leads Meta-side failure
The second pattern is the most useful operationally. Where a tenant runs its own per-contact frequency caps, the suppression events on the tenant side rise first and the Meta-side delivery failures follow only where the tenant cap is set too high or not set at all. The two curves have the same shape, separated by the gap between your ceiling and Meta's invisible one.
That ordering is the point. A suppression event in your send log is free: the send was never attempted, nothing was billed, and the contact's Meta budget was untouched. A Meta-side failure is the expensive version of the same lesson: the spend happened and the recipient's window absorbed the hit. Ninety days of data says the operators who read their own suppression count weekly and tighten when it climbs in the engaged cohort stay ahead of the cap; the operators who wait for Meta-side failures discover their ceiling by spending it. Your own throttle is the only cap in the chain whose threshold you set, which is exactly what makes it the leading indicator.
Pattern three — replies, not waiting, reset the window
The third pattern decides what a dormant list is worth. A dormant cohort left to cool does not recover: recipients who receive nothing still do not re-engage, and when the next campaign arrives their ceiling is as tight as it was a quarter ago, because nothing on their side reset. The cohorts that recover are the ones that take an action Meta counts: a read that leads to a reply, an inbound message, a Click-to-WhatsApp conversation. The quality-rating wave post covers the feedback side of the same coin; on the cap side, the reset event is engagement, and engagement only comes from sends the recipient asked for in some readable way.
The planning consequence is uncomfortable for a big-list program and good for a small honest one: the uncapped lanes (utility and authentication sends, service-window replies, free entry points) are the recovery machinery, and marketing templates are the spend you reserve for recipients who have shown they reset. Ninety days in, programs that treat the uncapped lanes as the recovery path grow their engaged cohort; programs that treat them as an afterthought watch the dormant end of the list fail campaign after campaign.
What the patterns mean for H2 2026 send planning
H2 carries the year's heaviest marketing weeks, and the cap does not pause for them. The patterns above compress into three planning rules.
First, build campaign audiences from engagement, not opt-in status. The Q4 push to the full list is precisely the send pattern the ninety-day failure tail is made of; the engaged and lapsed bands get different cadences, and the dormant band gets an entry-point re-acquisition burst instead of a template.
Second, set your own per-contact ceiling now and read the suppression count through the season. The ceiling belongs where your engaged recipients tolerate volume, observed from your delivery and read data. During peak weeks the suppression count is the dial to watch before anything Meta-side moves.
Third, keep categories honest under peak pressure. A utility template carrying promotional copy gets re-classified as marketing, and then it spends recipient budget, bills at the marketing rate under the per-message pricing model, and feeds the quality gates as marketing. The weeks where teams blur the category are the weeks the cap bites hardest, because the blurred send spends budget on recipients who never opted into a promotion.
Where Orbit sits
Every control the three patterns point at is tenant-owned configuration on Devotel Orbit; none of it claims a wire into Meta's cap, which Meta does not offer any vendor.
Cohorts and exclusion. Segments built on last-engagement traits under Audience decide who a campaign targets, and the outbound audience view includes or excludes whole bands before scheduling. The pattern-one fix lives here: lapsed and dormant contacts leave the high-frequency audience at the audience layer.
Per-contact frequency caps. The frequency caps settings are the pattern-two leading indicator: a per-contact, per-channel ceiling evaluated at send time, surfacing as suppression events in the send log per contact. The contact-level view turns "approaching the cap" into something readable on the contact profile instead of an invisible Meta property.
Pacing. Quiet hours hold sends outside each recipient's local sendable window and release inside it, which spreads a campaign across recipient-local days; a campaign spread over three evenings spends less of each ceiling per day than one released in a single burst.
Category discipline. Template category is part of authoring and submission under Channels, WhatsApp, Templates, so the pattern-three pricing and cap exposure is decided at review, not discovered at delivery.
The through-line is the same as the explainer's: the cap is Meta's counter, but the budget it counts is yours to allocate. Ninety days of logs say the operators who allocate it on their side of the line spend less of it failing on Meta's.
Frequently asked questions
Is there a published number of marketing sends per recipient to plan against?
No. Meta does not publish the threshold, and ninety days of data has not changed that; the ceiling varies with engagement and tightens on the disengaged. The workable number is your own per-contact frequency cap, set from your delivery and read data, so over-cadence sends suppress on your side before they fail on Meta's.
Does a cap-related delivery failure damage my quality rating?
They are separate counters. The quality rating grades your number on recipient feedback such as blocks and reports, while the cap limits deliveries per recipient regardless of rating. The two interact in practice: sends a capped recipient never wanted are the sends that get reported, so the failure tail and the feedback gates feed each other even though they measure different things.
Which send lanes stay outside the cap?
Utility and authentication templates as Meta classifies them, replies inside the 24-hour service window, and free entry-point conversations such as Click-to-WhatsApp ads. Ninety days in, those lanes are also the recovery machinery: they are the events that reset a capped recipient's window.
What should change in my H2 2026 campaign plan first?
The audience definition. Move lapsed and dormant contacts out of the full-cadence campaign audience before the peak weeks, set the per-contact cap from your engaged cohort's tolerated volume, and watch the suppression count weekly through the season. Cadence changes inside a bad audience move the failure tail less than fixing the audience does.
Resources
- Meta's per-user marketing template frequency cap, explained — the mechanics this post builds on.
- Tailoring WhatsApp send cadence under the cap — the operator playbook the ninety-day patterns validate.
- WhatsApp per-message pricing, decoded — the classification that prices the sends the cap counts.
- WhatsApp's quality-rating and template-pause wave — the feedback gates that run alongside the cap.
- Frequency caps docs — the tenant-owned per-contact ceiling, configured.
- Announcements hub — where Orbit logs shipped and industry updates as they land.
Published 15 October 2026.