11 actions, ranked, every one traced to a measured finding in the KPI pack. Sizes are measured, not estimated, except where the item says otherwise.
These are cheap and the evidence is unambiguous.
Measured from Shopify order records against each variant's unit cost: discount depth went 26.5% to 35.3% between Jul 26 and Sep 26, and gross margin fell 75.0% to 72.4%. On Sep 26's revenue those extra points are $207,798 of margin given away in one month. Revenue grew over the same period, so the top line reads well while the margin behind it does not. This is the same shape as campaign revenue per recipient falling while reach rises: more volume bought at a worse rate.
3 flows were earning last window and sent nothing this one: Abandoned Cart 2026 [Retention] (email, $5,099); [BFS][SMS] Abandoned Checkout (sms, $4,645); Browse Abandonment (2026) [Retention] (email, $2,216). A further 4 carry send steps and short delays yet have produced nothing in 30 days, worth $3,213 the month before: [BFS] Generic — Reorder Flow (legacy); NMN NAD+ LP Ebook Orders; [BFS] Post-Purchase Educational — NMNH; Berberine LP - 1 Day Detox Ebook Purchasers. This is not a performance problem, it is machinery that stopped. Five flows produce 60% of all flow revenue, so one broken trigger is a material revenue event.
Across the twelve months to Sep 26, campaign recipients went from 2,007,141 a month to 2,302,178 while sends to 20,000 or more recipients held at 10 to 13 a month, and revenue per recipient went from $0.348 to $0.173, -50%. Open rate fell from 43.6% to 37.2% over the same span. Total campaign revenue still rose because reach grew faster than the per-recipient return fell; reach is the only thing holding the line. This is the largest measured trend in the pack and no other item on this list changes it.
The April to June re-engagement worked: 7,660 people opened or clicked. Only 816 of them read as engaged here, because the suppression segment is built on this account's own open and click metrics and the backup account has its own. 6,808 people who demonstrably came back are still being suppressed from every send. Sampling both groups confirms the mechanism: of those released, 88% had opened an email in the main account and 32% had placed an order, so a purchase does release them automatically. Of those still held, none had done any of the three.
5,463 of July's 6,256 SMS opt-outs were people replying STOP. 2,564 of them came from the Independence Day series alone: six sends in six days, with the same message split into separate morning and afternoon versions. August repeated the shape with Black Seed Oil split three ways. The list has shed 10,531 people since the migration and this is the largest single cause.
Postscript stays installed on purpose for manual campaign sends, and that part behaves: 4,269 of the 4,659 link clicks since 1 July are Campaign. But 372 are Automation, and every automation still firing is one Klaviyo also runs: Browse Abandonment, Abandoned Cart, Welcome Series, Abandoned Checkout. Those are clicks, not sends, so the send volume behind them is several times larger, and the same person can sit in both. This count covers 1 July to 2026-08-28, which is where the Postscript click export stops; it is not evidence about the days since. There are also 17 pop-up clicks, so a Postscript pop-up is still live somewhere.
Bigger builds, all with a measured size attached.
9,049 people tried to opt in to SMS between 21 June and 18 August and 70% never landed consent. 4,830 of them never had a number captured at all, so no phone-based segment can see them. These people walked up and pressed the button, which is a stronger signal than anything else in the recoverable pool.
Total profiles, never-consented, suppressed-for-bounces and suppressed-for-complaints are not readable from this account today, so the gap between the list we have and the list we can send to cannot be stated. The segments that look like they would answer it do not. EVERYONE has a definition identical to CAN RECEIVE EMAILS, both being can_receive_marketing true, so reading it as a total reports the mailable count as the denominator and makes the gap vanish. Bounced, HARD BOUNCES and Spam all count metric events rather than suppression state, and an event is not a suppression. Reading segment counts is cheap once they exist: eight profile_count calls return in about thirty seconds.
18 of the 26 reorder flows live in both windows fell, median -27%, together $17,948 down. The group total looks nearly flat only because a few risers offset them, so the headline hides the breadth. Reorder, Skip-Save and Subscribe & Save are 63 of the 90 flow rows and 26% of flow revenue, at $0.394 per recipient against $0.976 everywhere else. The flows still sending reached +10% more people for -23% per recipient against the prior 30 days, so the block is being sent harder and returning less. The cohort is colder than an abandoned-cart audience by nature, so it will never match, but the direction is the campaign problem repeating inside the flow estate.
Not tasks yet; the answer changes what we build.
Anyone with no open or click in 180 days is already excluded from most sends, so the fall is not dead weight dragging an average down. It is happening among people who were engaged, which points at content, frequency or inbox placement.
Checked, not assumed: 9,581 of the 11,428 profiles on the TikTok affiliate list hold email marketing consent of SUBSCRIBED, timestamped to the 2026-04-07 upload itself rather than to any form submission. The other 1,847 have since unsubscribed. So they are consented in Klaviyo's eyes, they are in the mailable base, and they sit inside every rate on the KPI page.