The automated emails stopped on 10 July.
We found the line that did it.
On 10 July 2026 a single condition was added to 26 automated flows in one bulk edit. The condition asks for a
Shopify field that most subscribers do not have, so almost nobody has been allowed into those flows since.
Nothing is wrong with the emails, the domain, the deliverability or the audience. One line of setup is doing all of it.
The store has kept growing throughout, which is exactly why nobody caught it. Section 8 is honest about what that means
for the size of the loss. The last section is the part we would act on first.
Days down
62
10 July to 10 September
Flows affected
26
of 38 live flows
Flow revenue not credited
£28,379
£7,571 vs £35,950 expected. See section 8
Q4 revenue at risk
£32,651
of £65,428 flows earned last Nov–Dec
1 · What the collapse looks like
Automated flow emails sent per week. The step down lands on the week of 10 July and has not recovered in the nine weeks since.
2 · Six things that did NOT change
Before naming a cause, here is what the data rules out.
Demand
Subscribers, orders, checkouts and product views are all flat or up year on year. The people who should be triggering these flows are still arriving.
Deliverability
Campaigns went out to 671,845 people in the same window with a 0.24% bounce rate, better than last year's 0.41%. Open rate rose from 63.1% to 67.8%.
Domain and authentication
DMARC is published at p=reject with reporting, SPF and DKIM both resolve. Nothing in the DNS setup would block a flow but allow a campaign.
The emails themselves
Not one of the 188 email templates in the account was edited on 10 July. No redesign, no rewrite, no new sender address.
Blocked or skipped sends
Klaviyo logs a "Dropped Email" when it refuses to send. Drops average 10 a month, the same as always. Skipped sends since 10 July are negligible.
Consent
2,764 people granted email marketing consent in August alone, and every one we inspected is marked subscribed and unsuppressed.
Every ordinary explanation is ruled out by the account's own data. Campaign email out of the same account,
on the same domain, in the same nine weeks, is healthier than last year. Only the automated flows stopped.
3 · Demand held up while the flows went silent
August 2026 against August 2025. Grey means no meaningful change, within ten per cent either way. Every input to the flows is level or growing. Only the output fell away.
4 · The one thing that did change
We read the live setup of all 38 live flows. Twenty six of them share a timestamp and a new filter.
EDITED ON 10 JULY
26
live flows, all carrying the same new condition:
Shopify Tags does not contain "b2b"
NOT EDITED
12
live flows without that condition.
All of them are still sending normally.
The split is exact. Every flow that carries the condition collapsed. Every flow that does not carry it
is unaffected, including the transactional shipping emails, the Giftnote flows and the sunset flow. There is no
in-between case.
5 · Why one line stops the emails
The condition is not wrong in spirit. It is wrong in mechanics.
- What it was meant to do. Keep wholesale and trade customers out of the consumer flows. That is a sensible goal,
and it lines up with the B2B work done that same week: a "B2B Wholesale Sign Up" flow was created on 8 July, two days before, and the wholesale welcome flow was edited on 15 July, five days after.
- What it actually does. Klaviyo can only test "does not contain" on a profile that has the field.
A profile with no Shopify customer record has no
Shopify Tags field at all, so the test cannot pass and
the profile is never allowed into the flow.
- Who that removes. Exactly the people these flows exist for. Somebody who joins the newsletter from the pop-up
and has never ordered has no Shopify customer record, therefore no
Shopify Tags field, therefore no welcome email.
- Why the damage is uneven. Flows aimed at buyers still reach the slice of customers whose Shopify record carries
a tag. Flows aimed at non-buyers reach almost nobody. That is exactly the pattern in the table below.
The scale of it. The B2B Customers list holds 775 profiles and the TRADE list 2,284,
against a main subscriber list of 119,194. To keep a few thousand wholesale contacts out, the filter shut the door on
everyone who has never placed a Shopify order.
6 · The proof, at profile level
Thirty randomly sampled profiles. The split on one field is clean.
STILL RECEIVING WELCOME EMAILS
15 of 15
Every profile that received a welcome email since 20 July has a
Shopify Tags field. Values seen: ["newsletter"], ["Shop","Login with Shop"],
and empty []. An empty list still counts as present, so those profiles pass.
RECEIVED NOTHING
15 of 15
Every profile that joined the main list between 18 and 26 August has no
Shopify Tags field at all. All are consented, none are suppressed, and none entered the welcome flow.
One worked example: a subscriber joined the list on 24 August at 21:56, consent granted, no suppression.
Their next email of any kind was a marketing campaign on 4 September, eleven days later. No welcome email was ever queued.
7 · What it has cost, flow by flow
Recipients and attributed revenue, 10 July to 9 September, 2026 against 2025.
Same 62 days, this year against last: £30,388 of flow revenue became £7,571.
The store is about 18% bigger than it was a year ago, so the fair comparison is
£35,950, and the gap is £28,379, roughly
£3,204 a week and still running.
Two rows are not part of this story: Abandoned Cart – Evergreen was already declining before July after a
restructure in January, and the two Upstackified flows were retired last year.
8 · How big is the loss, honestly
This is the part a good finance director will push on, so here it is straight.
ORDERS, JUL+AUG
+28%
3,482 to 4,469
REVENUE, JUL+AUG
+32%
£107,976 to £142,657
NEW CUSTOMERS, JUL+AUG
+9.1%
3,031 to 3,306
- The store did not go backwards. Through the whole outage, orders, revenue and first-time buyers all grew
year on year. That is the honest counterweight to every number above, and it is why nobody noticed for nine weeks.
- So what does £28,379 actually measure? It is revenue that Klaviyo used to be
credited with and no longer is. When a welcome email does not send, an order that still happens through a paid ad,
a search or a direct visit simply stops being attributed to the flow. Some of that gap is genuinely lost sales.
Some is the same sale, credited elsewhere. The flow report cannot separate the two.
- We tried to separate them properly, and the method could not carry it. We measured the 30-day order rate
for four matched groups of new subscribers, roughly 800 people each, in the two months either side of the outage
and the same two months a year earlier. Only the bottom-right cell had the welcome series switched off.
- Read the bottom row against the top one. The group that received no welcome series at all converted at
44.1%, better than the 27.1% of the group a year earlier that received the full series. Taken literally that
says the welcome flow does nothing, which we do not believe.
- The real reason is in the right-hand column. Subscriber conversion fell 32 points
from May to August in 2025 and only 5 points in 2026. The seasonal shape of the business changed
between the two years, so the comparison has no stable baseline to measure against. We checked the obvious
explanation, that the 2026 signups were further along in a purchase when they joined, and it does not hold: the
share already shopping in the 24 hours before subscribing was 32.8% in 2025 and 31.8% in 2026.
- So we are not reporting a cohort-based figure. The swings that are nothing to do with email are several
times larger than any effect a single email series could have, at any sample size we can reach.
- What we will not do is put a precise pound figure of lost profit in front of you that we cannot defend.
The measured facts are in sections 7 and 9. The honest summary is that the flows stopped earning
£28,379 of credited revenue, an unknown but material part of that is real lost sales, and
the case for fixing it does not rest on knowing which.
Why fix it anyway. The fix costs nothing and takes an afternoon. The exposure in
section 12 is measured on flows that actually earned that money last Q4. And section 9 shows a promise being made on
every page of the site that the business is currently not keeping.
9 · The welcome code: a promise the site is still making
This is the clearest single consequence, and the one to fix first.
- Down 57% against last year, on a store taking
24% more orders. Against the two months immediately before the edit it is down
63%. No replacement code appeared, so it was not swapped out.
- There is no second delivery route. 21 live flow emails contain the code and
20 of them sit in flows edited on 10 July: every message in the welcome series and
every message in its Unveild twin. The only one outside is the wholesale welcome, a different flow for a different audience.
- The site still promises it. The signup form reads "To receive updates, offers & 10% off your first order".
Since 10 July roughly 5,200 people have accepted that offer and 393 have received anything back.
- Why no support tickets. WELCOME10 is a fixed code, unchanged since September 2024, so it circulates in old
emails and on voucher sites. Even before the outage it was redeemed 1,029 times in 62 days while
the flow was credited with about 388 orders. Enough people still find it that nothing looks broken from the help desk,
and someone promised a code who never gets one usually just leaves rather than writing in.
- A WELCOME10 order is worth £30.98 on average, median £25.39, against a store average of £33.52.
10 · Was the old baseline inflated by wholesale orders?
Asked and answered before the question comes up on the call.
- A fair question, because the wholesale channel is real. In the same 62 days last year the store took
78 wholesale-flagged orders worth £29,125, about
19% of store revenue, with single orders up to
£2,870. Order tags show Faire, Wholesale and b2b.
- But it did not inflate the flow numbers. Flow-attributed orders averaged £36 last year and
£44 this year, against store averages of £39.14 and £36.94. The old baseline ran
below store average, the current period runs above it.
- One wholesale order leaked into each period, and they cancel. £1,598 to
Nurture Flow on 20 Aug 2025, and £1,710 to
Coupon Reminder on 20 Jul 2026. Strip both and the fall is steeper, not shallower.
- The filter does not even catch them. Wholesale buyers carry Shopify tags, usually
["Faire"] and only
sometimes "b2b". In a sample of twelve wholesale orders, 4 of 12
would be excluded by a "contains b2b" test.
The proof it fails its own job. That £1,710 Faire wholesale order landed inside
Coupon Reminder, one of the 26 filtered flows, on 20 Jul 2026, ten days after the filter was added.
The filter shut out the entire prospect base and let the wholesale customer through anyway.
11 · The second cost: the list is being burned
Losing the revenue is the visible cost. Losing the subscribers is the one that carries into Q4.
- The sunset flow retires subscribers who have stopped engaging. It was not touched on 10 July and is the one flow
sending more: 1,426 emails in August 2025 against 6,541 in August 2026.
- Over the outage window it has gone from 2,595 people to 11,928.
- The reading is straightforward: subscribers who never received a welcome series never engaged, so they age into
"unengaged" and get retired. Paid media is buying email addresses that the account is then quietly retiring.
- About 5,200 people joined the main subscriber list between 10 July and 9 September.
393 of them received a welcome email. The rest are sitting on the list with no introduction to the brand.
12 · What this means for Q4
Half of last year's peak flow revenue sits in flows that are broken right now.
FLOWS EARNED NOV+DEC 2025
£65,428
across all live flows
FROM FLOWS BROKEN TODAY
£32,651
50% of the total
CHRISTMAS VARIANTS
10 clean
none carry the filter
- The good news. All ten Christmas flow variants are in draft, were last edited in December 2025 and January 2026,
and do not carry the condition. Switched on as normal they will work.
- The catch. The Christmas variants only take over in December. Last year the evergreen welcome flow carried
October and November on its own, sending 10,745 and 14,816 emails in those two months.
If this is not fixed, October and November run with the welcome series switched off.
- No seasonal cover at all. Post-purchase, winback, back-in-stock, review requests, low stock, price drop,
birthday and nurture have no Christmas twin. They stay broken until the filter is changed.
- Do not let the fix undo itself. Whoever applies the B2B exclusion again must not apply it as written, or the
Christmas variants will break the moment they go live.
13 · The fix
In order. The first item is the whole outage.
1
Change the condition, do not just delete it
In each of the 26 flows, replace Shopify Tags does not contain "b2b" with a condition that also allows a missing field: Shopify Tags does not contain "b2b" OR Shopify Tags is not set. The cleaner alternative is to drop the property test entirely and exclude the B2B Customers and TRADE lists instead, which is what it was for.
2
Decide what to do about the 10% promise in the meantime
Every page of the site offers 10% off a first order and the only thing that delivers the code is a flow that is off. Until the filter is changed, either pause that promise on the signup form or send the code by campaign to everyone who has joined since 10 July. Leaving the offer up while it goes unfulfilled is the part with brand cost attached.
3
Test with a profile that has no Shopify record
Create a test profile with an address that has never ordered, add it to the main list, and confirm the welcome email arrives within minutes. That is the exact case that fails today, so it is the only test that proves the fix.
4
Watch the volume come back within 48 hours
Welcome sends should return to roughly 2,700 a month and total flow sends to roughly 11,000 a week. If they do not, the change has not taken.
5
Recover the 62 days of missed subscribers
About 5,200 people joined the list during the outage and received no welcome series, and no WELCOME10 code. They are still on the list and still consented. Send them a one-off catch-up sequence carrying the code before the Q4 push, rather than letting them sit until the sunset flow retires them.
6
Pause or retune the sunset flow while you catch up
It is currently retiring people who were never given a chance to engage. Hold it for a few weeks, or exclude anyone who joined after 10 July, until the catch-up has run.
7
Test every Christmas variant before it goes live
They are clean today. Send one test entry through each in October so nothing else is discovered in the middle of Black Friday.
8
Put a weekly check on flow send volume
This ran for nine weeks without anyone noticing because the flows still showed as live and the configuration still looked healthy. A weekly count of flow emails sent would have caught it on day seven.
One loose end worth a look
On 1 July 2026, nine days before the outage, the account recorded 9,599 bounces in a single day,
all on the campaign side. That is a campaign sent to a stale or unverified segment rather than anything to do with the flows,
and reputation recovered: campaign bounce rate for the whole nine week window is 0.24%, better than
last year's 0.41%. It is worth identifying which send it was so it is not repeated before Q4.
Method
Everything above comes from read-only Klaviyo API pulls made on 9 and 10 September 2026.
Nothing in the account was changed. Flow configuration was read from the live flow definitions; sends, bounces, drops,
skips, subscriptions, orders and site activity from event metrics; revenue from Klaviyo's flow and campaign values
reports using the live Shopify "Placed Order" metric. Profile-level checks sampled real profiles and their event
timelines. DNS records were checked directly. Comparison windows are matched calendar dates year on year.
Sections 8 to 10 add later work from the same day: discount-code redemption read from the Discount Codes field on
every order in each window; order and profile Shopify tags read the same way; store orders, revenue and new-customer
counts taken from the Kova dashboard on the DTC basis; and subscriber cohorts built from list-subscription events and
followed for 30 days each.
Limits worth stating. Klaviyo's flow revenue is attributed revenue, so it falls when an email stops
sending even if the sale still happens elsewhere. We could not separate genuinely lost sales from re-credited sales,
because the 30-day order rate for new subscribers swings by more than thirty points between an ordinary spring month
and an ordinary summer one, which is far larger than any effect a single email series could have. Every figure here is
therefore reported as what it measures, not as profit lost.