Back to Blog
Email MarketingKlaviyoAutomation

Why Klaviyo Flows Stop Sending: 3 Months of Alerts

Andrew Beauchamp•November 18, 2026•8 min read

A Klaviyo flow can show "Live" and send nothing for days. Nobody gets an error. The revenue just stops, and you find out when someone happens to look.

So we built an alert with Claude Code. Every morning it checks every flow email across our client accounts and posts to Slack when one that normally sends has gone to zero. It has saved us money. I can't give you a clean number for how much, because the point is that we caught things before they ran for weeks.

It has also taught us that most of the time a flow goes quiet, nothing is broken. Here's what the first three months of alerts showed: why flows actually stop, which alerts mattered, and what a check like this can't see.

Three months of alerts

From August 14 to November 17, 2026, the alert posted on 48 days, about every other day. Grouped by cause, that was 20 separate incidents across 6 brands.

What it wasIncidentsWhat we did
Real outage2Fixed one the next day; the other cleared on its own
Side effect of a deliberate change2Found the cause, decided it was acceptable
Planned change10Nothing. The alert was right, but the change was on purpose
False alarm6Nothing. The flow was following its normal pattern

Two real outages in three months doesn't sound like much. But both were the kind that would have kept going: the flows still said Live, and nothing in Klaviyo flagged them.

Why flows actually stopped

1. A filter on a field the event doesn't have

This is the one that paid for the alert. A client was running a gift-with-purchase promotion and didn't want people viewing the free item to get browse abandonment emails. Their marketing lead edited the flow's trigger and added a filter: SKU is not [the gift item's SKU].

On that store, the Viewed Product event carries a product ID, not a SKU. A filter on a property the event doesn't have didn't just exclude the gift item. It excluded every event, so nobody entered the flow at all. Klaviyo saved the filter without complaint.

  • Day of the edit: 482 emails sent.
  • Next day: 213, all from people who had entered before the edit.
  • The two days after: 0 and 0. Meanwhile, browse events were arriving at above-normal volume.
  • The alert fired on the morning of the second zero day. The diagnosis took a few minutes in Claude Code, and the filter was switched to the product ID by late morning the next day. Sends were back to 482 a day within three days of the fix.

The detail that made this easy to miss: the same exclusion went into the add-to-cart flow in the same edit, and it worked there, because the add-to-cart event does carry a SKU. One flow kept running, so everything looked fine.

Lesson: after anyone edits a trigger filter, check that the flow had new entries the next day. Better yet, check that the event actually carries the property before you filter on it.

2. The trigger event stopped arriving

At the same brand, Shopify's Fulfilled Order events nearly disappeared for four days in September. Orders were normal and no flows had been touched. Our best read is that a returns-protection app's line item was getting fulfilled first, so Klaviyo only logged partial fulfillments. About 180 buyers never entered any of the six flows triggered by Fulfilled Order. It fixed itself after four days, and those entries can't be backfilled.

Only two of those six flows set off the alert. The other four were below its volume floor (more on that below).

Lesson: a flow triggered by another system's event is only as reliable as that event. If several flows go quiet at once, check the trigger metric before you check the flows.

3. A change outside Klaviyo killed a branch

A subscription brand's welcome flow split new subscribers on a profile property that recorded what they were shopping for. Only the brand's old Klaviyo signup form wrote that property. When the brand moved its signup popup to a third-party tool, a planned and announced change, nobody noticed that the new tool didn't write the property.

One branch's first email went from about 62 a day to 0 and stayed there. The flow wasn't edited, it showed Live the whole time, and the default branch kept sending, so nobody missed a welcome email. They just stopped getting the personalized track. The alert found it four days in.

Lesson: list every flow split and filter that reads data another tool writes. When that tool changes, check those flows.

4. A segment change dropped people mid-flow

A replenishment flow's trigger segment was narrowed from "ever bought" to "bought in the last 7 days." New buyers kept entering, but people who were already partway through the flow stopped getting it. In this case the dropped group hadn't been buying from those emails anyway, so we left it. It's still worth knowing that a segment edit can reach backwards into a running flow.

Most alerts aren't fires

Ten of the 20 incidents were planned: a flow swapped for its Black Friday version, a vendor's flow shut off when the client stopped using that vendor, review requests handed to the review app, a retired trial offer. The alert was right that sending had stopped. It just can't tell a swap from a failure, because it watches one email at a time.

The false alarms were stranger. 23 of one apparel brand's 25 alerts came from four post-purchase flows that trigger on Fulfilled Order and wait a fixed number of days (14 to 175) before each email. The warehouse doesn't ship on weekends or holidays, so Fulfilled Order logs zero on those days. A holiday next to a weekend makes three zero days in a row, which is exactly what the alert looks for. Memorial Day, Juneteenth, July 3 and Labor Day each set off a wave of alerts weeks or months later, as the delays played out. Every time, the flow was sending again the morning the alert posted.

Thanksgiving and Christmas will do the same thing into the spring.

How the alert works

  • Runs daily on a schedule in AWS, across 16 client accounts and roughly 650 flow emails per run.
  • Pulls delivered counts per flow email from Klaviyo's reporting API for two windows: a 14-day baseline that ends 4 days ago, and the last 3 full days. Offsetting the baseline keeps an outage from dragging down its own baseline.
  • Fires when an email's baseline is at least 10 deliveries a day and the last 3 days are at or below 2% of what the baseline predicts. In practice that means zero, or one or two stragglers.
  • Watches deliveries, not revenue. Revenue swings with a few big orders. A delivery count that drops to zero is unambiguous.
  • Posts to Slack with the client, flow, email, the last-3-day count and the baseline. It won't flag the same email again for 7 days.

What it can't see

  • Low-volume flows. Anything under 10 sends a day is skipped, which is why four of the six starved flows in September never alerted.
  • Partial drops. A flow that falls 70% won't trip a check that looks for zero.
  • People who never enter. At one brand, about half of trial buyers weren't entering the trial flow. Volume looked steady, so nothing fired. We found it by matching orders to flow entries.
  • SMS. It only watches email steps.
  • Lag. It needs three near-zero days, so the earliest alert lands on the fourth morning.

Build your own

If you have Claude Code connected to Klaviyo, you can build the same check in an afternoon. Start with something like this:

Build a script that runs every morning. Use Klaviyo's flow values report to get delivered counts for each flow email over two windows: a 14-day baseline ending 4 days ago, and the last 3 full days. Flag any email whose baseline is at least 10 deliveries a day and whose last 3 days are at or below 2% of the baseline rate times 3. Post the flagged flows and emails to Slack with both numbers, and don't flag the same email again within 7 days.

Based on what the first three months showed us, it's worth asking for a few more things up front: compare against the same weekday so weekly patterns don't trip it, check the trigger metric's own volume when several flows go quiet together, and keep a list of planned flow swaps so those alerts can be skipped.

If you'd rather have ours set up on your account, book a call and we'll walk you through it. For the flows themselves, our guide to Klaviyo flows covers what should be running in the first place.

FAQ

Why did my Klaviyo flow stop sending?

In our first 20 alert incidents, the real causes were a trigger filter on a property the event doesn't carry, a trigger event that stopped arriving from Shopify, a profile property that a new signup tool no longer wrote, and a segment edit that dropped people mid-flow. Half of all incidents were planned changes, like a flow swapped for its Black Friday version.

Can a Klaviyo flow show Live and not send anything?

Yes. In our data, a browse abandonment flow showed Live and sent zero emails for two days after a trigger filter was added on a property the event didn't have. Browse events kept arriving the whole time.

How do I know if a Klaviyo flow stopped sending?

Compare each flow email's deliveries over the last few days with its normal daily volume. Our alert flags any email that normally sends at least 10 a day and sent close to zero over the last 3 days, and posts it to Slack each morning.

Why do Klaviyo flows go quiet after holidays?

Flows triggered by Fulfilled Order follow the warehouse's schedule. If nothing ships on a holiday weekend, every fixed-delay email in those flows goes quiet for the same days, weeks or months later. Nothing is broken, and the flow picks back up on its own.

Want to know if any of your flows have quietly stopped? We'll check every flow in your account against its normal volume.

Methodology

  • Source: every alert our flow monitor posted to Slack from August 14 to November 17, 2026, across the client accounts it covers. Brands are kept anonymous.
  • Incidents: 48 alert posts flagged 107 flow emails. We grouped them by client, flow and cause into 20 incidents across 6 brands.
  • Causes: taken from Klaviyo flow data (daily deliveries per email, flow status, edit timestamps), trigger event volumes, and our own notes and client emails at the time. Where the cause is our best read rather than confirmed, the post says so.

Want results like these for your brand?

We help ecommerce brands build email and SMS programs that drive real revenue. Let's talk about what we can do for you.