The Automation Was Never Broken. The List Was.

HubSpot workflow errors mostly count the platform refusing to mail bounced and unsubscribed addresses. Read the log by cause before calling it broken.

· Updated November 19, 2026 · 5 min read

The first thing a HubSpot portal tells an auditor is how many HubSpot workflow errors it has. On the one we opened in September the answer was 22, out of 70 switched on, which made nearly a third of the automation look like it was failing. Our first report said as much, and went further. It said both of the company’s email nurture workflows were broken, on a portal whose whole business problem was that inbound had gone quiet.

That paragraph was wrong in a way that’s worth taking apart, because it’s the way most people read that badge.

What was the error count actually counting?

It was counting two things, and neither of them was “broken automation.”

The 22 was a lifetime figure. HubSpot records the last execution error on every workflow, and among the 22 enabled ones the earliest reached back to 2022. There’s nothing on the badge to tell an error from four years ago from one this morning. Parsed by date, seven enabled workflows had errored in the last 30 days, two of them on the day of the audit.

Four of the seven were property-copy loops, and a fifth was a smaller copy workflow. The busiest, a workflow that copies the company name onto the primary contact, had enrolled records 729,866 times, 8,634 of them in the previous seven days, in a portal with fewer than 5,000 contacts. It was erroring because it was thrashing, and it was being deleted anyway, one of 23 workflows that existed only because HubSpot can’t read a value across an association.

That left the two that mattered, the welcome sequence and the nurture sequence that follows it. Here the correction is the whole piece. HubSpot’s automation-issues view listed the welcome sequence’s errors by cause: six contacts with bounced addresses and five contacts unsubscribed from that subscription type. 1 Editor's note 1 A read-only service key can’t see the automation-issues view, because HubSpot’s service-key catalog has no automation scope. An audit built from a key alone can’t read workflow health at all. That’s eleven suppressed sends. Every one of them was HubSpot doing its job, refusing to mail someone it had already learned not to mail.

Across the portal the pattern held. Of 13 issues flagged, eight were send refusals: bounced, unsubscribed, quarantined, no address on the record, an invalid address, a non-marketing contact. The other five were record writes: two that couldn’t find their target, one copying from an empty field, one creating a record with invalid properties, and one trying to move a lifecycle stage backward. Not one was a sequence that had stopped sending to people it should have sent to.

The automation was not broken. The list was.

A badge nobody reads is how a portal decays

The distinction matters because the two diagnoses lead to different work. “Both nurture workflows are broken” sends a marketer into the workflow editor looking for a bad step. “Eleven of your welcome enrollments were bounced or opted-out addresses” sends her to the list, and it says something about every list in the portal: that addresses had been bouncing and unsubscribing for years, that the platform had been absorbing it, and that the one signal it raised about this was a number in a corner of the screen that a lean marketing team had no reason to open.

Every one of them was HubSpot doing its job, refusing to mail someone it had already learned not to mail. The automation was not broken. The list was.

A badge that counts makes a poor monitor. It only goes up, and once the number stops meaning anything, it stops being read. The errors were real. They’d been filed as background by everyone who walked past them.

The second example from the same portal had no badge at all.

When we pulled the images the company’s marketing emails pointed at, 197 distinct files on HubSpot hosting, 76 of them were already gone. Every URL form we tried returned nothing. They weren’t in the File Manager, the 30-day trash or the deletion history, which only reaches back to August 2026. The most-used of the missing files was the company’s own logo, referenced by 180 emails. With those images gone, 365 of the portal’s 442 marketing emails were rendering with broken images on the day we read them. Of those, 15 were in sends that were still automated: an account-based sequence with its A/B variants, and two surveys.

There was no badge and no error, because as far as the platform is concerned, a send with a dead image is a successful send. The company found out only because our archive pull during the exit counted the 404s.

Read the log. Do not count it.

The lesson for anyone running a HubSpot portal, leaving or not, has two halves.

The first is about reading. The number on the badge is useless; the list behind it isn’t, because every row names a cause. A few minutes in the automation-issues view, sorted by date, turns “22 broken workflows” into four copy loops, eleven bounced or opted-out addresses on one sequence, a promotion that tried to mail people who had opted out, and one stage-change rule fighting its own settings. Do that quarterly, and read it as a list of reasons rather than a score.

The second is about what the reasons are telling you. Eleven suppressed sends on a 25-enrollment welcome sequence are a list-hygiene finding wearing an automation costume. The portal was billing for 1,401 marketing contacts, and 1,157 of them could be sent to. The gap isn’t a HubSpot problem, and it doesn’t go away by leaving. It goes with you, into the first campaign on whatever platform comes next, unless the suppression state is loaded before the first subscriber. We’ve written separately about why marketable is not subscribed and about the suppression history a HubSpot export won’t give you; this piece is the reason both of those are day-one requirements of the new platform rather than cleanup tasks for the second month.

What the company gets on the other side is a different rule in the system being built to replace the portal. A failure lands in a queue a person can open and retry, and anything that needs attention goes to a shared channel the team can see. One person can’t be the monitoring system, and silence stops being the success signal. The migration guide covers where the list work sits in the sequence of a move, and the short answer is first.

The research is free. So is the diagnosis. 60 seconds →

Notes & Sources

1Editor's note: A read-only service key can't see the automation-issues view, because HubSpot's service-key catalog has no automation scope. An audit built from a key alone can't read workflow health at all.

Frequently asked questions

Often they mean the list has decayed, not that the workflow is broken. In one portal we audited, HubSpot flagged 13 automation issues. Eight of them were send refusals: the platform declining to email addresses that had bounced, unsubscribed, been quarantined, or had no valid address on the record. The welcome sequence's two flagged issues were 6 bounced addresses and 5 unsubscribed contacts. The sends were being correctly suppressed. The remaining five issues were record writes failing, four of them in workflows that existed only to copy a value onto a second record and were being retired anyway.
No. The count accumulates. The portal we audited showed 22 enabled workflows with execution errors, and that figure reached back to 2022. Parsed by date, 7 enabled workflows had errored in the last 30 days, and four of those were property-copy loops thrashing on re-enrollment, one of them with 729,866 lifetime enrollments in a portal of under 5,000 contacts. The number on the badge said nearly a third of the automation was failing. The dated view said two email sequences and some copy loops.

The next analysis lands Sunday. Get the Briefing →

Before the badge becomes the migration

Find out what your portal is actually telling you

We read the issue log, the suppression state and every automated send before the exit is scoped, so the list is clean on the day the new platform goes live, not six weeks after.