trial_ended last month and nothing emails them”, not just “send more email”.
Checks run for workspaces with more than 1,000 active subscribers that sent at
least one email (campaign, sequence or transactional) in the last 14 days. A
workspace that stops sending is skipped until it sends again.
What it checks
The rules decide what is a problem. The AI step ranks findings, writes a
short summary for your business and sharpens each recommendation. It can leave
a finding out (an event it traces to an active sequence, an event that is
routine product activity, a draft not worth sending). The only findings it adds are missing sequences, described below.
Campaign drafts and new missing sequences need its judgment, so they only appear
in checks where the AI step ran. A missing sequence it suggested earlier stays
in a check without it, until something covers it or you close it.
Missing sequences
The AI step also compares the sequences you run with what businesses like yours usually run, using what your workspace says it does, the campaigns you sent in the last 30 days or have scheduled, your sequences and their emails (including drafts), how many contacts joined in the last 30 days and the events you send. When your emails and your workspace description disagree, it goes by your emails. It suggests up to three missing sequences, most important first, each with why you need it and a plan: the trigger and its emails (usually two to five) with their timing. Before it answers, the check can look things up: read one of your campaigns or sequence emails in full, read a page of your own website, and find how similar brands in the email gallery run the sequence it is about to suggest. It only reads, makes at most six lookups per check, and only uses pages on your own domain (a page that redirects elsewhere is skipped). What it finds shapes the plan’s emails and timing. It picks from standard playbooks (welcome, onboarding, lead nurture, trial conversion, re-engagement, win-back, post-purchase, abandoned cart, failed payment, upsell, feedback, referral, event follow-up, and launch or webinar sequences timed from registration). A suggestion starts when a contact is added, on an event your workspace already sends, or by manual enrollment for people already on your list, such as a re-engagement sequence. Rules then drop anything an active sequence already covers and any event that already has its own finding. A suggestion you dismiss is never made again, and one clears by itself once an active sequence covers it.How often people hear from you
Two findings look at email from each subscriber’s side. Both count marketing email only: campaigns, sequence emails and marketing Send API emails. Transactional email such as receipts and password resets, and test sends, never count. Too many emails per person counts, for each subscriber, the most marketing emails they got within any 24 hours over the last 7 days, the same rolling window as the frequency cap. It names the campaigns, sequences and Send API email that sent those people the most and says what to change. A sequence set to Unlimited Re-entry that took the same person in 3 or more times that week comes first: switch it to One-time Only, or end it with a wait so nobody re-enters while still in it. Otherwise, turn on the frequency cap if it is off and switch it on in the sequences that can wait, such as tips and nudges, or space out campaigns, which never wait for the cap. The cap does not limit marketing email sent through the Send API, so when that is behind it, the finding tells you to space those emails out in your app. It is high severity when 1,000 or more people, or 10% of everyone you emailed, are affected. Subscribers left out counts active subscribers with an email address who joined over 90 days ago and got no marketing email in the last 90 days. People who go months without hearing from you forget they signed up and are more likely to mark your next email as spam. On lists larger than 2,000 such subscribers, the number is estimated from a sample of 2,000 and shown rounded. When nobody, new subscribers included, got a marketing email in those 90 days, this finding stays out. It also stays out while a No recent campaigns finding applies, even if you dismissed or snoozed it, since that finding already asks you to email your list again. If you leave these people out on purpose, dismiss it.Signups that never engage
Signups that never engage looks at the newest 5,000 active contacts who joined between 7 and 60 days ago. It groups them by consumer mailbox provider (Yahoo covers yahoo.com, att.net and the other addresses Yahoo hosts) and by the list each one joined first. Addresses at a business’s own domain are never reported, since their mail servers often hide opens; they only count towards the comparison. A group is reported when at least 100 of its people got 2 or more emails each and 1% or fewer of them opened or clicked any, while at least 8% of your other new signups did. It compares against your other new signups so a workspace whose mail as a whole goes unopened is not blamed on one group. A group like this usually comes from a bot submitting real people’s addresses through a signup form, or from an old or bought list. Mailbox providers read email nobody opens as unwanted, so sending to them pulls down where the rest of your email lands. Turn on bot protection and double opt-in for the form that feeds the list, then unsubscribe the people in the group who never opened. It is high severity when 1,000 or more people are in the group, or when it makes up a quarter of your recent signups. At most 2 groups show per check, the largest first. If you know they are real people, dismiss it.Which events are reported
A product that tracks every action has dozens of events, and most do not need their own sequence. An event each person fires 10 or more times on average (page edits, messages, generations) is product activity, so it is never reported, except failed payments, cancellations, downgrades and events with $1,000 a month of revenue behind them. Of the rest, the AI step picks up to 5 that mark a moment worth a sequence, such as a purchase, a cancellation, finishing setup or an error that blocks someone. It leaves out sign-ins, each step of a flow whose finish is also tracked, and verification or invite emails. Events with a drafted fix, revenue at risk, or $1,000 a month behind them always stay, even past 5. In a check without the AI step, only the events already shown stay (and one whose handling sequence was turned off), so events an earlier check left out do not come back (on a first check, the 5 with the highest priority show instead). Events you dismissed or snoozed never take one of the 5 places. An event finding is high priority with $1,000 a month of revenue behind it, medium with 250 or more subscribers or when it puts revenue at risk (failed payments, cancellations, downgrades), and low otherwise. Within the same priority, findings with more revenue behind them come first, then sending problems (a failing domain, a bounce or complaint spike, signups that never engage, too many emails per person), then suggestions.How “nothing reacts to it” is decided
An event counts as handled when something ends up contacting the person, directly or down a chain. The check follows each event through:- sync rules that tag the contact (
saas.payment_failedaddspast-due); - sequences that start on that tag, or on a segment the tag or the event puts the contact in (a segment counts when that tag or event alone is enough to enter it);
- the tags and lists those sequences add, and the sequences that start on them, up to three sequences deep.
Revenue behind a finding
When a payment integration syncs revenue to your contacts (theirmrr
attribute), an event finding also says how much the people behind it pay you
each month, for example “$3.1k/mo at risk" for `saas.payment_failed`. Events
that put that revenue in danger (failed payments, cancellations, downgrades)
say "at risk". $1,000 a month or more makes a finding high priority, and among
findings of the same priority the one with more revenue behind it comes first.
The amount is in evidence.monthlyRevenue (USD) in the API.
If the sequence that would catch the event is paused or a draft, the finding
offers Turn it on and says how the event reaches it, for example “It
starts when this event adds the past-due tag.” The AI step then
double-checks the rest against your sequences, their triggers and what they
do, and leaves out an event it can trace to a specific active sequence. That
verdict is kept while the sequence stays active, so a check without the AI
step does not bring the finding back.
Where you see it
- Dashboard. The Account Health card sits at the top of your dashboard: Swirl’s summary, the most important finding and its fix button. See all opens every open finding, most severe first. Findings about a sequence, campaign or domain link to the screen where you fix them. Run now checks immediately; progress shows on the card and the list updates when it finishes.
- Email. After each scheduled check that finds something, the findings reach everyone who has the check on (see In your weekly report). The standalone email starts with your sending stats for the period against the one before: emails sent, open, click and bounce rates. Every link opens the full list on your dashboard, scrolled to the finding it is about.
- API, CLI and MCP. See Get Account Health,
sequenzy account-health, and theget_account_healthMCP tool.
Drafted fixes
With Draft fixes on (the default), Swirl may lay out up to two draft sequences per check, for example a sequence triggered by an event nothing reacts to. Swirl names each draft for your business, so you can tell what it does from your sequence list, and lays out the trigger, the timing and one titled email step per message, with a short brief for each about your product. The check itself never writes the emails: they are written with AI the first time you open the draft, so you land on emails to review rather than empty steps. Review them, then turn the sequence on yourself. A sequence with an empty email step cannot be turned on. The agent never drafts the same fix twice, even if you delete the draft. When you click Build this sequence yourself, the emails are written too, the same as a sequence from a template. Review them, then turn it on.Fixing findings in one click
Every finding comes with buttons that do the fix, not just describe it:
Anything that changes live sending asks you to confirm first. When a button
fixes the finding, it is marked fixed and Swirl, the account health snail,
celebrates. The same actions are listed in the API response, so agents can
run them too.
You can also triage from the menu on each finding:
- Mark as fixed clears a finding. If the next check still finds it, it comes back.
- Snooze hides it for a week or a month.
- Dismiss hides it for good.
Was this useful?
Each finding asks Was this useful? with a thumbs up and a thumbs down. Your answer goes to the Sequenzy team so the suggestions get better, and it changes nothing in your workspace. After a thumbs down you can say what missed; after a thumbs up you can add a comment too. Everyone on the team, viewers included, answers for themselves, and you can change your answer at any time. From the CLI,sequenzy account-health feedback ins_123 not_useful --comment "We already send this from Stripe". Through the API, use
Send Account Health Feedback. With
MCP, your agent can pass on what you said about a finding with
submit_account_health_feedback.
In your weekly report
Checks run early Monday (05:00 UTC), a few hours before the Monday weekly report. Swirl’s summary and the most important fixes, each with its button, then appear as an Account health section inside that report, so you get one email instead of two. If you do not get the weekly report (it is off for you, the workspace was not in it, or the week was too quiet), the findings arrive as their own email a few hours later. Either way you get each check once, and only when something needs fixing. Who gets it is each person’s own choice, in Settings → Notifications under Account health check: on by default for the workspace owner, opt-in for everyone else. Turning it off stops both the report section and the separate email. With the CLI:sequenzy notifications set --account-health off.
Settings
Checks run every Monday at 05:00 UTC, a few hours before the weekly report. The AI cost is on us, so there is no budget to manage. Owners and admins open Settings from the … menu on the card:- Weekly check: turn scheduled checks on or off. Run now works either way, at most once every 10 minutes. After a check that found something to fix, it waits a day: work through the findings first.
- Draft fixes: let Swirl lay out inactive sequences for the gaps it finds. Their emails are written when you open them; you review them and turn them on.