The failures that look like success.
A tag deployment drops the snippet and audiences quietly stop growing. An ad token expires and the ad set keeps spending against a customer list frozen the day it died. A relay password is rotated and the failed column climbs where nobody looks. Every one of these is invisible until somebody is told — so SegmentHub tells them.
CRITICAL · Ad activation failed
Meta Ads — VIP Customers
The network replied: token has expired. Last successful sync 3 days ago.
WARNING · Ingest volume drop
purchase — down 84%
61 events this hour against a median of 380 for Fridays at 14:00.
RECOVERED · Email sending
Relay accepting mail again
Ingest volume drop
An hour of event volume compared against what that hour is normally worth. Watch the whole account to catch a tag that stopped firing altogether, or name individual events to catch one broken integration among six.
Ad activation failure
The current state of every audience and ad network pair. A failed sync is critical and quotes what the network actually replied — normally the only thing that says what to fix. A high rejection rate is flagged too, because it usually means a format mismatch rather than an unmatchable list.
Sender failure
The failure rate of email, push and SMS sends since the last sweep — minutes, not a running daily average. Two million clean sends yesterday would bury four hundred consecutive failures this morning in a number nothing could move.
A baseline that survives a normal week
Being paged every weekend is how alerting gets switched off. The comparison is built to avoid that.
The same hour, on the same weekday
Traffic has a strong weekly shape. Comparing a Sunday morning against an average that is five-sevenths weekday invents a drop every weekend. The baseline is the median of that same hour on that same weekday over the previous four weeks.
Median, not mean
One week containing an outage or a Black Friday would drag an average far enough to either hide a real drop or invent one. A minimum-volume floor applies to the baseline, so an 80% fall on an hour that normally carries four events stays what it is: a quiet Tuesday.
Zero is recorded as zero
An hour in which nothing arrived is written down as an hour in which nothing arrived. Without that, a total outage and a quiet night look identical — and a total stop is the headline alert, the one that must never be the one that gets missed.
Severity from your own threshold
A 20% failure rate is critical to somebody who asked to hear about 10% and barely news to somebody who asked about 50%. Severity is measured against the number you set, and anything reaching total failure is always critical.
One alert per broken thing
A token that expired on Friday is one alert with a rising last-seen time and one notification — not eight hundred rows and eight hundred emails.
Cooldowns and quiet hours
A continuing problem refreshes its evidence silently and re-notifies only after the cooldown. Notifications inside quiet hours are held rather than dropped — and a recovery notice is never held.
Recoveries announced
When a sync succeeds or sending resumes, the alert closes itself and says so — but only if you were told about it in the first place. Telling somebody that a thing they never heard about has recovered is pure noise.
"Not enough evidence" is a real answer
A sync still running, or a channel that attempted nothing this sweep, is left exactly as it is. Treating that as healthy would close an open alert because the failing thing stopped being attempted — the most misleading possible moment to declare a recovery.
Delivery that does not run through what it watches
One of the things this module watches is whether your own message senders are working. Asking a relay that has started refusing mail to carry the message saying it has started refusing mail produces silence exactly when it matters most.
Alert email leaves over the platform's own relay, never the bring-your-own relay your campaigns use.
The webhook is a direct call, not a rule inside the webhooks module — an alert must not depend on the machinery it reports on. The payload is Slack-shaped by default, which Mattermost, Rocket.Chat and Google Chat all accept, with a plain JSON alternative for anything else.
A notification that failed is retried, not swallowed, and the delivery error is shown on the alert — which is how somebody finds out their webhook has been returning errors for a month.
Your webhook address is validated when it is saved and again when it is called, with redirects never followed.
Alert history
Meta Ads — VIP Customers
Activation failure · opened 3d ago
purchase — ingest drop
84% below the Friday 14:00 median
Email sending
Resolved automatically after 41 min
Google Ads — Cart Abandoners
31% of records rejected
Rules ship paused, so switching the module on never floods a new account. Illustrative interface values.
Find out in minutes, not in the quarterly review.
Alerting covers the whole platform — ad activations, email, push, SMS and the event stream that feeds all of them.
Book a Demo