Skip to content
Back to Articles

Articles

How to monitor failed business automations

Written by Roberto Postolovski, Founder & Tech LeadUpdated

Loud failures and silent failures

A loud failure throws an error: an API rejects a request, a record fails validation. These are easy to catch if you log and alert.

A silent failure produces no error at all. The automation simply stops receiving work, and everyone assumes things are quiet. These are the ones that cost money, because nobody notices until a customer asks where their order is.

Where silent failures come from

  • Expired subscriptions: Microsoft Graph change-notification subscriptions expire and must be renewed, and a Gmail watch must be renewed at least every 7 days
  • Disabled webhooks: Aircall disables a webhook after 50 consecutive failed deliveries
  • Missed triggers: Microsoft notes the Outlook connector's trigger can occasionally miss emails when many arrive at once
  • Revoked credentials: a password change or expired token stops access without anyone being told
  • Changed screens: screen-automation (RPA) steps stop finding the button they click

What to monitor

  1. 1.Every run

    Start, end, items processed, items failed. A run log you can search by date and by item.

  2. 2.Every item

    What arrived, what was extracted, what was written where, and who approved it. This is also your audit trail.

  3. 3.Errors

    Alert a named person the same day, with the item and the reason, not just a stack trace.

  4. 4.Heartbeats

    If a mailbox usually receives orders every day and nothing arrived since yesterday, that is an alert, even though nothing failed.

  5. 5.Subscriptions and webhooks

    Check daily that every subscription, watch and webhook is active and not about to expire.

  6. 6.Queues

    Alert when items wait longer than they should, or when the retry queue keeps growing.

Reconcile regularly

Monitoring tells you what the automation saw. Reconciliation tells you what it missed. Compare the source with the result on a schedule: emails in the mailbox versus emails processed, calls in the phone system versus calls logged in the CRM, invoices received versus invoices posted. Every gap is either a bug to fix or an item to process.

Make failures easy to recover from

  • Keep failed items in a review list where a person can fix and re-run them
  • Make every step safe to repeat, so a re-run doesn't create duplicates
  • Write down, per automation, who gets the alert and what they should do

Sources

Related

Have a process like this?

Tell us how it runs today. We'll tell you whether it can be automated, what it would take, and whether the numbers make sense.