Why Business Automations Break and How to Notice
By CodexierPublished 6 min read
An automation that worked perfectly at launch will break at some point. That is not a sign it was built badly; it is a consequence of connecting systems you do not control. The question is whether you find out within an hour or after three weeks. This guide covers the usual causes and a monitoring setup a small team can actually run.
The short version
Expired logins and tokens
Most automations connect to other systems through an access token, not a password. Tokens expire by design. Some last an hour and are refreshed automatically; others are refresh tokens that stop working if they are not used for a period, or when the person who authorised the connection changes their password, leaves the company or loses admin rights. A Fortnox, Google or Microsoft connection authorised by an employee who has since left is one of the most common reasons a flow dies on a Monday morning.
- Authorise connections with a shared service account, not a personal login, where the system allows it.
- Write down which account authorised each connection and when it was last renewed.
- Put a recurring reminder in the calendar for connections that must be re-authorised manually.
Changed fields and formats upstream
The second classic is a change nobody thought of as technical. Someone renames a field in the contact form, adds a new product category, switches the date format in a spreadsheet or introduces a new VAT code in the accounting system. The automation still runs, but it maps data to the wrong place or skips rows it no longer recognises.
| Change | Typical symptom | How to catch it |
|---|---|---|
| Form field renamed or removed | Leads arrive in the CRM without phone number or message | Required-field check that alerts on empty values |
| New option in a dropdown | Records routed to a default or dropped | Alert when a value does not match a known list |
| Spreadsheet column moved | Wrong values in the wrong fields | Validate by header name, not column position |
| New VAT code or account | Invoices booked incorrectly or rejected | Reconcile totals against the source weekly |
The fix is rarely technical: agree that changes to forms, sheets and account plans go past whoever owns the automation first.
Rate limits and volume spikes
Every API limits how many requests you can make per second or per day, and most automation platforms cap tasks or operations per month on your plan. A flow built and tested with ten orders a day can fail when a campaign brings in three hundred. Requests get rejected, the platform pauses the scenario, or you simply run out of tasks halfway through the month. Well-built flows queue work, retry with a delay and warn you before the monthly quota runs out.
Silent failures vs loud ones
A loud failure is the good kind: the step errors, the platform sends an email, someone fixes it. A silent failure is when every step reports success but the result is wrong or missing. The trigger stopped firing because a webhook was deleted. A filter now excludes everything. The flow writes to an old sheet nobody reads. These can run for weeks, and the only protection is checking outcomes rather than errors.
Count check
Compare how many items entered with how many arrived. Twenty orders in the webshop and twenty in Fortnox is a pass; twenty and fourteen is an alert.
Heartbeat
A flow that normally runs daily pings a monitor each time. No ping within the expected window means something upstream stopped.
Spot check
Once a week, a named person opens three random records at both ends and confirms they match.
A monitoring setup for small teams
You do not need an operations team. You need a short list, a place where alerts land and a habit. This is the minimum we set up on every workflow automation we deliver, and what we recommend checking before any flow goes live (see our automation go-live checklist).
- An inventory: every flow, what it does, which systems and accounts it uses, and who owns it.
- One alert channel, such as a shared inbox or Teams/Slack channel, that more than one person watches.
- Error notifications turned on in the automation platform and sent to that channel, not to a former employee.
- A daily or weekly count check for every flow that moves money, orders or customer data.
- A run log kept long enough to replay what failed once the cause is fixed.
- A quarterly review: remove flows nobody uses and renew connections that are about to expire.
When you do not need outside help: if you run two or three simple flows, such as form to email or new order to a spreadsheet, the platform's built-in error emails plus a weekly glance are usually enough. Paying for a monitoring build makes sense once flows touch bookkeeping, customer data or anything you would struggle to reconstruct. If you are unsure which category you are in, book a short call and we will tell you honestly.
Frequently asked questions
How often do automations actually need fixing?
It depends on how many external systems they touch and how often those systems change. A flow between two stable systems can run untouched for a long time; one that depends on forms, spreadsheets and several APIs will need attention whenever any of them changes.
Is Zapier or Make more reliable than a custom integration?
Neither is inherently more reliable. Platforms give you retries and error emails out of the box; custom code gives you full control over validation and logging. What decides reliability is whether someone monitors the outcome. Our comparison of Zapier, Make and n8n covers the trade-offs.
Who should own an automation in a small company?
The person who owns the business process, not necessarily the person who built it. They know when the result looks wrong. The builder or supplier supports them, but the owner watches the alerts and approves changes to the systems it depends on.
Can AI automations fail in different ways?
Yes. Besides the usual token and field issues, an AI step can return a plausible but wrong answer or change behaviour after a model update. Those steps need sample-based checks of the output, not just error monitoring.
Want a second pair of eyes on your flows?
Tell us which automations you run and what they touch. In fifteen minutes we can point out the weak spots and whether monitoring is worth building.
Book a free 15-minute call