On this page
TL;DR: "Set it and forget it" is half right. A good automation should not need anyone day to day, but it depends on logins, plans and forms you do not control, and when one of those changes it usually fails quietly. Give every automation an owner, a failure alert and a short monthly check, and it keeps paying you back.
The pitch for automation often ends with "and then you never think about it again." The first half of that promise is fair. Once a workflow runs, nobody should be copying data or chasing a reminder by hand. The second half is where small businesses get burned. An automation is a small system, and every system needs someone watching it.
Why do automations stop working when nobody touched them?
Because they lean on things you do not control. Every automation connects to other tools through a login, an API or a form, and each of those can change underneath it.
- Connections expire. Google's developer documentation lists the reasons a saved authorization can stop working, including a refresh token that has not been used for six months and, for connections with Gmail access, the user changing their password. Apps still in testing mode get tokens that expire in 7 days. Source: Google for Developers, OAuth 2.0.
- Tools switch work off for you. Zapier's help center says it automatically turns off a Zap that errors 95% of the time and has run more than 20 times in the past 7 days. If the account reaches its task limit, Zapier holds the actions for every Zap in the account until the billing cycle resets or the plan changes. Source: Zapier Help, Zap is not running.
- Inputs drift. Someone renames a form field, adds a dropdown option or moves a spreadsheet column. Nothing errors. The data just lands in the wrong place.
The quiet failure is the expensive one
A loud failure is annoying but cheap. The automation stops, someone gets an error, it gets fixed the same day. The expensive failure is the quiet one: the automation keeps running but does the wrong thing, or it stops and nobody notices. Review requests stop going out for three weeks. New leads from the website form land in a column nobody reads. Invoices go out with last quarter's rate.
The cost is rarely the fix itself. It is the gap between the moment something broke and the moment a person noticed. Maintenance is mostly about shrinking that gap.
What breaks, and how you find out
| What changes | What you see | What catches it |
|---|---|---|
| A login or connection expires | Runs stop or error | A failure alert to a named person |
| The tool hits a plan or usage limit | Runs are held and nothing goes out | A usage check in the monthly review |
| A form field or column is renamed | Runs succeed, data lands in the wrong place | A spot check of a few real records |
| A business rule changes, such as a new price or step | The automation follows the old rule | The owner updates it when the process changes |
| The person who built it leaves | Nobody knows how it works | A one page note and a named owner |
What maintained looks like in practice
It does not take an IT department. It takes four habits.
- One owner per automation. A name, not "the team." Usually the person who owns the process the automation serves.
- A failure ping. The automation alerts its owner when a run errors, and also when it has not run in the time it should have. The second alert matters more, because silence is how quiet failures hide.
- A monthly fifteen minute check. Look at the run history, follow three real records from start to finish, compare usage against plan limits and note any connection that needs to be reauthorized soon.
- A one page note. What the automation does, what it connects to, which account it runs under and who to ask. Where the login lives, never the password itself.
This is the same thinking behind giving AI agents a defined stop and why a pile of zaps is not a workflow. When we scope a first build, the failure ping is part of the build from day one, not an extra.
Does this mean automation is fragile?
No. A well built automation can run for months without anyone touching it. The point is that "without touching it" and "without watching it" are different things. The manual process it replaced also broke regularly. It was just called "someone forgot," and nobody got an alert when it happened.
Most tools already give you what you need to watch. Zapier, for example, keeps a run history with a status for every run, and its troubleshooting guide explains how to replay failed runs once the cause is fixed. The work is deciding who looks, and how often.
Frequently asked questions
How much time does maintaining an automation take?
For a small business with a handful of automations, it is usually a short monthly check plus fixing whatever the alerts flag. If one automation keeps needing attention, it is probably doing too much in one piece and should be split.
Who should own an automation?
The person who owns the business process it serves, not necessarily the person who built it. They are the one who knows when a price, a step or a rule changes.
What should a failure alert include?
Which automation failed, which step, the record it was working on, and a link to fix or rerun it. An alert that only says something went wrong gets ignored.
Can the tool not just retry on its own?
Some can. Zapier, for example, offers Autoreplay on some plans to retry runs that failed because of a temporary issue. Retries help with a brief outage. They do not fix an expired login or a renamed form field.
The takeaway
Automation is not set and forget. It is set and watch, lightly. Name one owner, add a failure ping that also fires on silence, and spend fifteen minutes a month checking the runs. That small habit is what keeps an automation saving time in month twelve instead of quietly costing you in month three.
Like these posts? Get each new one by email with SergeBot Insights.
Keep reading
Hot Take
You Have Six Zaps. Your Team Is Still the Glue.
AI Statistics
62% Are Building Agents. Nearly Half Already Ban Autonomy Somewhere.
AI Statistics
Most Small Businesses Use AI. Only 23% Have Written Rules for It.
Ready to automate?
If you have automations running that nobody has checked in months, a 30-minute discovery call is enough to find the ones most likely to fail quietly and give each one an owner.