What breaks when you automate a process nobody is watching?
Automation rarely fails loudly. It fails quietly, and keeps reporting that it is fine.
What breaks first is your ability to tell that anything broke. An automation that fails with an error is the easy case, because something shouts. The expensive case is the one that keeps returning success while quietly doing nothing, and the reason it is expensive is time. You find out weeks later, and by then you have been making decisions on the assumption that the work was happening.
Why does a broken automation keep looking healthy?
Because most monitoring watches the wrong thing. It checks whether a job returned an error, not whether the job produced anything. Those are different questions, and a job that never runs at all never returns an error. It sits there looking perfect.
We found exactly this in our own systems. A weekly job had been switched off and nothing flagged it, because the health check was reading the last recorded exit code. The last recorded run had succeeded. It had simply been a long time ago, and nothing was measuring that.
What does a silent failure actually look like?
It looks like a green dashboard. Here are patterns we have hit ourselves, and in each one the report was true and useless at the same time.
The gap between what a system reports and what is happening
| What the system reported | What was actually happening | Why nothing caught it |
|---|---|---|
| Last run: success | The job had been disabled for weeks and had produced nothing | The check read the last exit code, and a job that never runs never fails |
| Notification delivered | Nobody replied, so the approval it was waiting on never happened | Something sent the message. Nothing watched for the answer |
| Migration complete | The old job was switched off and the new one was never created | Turning the old thing off was verified. Turning the new thing on was assumed |
| All checks green | A temporary workaround from an outage was still the live configuration | The workaround worked, so nothing ever complained about it |
| Report published on schedule | The report was being built from a source that had stopped updating | The existence of the report was monitored. Its contents were not |
What is an open loop, and why is it not automation?
An open loop is any workflow that hands off to a human and then stops caring. The system sends the message that says reply to approve, and nothing anywhere reads the reply. If the person never answers, the work simply never happens, and no part of the system treats that as a problem.
This is the most common thing we see described as automated, and it is worth being blunt about. A notification that nobody catches is not a step in a process. It is a hope with a timestamp on it.
How do you catch it before it costs you a month?
Five checks that catch a silent failure
- 1Monitor output, not exit codes. The question is never did it run without error, it is what did it produce and when. If the answer is nothing for longer than the schedule allows, that is a failure regardless of what the status says.
- 2Make every automation report what it did, not that it ran. A heartbeat that says finished is worthless. A heartbeat that says wrote four records, sent two emails, skipped one is something you can actually read.
- 3Treat every human handoff as an open loop until something reads the answer. If a step waits on a person, something has to notice when the person does not act. Otherwise you have not automated the process, you have automated the reminder.
- 4When you move a job, confirm it exists at the destination before you disable the source. Moving something is two actions, and it is always the second one that gets assumed.
- 5Schedule the revert whenever you switch on a temporary fix. Emergency settings become permanent settings because nothing is tracking them. Decide when it goes back at the moment you turn it on, not later.
The uncomfortable part
Everything above came from our own systems, not from a case study. We build automation for a living and we still shipped a job that sat dead for weeks while a monitor reported it healthy. That is not a confession for effect, it is the actual point. Silent failure is not a sign of carelessness, it is the default behaviour of any process nobody is measuring.
So the rule we run by now is that an automation is not finished when it works. It is finished when something would tell you if it stopped.
Questions people ask
How would I know if this is already happening to me?
Pick one automation you rely on and ask what it produced in the last week, by name. Not whether it ran, what came out of it. If you cannot answer that without logging into something and hunting, nothing is watching it and you are relying on luck.
Is more monitoring the answer?
Better monitoring, not more. A single message that reports what an automation actually did is worth more than a dashboard full of green lights that only prove nothing threw an error.
What about steps that genuinely need a person?
Plenty do, and that is fine. The rule is to say so plainly and then build the catch, so something notices when the person has not acted and escalates. A manual step you have named is a design choice. A manual step you have forgotten is a broken process.
Does this mean automation is not worth it?
The opposite. It means the value sits in the whole loop rather than the trigger. Anyone can wire up something that fires. The part worth paying for is the part still telling you the truth about itself six months later.
If you have automations running that nobody has looked at in a while, working out which of them are still doing the job is a short conversation and usually a surprising one.
More notes
- Why does AI make things up, and how do you stop it?A model handed something unusable does not stop. It produces something plausible, and it will produce a different plausible thing the next time you ask.
- Why did your cold outreach run out of leads?Usually it is not the message. You have simply run out of people to send it to.
- Do you need to hire a salesperson, or a system?The honest answer depends on one thing, and it is not budget.
- What do you do about sales when you are already at full capacity?The trap is not that you are too busy. It is that being busy feels like being safe.
Want the version of this for your business?
Twenty minutes, your actual situation, a straight answer about what we would do first.
20 minutes. Straight answer, no pitch.