The problem
Nobody tells you that the automation stopped working. You find out three weeks later when you notice the leads stopped arriving, or when a client asks why they never got onboarded.
I see this constantly, and it is the single most expensive pattern in this industry. Not the build cost. The silent failure cost.
After 4 years of owning automation systems end to end, including systems I built in 2023 that are still running, here is what I think most companies get wrong.
Automation is not a project. It is infrastructure.
Companies buy automation the way they buy a website redesign. There is a scope, a build, a handoff, an invoice, and then it is done.
But a workflow is not a deliverable that sits still. It is a live system with dependencies on things you do not control.
Every one of those is routine. None of them are your fault. All of them break your system, and none of them announce themselves.
The n8n example everyone in this space knows
I build a lot on n8n, and I recommend it. It is flexible, self hostable, and far more capable than Zapier for serious work.
It also ships breaking changes. A bad upgrade can take down your entire automation stack overnight. If you are running business critical workflows, that is not a theoretical risk, it is a Tuesday.
That is not an argument against n8n. Every platform in this category has some version of it. It is an argument for someone owning the upgrade path instead of clicking update and hoping.
Silent failure is the real problem
A loud failure is fine. The workflow errors, you get an alert, you fix it in an hour.
The dangerous one is partial. The workflow still runs. It still reports success. But the enrichment step now returns empty, so leads are entering your CRM with no company data, and the routing logic that depends on company data is quietly sending everything to the wrong person.
Nothing is red. Everything is broken.
I have walked into systems that had been degrading for four months. The client thought their lead volume had dropped. It had not. The leads were arriving and going nowhere.
What ownership actually looks like
This is the part I build into everything I ship, and it is the reason my systems are still running years later.
Monitoring the outcome, not the execution
Do not alert on "did the workflow run." Alert on "did the thing the workflow exists for actually happen."
If a workflow is supposed to create 20 to 40 CRM records a week, alert when it creates fewer than 10. The workflow succeeding is not the goal. The records are the goal.
Validation at every boundary
Every time data crosses from one system to another, check it. Is the field present. Is it the right shape. Is the value plausible.
If enrichment returns empty for 30% of records in a run, that run should stop and tell someone, not continue cheerfully.
Pinned versions and a real upgrade path
Do not auto-update a production automation platform. Pin the version. Test upgrades against a copy. Upgrade deliberately.
This is basic software practice that the no code world skipped, and it is why so many automation stacks fall over.
A named human
Somebody has to be responsible. Not "the team." A person, with access, who will notice.
For my clients that is usually me, because a bookkeeper with a Zapier login is not going to catch a schema change in a vendor API.
What to ask before you buy
If you are hiring someone to build automation, the build is the easy part. Ask about after.
If the answer to any of those is silence, you are not buying a system. You are buying a thing that will work for a while.
The uncomfortable math
A workflow that saves 15 hours a week is worth a lot. A workflow that saves 15 hours a week for five months and then silently corrupts your pipeline data for three months is worth less than nothing, because now you also have to clean up the data and rebuild trust with whoever relied on it.
Unmaintained automation does not decay to zero. It decays to negative.
That is why I own the full lifecycle of everything I build, from architecture through deployment to long term reliability. Not because it is a nicer service model, but because the alternative does not actually work.
If you have automation that someone built and left, it is worth having it audited. Most of the time I find at least one thing that has been quietly failing for months.