People ask which no-code tool to buy. I start with three automations that failed — including ones I approved. The pattern matters more than the logo.
Autopsy 1 — the invoice ping that cried wolf
Intent: Slack alert when an invoice passed due.
What shipped: Alert on every status change, including drafts and manual edits.
Failure: Channel muted in four days. Real overdue invoices vanished into noise.
Fix: One alert only when status = overdue AND days_late >= 3, with a daily digest ceiling of five messages. Automation is not “notify on everything.” It is notify on the decision you will actually take.
Autopsy 2 — CRM enrichment that invented people
Intent: Auto-create contacts from email form submissions.
What shipped: New contact per submission, no dedupe.
Failure: Three records for one customer; sales called the same person twice.
Fix: Match on email before create; otherwise update. Add a weekly “duplicate suspects” view. Speed without identity rules creates cleanup debt that costs more than the typing you saved.
Autopsy 3 — the Zap nobody owned
Intent: Copy signed contracts into a folder.
What shipped: Worked for six months until an API token expired Friday night.
Failure: Weekend contracts sat in email. Monday blame circled the tool, not ownership.
Fix: Named owner, calendar reminder for token rotation, and a Friday check that the last file landed. Unowned automations are invisible employees that quit without notice.
What to automate first
High volume, low judgment, expensive when forgotten: invoice reminders, backup confirmations, access offboarding checklists, form-to-tracker with dedupe. Avoid automating nuanced client tone, pricing exceptions, or anything you cannot monitor weekly.
A 30-minute go / no-go test
- What human action happens when this fires?
- What happens if it fires ten times too often?
- Who gets paged when it stops?
- Can you turn it off in under two minutes?
If answers are vague, keep the manual step. Manual and reliable beats automatic and ignored.
How I monitor something after it ships
For the first two weeks, I watch the automation the way I would watch a new hire: a daily glance at the last five runs. After that, a weekly check is enough if there is a named owner and a way to see failures (email, a log, a folder that should not be empty). If you cannot see the last successful run without opening five admin screens, you will not monitor it. Then it is not an automation. It is a future incident.
Write the kill switch in the same note as the Zap: which login, which toggle, who else has access. Friday-night token deaths are expensive because the only person who could stop the job is already off the clock.
When no-code is the wrong layer
Multi-step money movement, anything that must be auditable for a regulator, and workflows with more than two “if this, unless that, except on Tuesdays” branches. Those belong in a spec and a developer, or they belong back in a checklist. No-code shines when the path is boring and the cost of a miss is an annoyance, not a lawsuit or a double charge.
If a vendor demo requires three workarounds before it matches your real invoice, you are not evaluating a tool. You are being trained to accept a worse process. Walk away or write the spec.