Checklist Before Switching On a Business Automation
By CodexierPublished 6 min read
An automation that works in a demo is not the same as one that is safe when a real invoice, a real customer or a real bank transfer is on the other end. The difference is preparation: test cases, a named owner, a log you can read and a way back. This is the checklist we run with clients before any workflow goes live, whether it was built in Make, n8n, Power Automate or custom code.
The short version
Automations fail quietly. A form field renamed, an API token expiring, a supplier changing their invoice layout: none of those raise an error you will see unless you designed for it. The cost of a silent failure is paid by the customer who never got a confirmation, or by the bookkeeper who finds duplicate entries in Fortnox three weeks later.
Test data and edge cases
Most automations are tested with the happy path only: one clean order, one tidy lead, one well-formed PDF. Real input is messier. Write the cases below down, run each one, and record what happened.
- Empty and missing fields: a lead without a phone number, an order without a shipping address.
- Swedish characters and formats: å, ä, ö in names, personal numbers with and without the century, dates written 2026-09-25 and 25/9.
- Duplicates: the same webhook delivered twice, the same customer submitting the form twice in a minute.
- Amounts: zero, negative (a credit note), decimals with comma, VAT included versus excluded.
A named owner and alerts that reach a person
An automation without an owner belongs to nobody, which means the first person to notice a failure is a customer. Ownership is a name, not a department, and it comes with the login to the tool, the right to pause the flow and the alert channel. An error email to a shared inbox nobody reads is the same as no alert.
| Event | Who is told | How | Expected reaction |
|---|---|---|---|
| Single run fails | Owner | Email or chat message with the record id | Retry or fix by hand within the day |
| Three failures in a row | Owner and their backup | Chat message plus phone if outside office hours | Pause the flow, investigate before resuming |
| Nothing has run for 24 hours when it should | Owner | Scheduled heartbeat check | Confirm the trigger still fires |
| Token or connection expired | Owner | Alert from the tool | Re-authorise the connection, check missed runs |
Logging and an audit trail
A log is what lets you answer the question that always comes: did the automation handle order 4711, and what did it do? The built-in run history in Make or n8n is a start, but it is kept for a limited time and is hard to search. For anything touching money or personal data, write one line per run to a place you control: a sheet, a database table or a log channel.
- Timestamp, the record id in the source system and the id created in the target system.
- The outcome: success, skipped (and why) or failed (with the error text).
- Which version of the workflow ran, so a change on Tuesday can be linked to a problem on Wednesday.
- For personal data: only what you need to trace the run, never a copy of the whole record. GDPR applies to your log too.
Rollback and a manual fallback
Rollback means two things. First: how do you stop the automation? Second: how do you undo what it already did? Stopping is usually one toggle, if the owner knows where it is. Undoing is harder: a sent email cannot be unsent, and an invoice posted to the ledger has to be credited, not deleted.
Reversible actions
Creating a draft, tagging a contact, adding a row to a sheet. Let the automation do these freely; the cleanup is a filter and a delete.
Hard-to-reverse actions
Sending email or SMS, posting a payment, booking a supplier invoice, changing a price. Put a human approval in front of these for the first weeks.
The manual fallback
Write down in five lines how the job is done by hand if the automation is off for a week. If nobody can, the process was never understood well enough to automate.
The first-week watch list
Go-live is the start of a trial period, not the end of the project. Set a daily ten-minute check for the first week and a weekly one for the first month, to catch the slow failures: the flow that runs but skips every second record, or works on weekdays and not at all on weekends.
- Compare counts: records in the source versus records created in the target, every day.
- Open three random records end to end and check every field.
- Read the error log even when no alert fired.
- Ask the people downstream, such as the bookkeeper, whether anything looks different.
- After a month, decide in writing: keep, adjust or remove the approval step.
One honest note: not every process deserves this much ceremony. If the automation moves internal notes between two tools and the worst case is a missing reminder, run it, watch it for a day and move on. The full checklist is for flows that touch customers, money, personal data or your bookkeeping. We build and harden workflows as a fixed-price workflow automation project and the go-live checklist is part of every delivery; see the pricing page, or book a call and we will tell you honestly whether your flow needs it.
Frequently asked questions
How long should the test phase take before go-live?
For a simple flow between two systems, a few days running in parallel with the manual process is enough. For anything that posts to your bookkeeping or contacts customers, run it in shadow mode for a full billing cycle so you see month-end behaviour.
Should we keep a human approval step permanently?
Keep it wherever the action is hard to reverse and the volume is low enough to review in minutes a day. Remove it where volume makes the review a rubber stamp; better logging is the right control then.
What if the tool we use has no proper logging?
Add your own. Almost every automation tool can append a row to a spreadsheet or post a message to a channel at the end of a run, which gives you a searchable trace that outlives the tool's built-in history.
Do we need this checklist for a tiny internal automation?
No. If the worst outcome is a missed internal reminder, test it once and switch it on. The full checklist is for automations that touch customers, payments, personal data or the ledger.
Want a second pair of eyes before you switch it on?
Fifteen minutes, you walk us through the flow, we tell you which of these checks actually matter for it and what a hardened version would cost. No obligation.
Book a free 15-minute call