Test automation: so silence never means it is working
Checks by itself that your automations still do what they should
You hear that something broke from the system, not from a customer
A broken automation does not raise its hand. It keeps looking like it runs, until somebody notices nothing has arrived for three days.
What you get
Failure is loud
A flow that did not run, or ran empty, raises an alert. The absence of a result is itself a result.
Changes do not break the rest
Before anything new goes live, what already worked is checked automatically.
You catch external changes
When a supplier changes an API or a format, the check flags it on the first run.
Repair costs less
A problem found on the day it appeared costs far less than one discovered at month end.
How it works
1. What correct means
For each flow we define what should happen: how many records, in what window, in what shape.
2. Checking the run
After each run the result is confirmed. A flow that finished producing nothing does not pass.
- n8n
- PostgreSQL
3. Regression checking
Before a change, existing flows are run against test data and compared with known results.
- Playwright
4. The alert
Whatever fails reaches a person, on the channel they read, with what was expected and what arrived.
- Slack
- WhatsApp Business API
Where it works well
- Scheduled flows that can fail quietly
- Integrations with systems you do not control
- Automations touching money: invoicing, payments, stock
- Systems where several flows depend on one another
Where it does not
- It catches nothing that was never covered — coverage is a choice, not a guarantee
- Checks that are too strict produce false alarms, and false alarms get ignored
What we use
- n8n
- PostgreSQL
Frequently asked questions
- Is it not enough to check now and then?
- That works until the day nobody looks. Automated checking is cheap precisely because it does not depend on discipline.
- What does it add on top of the automation itself?
- Roughly ten to fifteen per cent of the flow's effort. On flows that touch money, it is the cheapest insurance available.
- Will we be getting alerts constantly?
- If so, the checks are badly calibrated and we retune them. An alert you start ignoring is worse than no alert.
- Can it be added to existing automations?
- Yes, and it is often the first thing we do on systems built by someone else. They need no rewrite to be checked.
- What happens when an alert is real?
- It depends on the contract: on the managed service we act, otherwise you get the alert with a diagnosis and decide.