How Do I Test Business Automations Before Turning Them On?
By Bill Stilwell · Last reviewed October 17, 2026
The short answer
Section titled “The short answer”Before an automation goes live, run it on test contacts you control. Check that it starts at the right moment, uses the right data, waits the right amount of time, never sends duplicates, stops on opt-outs and replies, shares information only with the right people, and logs errors. For AI tools, also test strange inputs, costs, and human approvals. Write down every test and every change.
Why it matters
Section titled “Why it matters”Automations repeat mistakes at scale. A typo in a manual email hits one person. A broken trigger can hit every customer on your list overnight. Ten minutes of testing with fake contacts is cheap compared with apologizing to 500 people.
18.10.1 Trigger test
Section titled “18.10.1 Trigger test”What it means: confirming the automation starts only under the right condition.
How to do it: create a test contact with your own email and phone. Make the trigger happen, like submitting the form or marking an estimate sent. Confirm it starts. Then do something close but different, and confirm it doesn’t start.
Common mistake: testing that it starts but never testing that it stays off when it should.
18.10.2 Data test
Section titled “18.10.2 Data test”What it means: confirming the right records and fields are used.
How to do it: check that names, dates, prices, and links fill in correctly. Test a contact with a missing first name to see what happens.
Common mistake: “Hi {first_name},” sent to real customers.
18.10.3 Timing test
Section titled “18.10.3 Timing test”What it means: confirming delays and schedules work.
How to do it: shorten delays to minutes for the test, or check the scheduled send times in the tool. Confirm time zones and quiet hours (see 18.4 Text Message Marketing Rules).
Common mistake: a “3 days later” step that sends in 3 minutes.
18.10.4 Duplicate test
Section titled “18.10.4 Duplicate test”What it means: confirming one person doesn’t get the same message twice.
How to do it: trigger the automation twice for the same test contact. Confirm they only enter once. Check what happens if the same person is in two sequences.
Common mistake: a customer who fills out the form twice gets every message twice.
18.10.5 Stop test
Section titled “18.10.5 Stop test”What it means: confirming opt-outs, purchases, replies, and finished actions stop the sequence.
How to do it: for each exit, run a test contact and do that action: reply, book, buy, text STOP, and unsubscribe. Confirm no more messages go out. Federal Communications Commission (FCC) rules require honoring text opt-outs within a reasonable time, no more than 10 business days, and the Federal Trade Commission (FTC) sets 10 business days for email.
Common mistake: STOP works for texts but the email sequence keeps going.
18.10.6 Permission test
Section titled “18.10.6 Permission test”What it means: confirming only authorized people and systems get the information.
How to do it: check who gets internal alerts, who can see records, and what data is sent to other tools. Remove anything a tool or person doesn’t need (see 11.7.6 Least Privilege Access).
Common mistake: an alert with a customer’s full details sent to a group chat that includes former staff.
18.10.7 Failure test
Section titled “18.10.7 Failure test”What it means: confirming errors are logged and someone is told.
How to do it: break something on purpose. Use a bad email address or a disconnected test number. Confirm the error is logged and an alert reaches the owner (see 18.9 Internal Notifications).
Common mistake: failures that only show up in a log nobody reads.
12.5.3 Automated texts
Section titled “12.5.3 Automated texts”The communication systems view points here. For automated texts, test the trigger, delays, duplicate prevention, and stop conditions using 18.10.1, 18.10.3, 18.10.4, and 18.10.5. Use only phones you own for testing.
25.10.1 Functional test (AI and advanced systems)
Section titled “25.10.1 Functional test (AI and advanced systems)”Confirm the system does the job it was set up to do, like an AI tool that drafts replies actually producing a usable draft from a real example.
25.10.2 Boundary test
Section titled “25.10.2 Boundary test”Try unusual, missing, extreme, and conflicting data. Give an AI tool a blank message, a very long one, a message in another language, and an angry complaint. Check that it doesn’t make promises or share private details.
25.10.3 Permission test
Section titled “25.10.3 Permission test”Confirm the AI tool or system can reach only the data and actions it needs, as in 18.10.6.
25.10.4 Failure test
Section titled “25.10.4 Failure test”Confirm errors are captured and routed, as in 18.10.7. Check what happens when the AI service is down.
25.10.5 Cost test
Section titled “25.10.5 Cost test”Check how the tool charges, like per message, per task, or per user. Run a small test and look at the bill. Set spending limits or alerts if the tool offers them.
25.10.6 Human review
Section titled “25.10.6 Human review”Confirm required approvals happen before anything goes to a customer (see 25.3 Human Review of AI Output). Test that an unapproved item can’t be sent.
25.10.7 Change log
Section titled “25.10.7 Change log”Record each change, who made it, the date, why, and the test results. Retest after every change (see 25.1 AI and Automation Inventory).
When to get help from a pro
Section titled “When to get help from a pro”An automation specialist can build a test plan for complex systems. Questions to ask:
- Which exits and failures should I test in this automation?
- Can I run tests without messaging real customers?
- How do I set spending alerts on this tool?
Bill’s note
Section titled “Bill’s note”[Bill: add a short note. For example, how SiteSmarty tests client automations before launch, or a broken automation you’ve had to clean up.]
How I did this for SiteSmarty
Section titled “How I did this for SiteSmarty”- [Date I did this:]
- [What I actually did, step by step: the test contacts and test checklist I use for SiteSmarty automations]
- [What it cost and how long it took:]
- [What surprised me or what I’d do differently:]
- [Screenshot or photo to add, with private info blurred: the automation test log]
- [Result so far:]
Questions people ask
Section titled “Questions people ask”What should I test in an automation? The trigger, data, timing, duplicates, stop rules, permissions, and failure alerts.
How do I test that opt-outs work? Use a test contact you control, text STOP and click unsubscribe, and confirm no more messages arrive in any tool.
How do I test an AI tool? Give it normal, strange, and difficult inputs. Check that it stays within limits, keeps private details private, and waits for human approval where required.
Do I need a change log? Yes. Record each change, who made it, why, and the retest results.
Audit check
Section titled “Audit check”| Outline line | Question | Pass looks like | Fail looks like | Evidence to collect |
|---|---|---|---|---|
| 18.10.1 Trigger test | Does the automation start only under the right condition? | Positive and negative tests passed | Only start tested | Test log |
| 18.10.2 Data test | Do fields fill correctly, including blanks? | Correct merge fields | Raw placeholders sent | Test messages |
| 18.10.3 Timing test | Do delays and time zones work? | Correct send times | Wrong delays | Test log |
| 18.10.4 Duplicate test | Is double entry blocked? | One entry per person | Duplicate messages | Test log |
| 18.10.5 Stop test | Do replies, bookings, purchases, and opt-outs stop it? | All exits tested | Messages after STOP | Test log |
| 18.10.6 Permission test | Do only authorized people and tools get data? | Minimal access | Over-sharing | Access review |
| 18.10.7 Failure test | Are errors logged and alerted? | Alert reached owner | Silent failure | Test log and alert |
| 12.5.3 Automated texts | Were automated texts fully tested? | Trigger, delay, duplicate, stop tested | Untested | Test log |
| 25.10.1 Functional test | Does the system do its job? | Works on real examples | Unusable output | Test results |
| 25.10.2 Boundary test | Were odd inputs tested? | Safe handling | Promises or leaks | Test results |
| 25.10.3 Permission test | Is AI access limited? | Least access | Broad access | Access settings |
| 25.10.4 Failure test | Are AI errors captured? | Logged and routed | Silent | Test results |
| 25.10.5 Cost test | Are costs checked and limited? | Limits or alerts set | Surprise bills | Billing check |
| 25.10.6 Human review | Are approvals enforced? | Unapproved items blocked | Sent without review | Test results |
| 25.10.7 Change log | Are changes and retests recorded? | Dated log | No log | Change log |
Where this fits
Do this first: 18.5 Follow-Up Sequences, 18.9 Internal Notifications
Do this next: 18.11 Monitoring Automation
Related pages: 12.6 Communication Consent Records, 12.7 Testing Communication Channels, 18.4 Text Message Marketing Rules, 11.7.6 Least Privilege Access
Sources
- Electronic Code of Federal Regulations, 47 CFR 64.1200, Delivery restrictions (checked October 17, 2026)
- Federal Trade Commission, "CAN-SPAM Act: A Compliance Guide for Business" (checked October 17, 2026)
This page explains general rules and Bill's own experience. It is not legal, tax, or insurance advice. Talk to a qualified professional about your situation.