GET AI WORKING — AI WORKFLOW TESTING CHECKLIST
https://getaiworking.co.uk/resources/ai-workflow-test-checklist

Use this checklist with an agreed demonstration or test environment.
Use suitable sample data. Prevent tests from sending customer messages
or changing live records unintentionally.

1. DEFINE THE WORKFLOW
Task and process owner:
Systems or connections involved:
Input and expected output:
Which steps may prepare a draft:
Which steps may change something:
Who must approve those changes:
How to pause the workflow and continue manually:

2. DEFINE AND RUN THE TESTS
[ ] Complete, ordinary input reaches the right review step.
[ ] A missing required field is identified; missing details are not invented.
[ ] An invalid input follows the agreed correction or stop route.
[ ] Running the same reference twice avoids an unintended second action.
[ ] Unclear output can be checked against its source and rejected.
[ ] Important names, figures and commitments match the source.
[ ] An unavailable connection produces a visible problem for the owner.
[ ] Retrying after a failure does not repeat completed actions.
[ ] An action needing approval stays pending until approval is given.
[ ] The reviewer can understand what happened and use the manual route.

3. TEST LOG — COPY THIS BLOCK FOR EACH CASE
Test reference:
Date and reviewer:
Input or source example:
Expected result (write this before the run):
Actual result and output:
Pass, fail or follow-up needed:
Reason for the decision:
Person responsible for a correction:
Affected tests to repeat after the correction:
Retest result:

4. BEFORE EVERYDAY USE
[ ] Representative examples and awkward cases have been reviewed.
[ ] Failed tests have an agreed resolution and affected checks were repeated.
[ ] The process owner can find the source, output and approval information.
[ ] The team knows who maintains the workflow and who handles a problem.
[ ] Instructions explain how to pause and continue the task manually.
[ ] There is a plan to check changes to sources, instructions and connections.

EXAMPLE TO ADAPT — FICTIONAL ENQUIRY PROCESS
Input: enquiry ENQ-204 has no contact email.
Expected: identify the missing email, retain the enquiry, prepare no reply.
Next test: correct the email, complete the review, then rerun ENQ-204.
Expected: do not prepare an unintended second response.
Record what actually happened; do not assume the expected result occurred.

One successful run does not establish reliability. Repeat with representative
inputs, including changed or incomplete information.

NEXT STEP
Discuss a practical workflow and its review steps:
https://getaiworking.co.uk/services/workflow-automation
https://getaiworking.co.uk/book-a-call