A failed test after an AI suggestion is a Workday configuration defect or a bad suggestion. The ticket has to say which. Workday does not provide a ticket form for this. The fields below are the ones a second administrator can retest from the admin guide: the event, the security report, and the exact fault. Paste the prompt. Do not paste a worker’s pay.
Ticket
Title: [Tenant] [task or rule name] failed after an AI suggestion
Environment
- Tenant name and type: Implementation, Preview, Sandbox, or Customer Central. Deployment Agent is not a Production assistant. If the failure is Production, say the change was moved there and name the migration (for condition rules, Object Transporter is the method Workday documents).
- Workday user who applied the change.
- Workday user who retested. If this was a proxy session, name the proxy user and the person they acted for, and confirm Stop Proxy was run. The Adaptive Planning procedure is the published step list for Start Proxy and Stop Proxy in non-production. Say if you used that pattern or a proxy policy you confirmed for this tenant.
What the assistant was
- Product: Deployment Agent in non-production, Self-Service Agent, or another assistant you can name from Workday’s catalog. Do not write “the AI.”
- Skills you believe were on. For Deployment Agent Ask, Workday says you need Tenant Inspection and Implementation Knowledge, and there is no on-screen indicator. Write whether the prompt included “in my tenant” or a Customer Central tenant name.
- Thumbs down: yes or no. On Self-Service Agent, detailed thumbs-down is on the latest reply only. Say if you could not leave a comment. Thumbs-down does not revert a saved field or an activated policy.
Prompt and answer
- The full prompt, including role and format.
- The answer’s task names, domain names, and field names. Quote those. Summarize the rest.
- Any follow-up, such as a renamed management level (“Lead” instead of “Manager”).
What you changed
- Task you actually ran (Create Calculated Field, Edit Condition Rule, Edit Security Policy Permissions, Edit Custom Report, template model on an EIB).
- Whether Activate Pending Security Policy Changes was confirmed. A test before activation is not a product failure.
- For a calculated field: tenant-wide or report-specific, business object, function, and the condition order. Evaluate Expression stops at the first true row and otherwise returns the default.
What the tester showed
Pick the row that matches the object. Leave the others blank.
| Object | Paste this result |
|---|---|
| ObjectCondition rule | Paste this resultRule Tester flow from a business process event. Note the first And row that was false or the first Or row that was true. Say if the source on initiation was Initiator. Say if the step was skipped for the processing user. Rule Tester does not exist for Workday-owned rules. |
| ObjectAgent skill security | Paste this resultView Security for Agent Skill functional status: Full Match, Review Required, Config Required, or Not Evaluated. Tool name. Available To group. Whether Test My Setup was run. |
| ObjectDomain change | Paste this resultDomain name, View or Modify, group added, and the result as a user in the group and a user outside it. |
| ObjectSOAP or EIB | Paste this resultExact fault text. What you already confirmed was present (the published example is a parent element that was in the XML). Web service name. Whether you asked for the ISU’s functional area and domain permissions. |
| ObjectCalculated field | Paste this resultOne worker who should match, one who should hit the default, and any Calculated Field Exception Audit row. |
Expected and actual
- Expected, in one sentence from the design note.
- Actual, in one sentence from the tester. Not from the assistant.
Scope
- Other tenants where the same rule or field exists.
- Whether a Sandbox refresh could copy a Production agent configuration back. For Deployment Agent, deactivating Sandbox alone is not enough if Production still has it for refresh.
How to title the cause
Use one of these, and only after the tester supports it.
| Cause label | Evidence you already pasted |
|---|---|
| Cause labelSuggestion does not match the tenant | Evidence you already pastedDomain or field name in the answer does not exist on the task you opened. |
| Cause labelShort-circuit | Evidence you already pastedRule Tester stopped before the row you care about. |
| Cause labelInitiator on initiation | Evidence you already pastedSource is Initiator and the value is empty. |
| Cause labelStep skipped for the processing user | Evidence you already pastedAdministrator sees the step. The previous step’s user does not. |
| Cause labelSecurity still pending | Evidence you already pastedActivation was not confirmed. |
| Cause labelAgent skill not actually evaluable | Evidence you already pastedStatus is Not Evaluated or Config Required. |
| Cause labelParent SOAP fault | Evidence you already pastedThe named parent element was present. Child elements or domain permissions were not checked. |
| Cause labelWrong product | Evidence you already pastedA Production employee scenario was tested in Deployment Agent, or an ASOR client was used for a non-ASOR integration. |
If you cannot pick a row, the ticket is not ready. Ask for the missing tester output instead of asking the model to try again inside the same chat.
What not to attach
- A full worker profile or a payroll result row. Name the worker ID only if your ticket system is already allowed to hold it. The condition outcome is enough.
- A client secret or an API client credential. The failure is the Workday task, not the credential file.
- A Community thread as the expected result. Expected comes from the design note and the admin-guide behavior you are testing.
After the ticket
The person who fixes it re-runs the same tester and records the new flow in the ticket. They do not replace the ticket with a fresh chat. If the original suggestion was wrong, leave the thumbs-down in place so the rating and the defect are both true.
A second failure with the same prompt and a new task name is a new ticket. Otherwise you will “fix” a condition rule that was never the object under test.
Frequently asked questions
Is this a Workday-delivered template?
No. Workday documents Rule Tester, View Security for Agent Skill, security activation, and SOAP examples. This form is a way to file those results so a fix does not depend on a 14-day Deployment Agent chat history. Histories are deleted after 14 days and do not cross tenants.
Should the assistant be deactivated because a test failed?
Only if the failure is “people are using an agent that should be off,” or finance wants Flex Credit consumption for Deployment Agent to stop in that tenant. A wrong calculated-field suggestion is fixed in Create Calculated Field. Deactivate is a separate decision, and it has to include every tenant you configured, including Production if refresh would turn Sandbox back on.
Can we reopen the chat instead of filing this?
Reopen it to copy the prompt if the 14 days have not passed. File the ticket with the tester output. The next administrator may not have your chat, and Customer Central history is not the Sandbox history.
What if Test My Setup and the proxied user disagree?
Put both results in the ticket and do not pick a cause label until someone looks at pending security changes and at optional field-level policies. Disagreement is the defect. Hiding one of the results makes the next edit guess.
Sources checked
Workday pages in the frontmatter, fetched 2026-11-16. The form itself is not a Workday screen.