(CIO Office) - Enhancement Request — LLM-Inferred Dropdown Value Persistence

Current behavior

When a User Activity contains a dropdown/choice field and the user provides input that does not exactly match the spelling of one of the available dropdown options, the LLM is able to correctly infer the intended option and the flow proceeds using the inferred value.

However, the dropdown itself is not populated with the inferred value. As a result, when the user exits and re-enters the thread, the dropdown continues to display the default “Choose an Option” placeholder rather than the option that was previously inferred and used by the flow.

Example

Dropdown options:

• Benefits and Leaves
• Payroll
• Compensation

User enters a request such as:
“I have a question about my benefits and leave.”

The LLM correctly infers “Benefits and Leaves” and the flow proceeds with that value. However, the dropdown UI remains unpopulated, and after thread re-entry it again displays “Choose an Option.”

Enhancement requested

When the LLM infers a value that corresponds to one of the available dropdown options, the platform should:

1. Populate the dropdown with the inferred option.
2. Persist the resolved selection as part of the thread/message state.
3. Restore the inferred selection when the user exits and re-enters the thread.
4. Ensure the behavior is consistent with a value selected manually from the dropdown.

Expected behavior

If the LLM infers “Benefits and Leaves” from the user's input and the flow proceeds using that option, the dropdown should also reflect “Benefits and Leaves” as the selected value. On thread re-entry, the same value should remain populated rather than reverting to “Choose an Option.”

This would provide a consistent user experience between manually selected dropdown values and values resolved through LLM inference, while ensuring that the UI accurately reflects the value that was used by the flow.

Context

This enhancement is related to the existing behavior identified during investigation of dropdown persistence. The exact-text-match scenario is being tracked separately as a defect. This request specifically covers the LLM-inferred/non-exact-match scenario, which has been identified by the development team as a feature enhancement.