Best Practice for Validating AI-Populated Reference Fields in Incident Creation

[MiftahF]

Hi everyone,

I’m currently building an AI Agent for Incident Creation that also performs classification (e.g., Business Service and Service Offering).

What is the best practice to ensure that the values selected by the AI Agent strictly match existing records in the system?
For example, when populating Business Service and Service Offering, how can I guarantee that the agent only selects valid values that already exist in the respective tables (instead of generating free text or invalid entries)?

Feel free to suggest the best approach or tools to reach that

Thanks in advance!

[TharaS657398130]

Hey

Best practice is: don’t let the AI directly fill reference fields like Business Service or Service Offering. Instead, let the AI only suggest a name, then in your Flow Designer validate it against real records (like in cmdb_ci_service and service_offering) using a lookup.

If a match is found, use the sys_id to populate the field; if not, leave it empty or use a default. Even better, maintain a small mapping table where AI outputs are pre-mapped to valid records. This way, you ensure only existing values are used and avoid any invalid or free-text entries.

[Brian Bakker]

@MiftahF 

In the Now Assist for IT Service Management (ITSM) application, we have an out-of-box "Classify service and CI AI agent" that uses a script for validation purposes to ensure it only recommends valid values in the system.

We recommend duplicating this out-of-box AI Agent and then to customize it, as per your requirements instead of building it from scratch, as this should save you significant time.

[MiftahF]

Thanks for the suggestion, 
I've tested the ootb Agent, but somehow the accuracy is not good because we have custom term or keyword for each BS or SO. Is there any appoach to refer to existing populate value from existing table?