We're using a Flex Prompt Template with Einstein Search Retriever to fetch historical case resolutions grounded on structured case data from a Data Cloud index. The retrieval should match strictly on two fields: Type and Subtype.
Despite reinforcing these constraints through prompt instructions, the retriever still returns fuzzy or partial matches, reducing response relevance. Our goal is to only return results where all three fields exactly match the user input.
We've tried various prompt phrasing strategies but haven't seen consistent improvements.
Has anyone successfully enforced strict matching conditions with Einstein Search Retriever in a similar use case? Is there a way (via filtering, metadata constraints, or another approach) to ensure that results respect exact field-level matching in the retrieval phase?
Any guidance or examples would be much appreciated.
Hi, sorry for the inconvenience. If you don't receive a response here, we recommend reaching out to our support team for further assistance. You can do so by visiting https://help.salesforce.com/s/articleView?id=000393090&language=en_US&type=1