Spring ’25 expanded Salesforce’s agentic and retrieval capabilities, including richer grounding options. For service leaders, retrieval-augmented generation stopped being an AI-side experiment and became part of the core knowledge architecture.
The hard problem was not returning more passages. It was selecting authoritative evidence for the customer, product, market and moment.
Cloud Group point of view
Retrieval quality depends more on knowledge operations and metadata than on vector search alone. Hybrid search is powerful because it combines semantic similarity with exact business constraints.
A practical playbook
The strongest next step is narrow enough to govern and useful enough to produce evidence. We would structure the work around these moves:
- Define which sources may answer which classes of question.
- Add product, region, audience, version and effective-date metadata.
- Use keyword constraints for identifiers, policy terms and regulated language.
- Measure retrieval separately from final-answer quality.
- Create a content retirement and freshness operating rhythm.
The architecture and operating implication
Build a retrieval layer that can filter by customer entitlement and record context before ranking content. Keep citations attached to the generated answer, and preserve the retrieved evidence in traces. Use structured Salesforce facts directly rather than forcing every field through document retrieval.
Measure what changes
Model activity is not a business result. Track a small set of indicators that connect behavior to accountable work:
- Relevant evidence in the top retrieved results
- Answer support and citation validity
- Stale or conflicting content incidents
- Resolution and escalation by knowledge domain
Service AI becomes dependable when knowledge is treated as a product with ownership, lifecycle and measurable performance—not a folder connected to a model.
Primary sources
This field note is grounded in the product and market context available at the time of publication.
