Section 3 · Data 360 Fundamentals · 20% of exam
3.4 Indexing
Exam objective
Describe the role of a Data 360 search index and explain how content preparation, indexing status, and search type affect Agentforce retrieval.
Why an index is needed?
A search index organizes prepared content so a query can locate relevant records or passages without scanning every source manually. For unstructured content, the preparation process commonly includes selecting fields, extracting text, creating chunks, generating vector embeddings, and storing searchable information. An embedding is a numerical representation that helps a system compare semantic similarity. Data 360 search index configurations define how selected content is prepared and made available for search. A data library automates creation of a search index and retriever for supported source choices, while advanced scenarios can configure those components separately.
Indexing is a lifecycle
Creating a library or index does not mean all content is immediately searchable. Data must be ingested or selected, processed, chunked where applicable, and indexed. Processing can take time depending on the volume, size, structure, and complexity of source data. Check the status card or index status to confirm readiness. Salesforce Help directs users to ensure a library is Ready before connecting it to an agent for Answer Questions with Knowledge. If content was recently added or changed, confirm that the relevant updates have been processed. Testing before the index is ready can look like a prompt or agent problem even though the source content is not yet available to retrieval.
Vector and hybrid search
Vector search compares embeddings to find passages that are semantically similar to the query. This helps when users phrase an idea differently from the source. Keyword search emphasizes exact terms. Hybrid search combines vector and keyword search and can be useful when both meaning and exact tokens matter. For example, a user asking about a named product plan may benefit from exact product-name matching, while a user asking what to do after a billing failure may need semantic matching for related guidance. The exam objective is to match the search approach to the retrieval requirement, not to assume that one search mode is always best. Search quality also depends on the text selected for indexing, the query, metadata, and the retriever configuration. Inspect configuration and processing
When troubleshooting, review which objects and fields are included, the chunking strategy, the index type, and the last processing run. Salesforce Trailhead notes that the configuration view can show selected fields, chunking strategy, and chunking configuration; process history can show records processed and chunks created. These details help distinguish an ingestion issue from an indexing or retrieval issue. For example, if an answer exists in a file but the relevant text field was not selected, adjusting the prompt will not make the source searchable. If the index status is not Ready, confirm the processing error or wait for completion before evaluating the agent. Keep the index trustworthy
Index only information that should be available to the intended AI use case. Remove obsolete documents or correct outdated source records, then confirm that the search pipeline reflects the change. Organize source data with useful titles, metadata, and structure. Avoid treating a successful indexing status as proof of content quality: a technically Ready index can still contain stale, duplicated, irrelevant, or poorly formatted material. After an index or source change, retest important questions and verify that returned passages support the expected answer. Consider credit usage, data governance, and operational ownership as part of the index design, especially when indexing large or frequently changing collections.
If a correct source is not appearing in an agent response, verify the index is Ready and includes the relevant object, field, file, or updated content before tuning the prompt.
An index being Ready confirms processing status, not that every result is relevant or every answer is correct. Evaluate retrieval quality with realistic queries.
Extra concept: Vector embeddings and keyword search solve different matching problems; hybrid search combines them when a use case needs semantic relevance and exact-term matching.
Official references:
- Choose a Data Source for Your Library https://help.salesforce.com/s/articleView?id=ai.data_library_choose_data_source.htm&language=en_US&type=5
- Master Search Index Configurations for Effective Data Management https://trailhead.salesforce.com/content/learn/projects/unstructured-data-in-data-cloud/create-a-search-index-configuration
- Enhance Agents with Business Knowledge Using RAG https://trailhead.salesforce.com/content/learn/modules/retrieval-augmented-generation-quick-look/augment-prompts-with-relevant-knowledge
Key takeaways
- A search index makes prepared source content discoverable for retrieval.
- Check selected fields, processing history, index status, and search configuration when results are missing.
- Ready status is an operational check; realistic query testing is still required.
