How Document Search Gives Your Agent Deep Knowledge

Documents let you give your agent deep, detailed knowledge without manually converting everything into individual facts and FAQs. Upload your employee handbook, service menu, insurance guide, or policy manual — your agent searches it during calls to find exactly the information a caller needs.

How Document Search Works

When you upload a document, it's processed through a pipeline:

1. Chunking — The document is split into smaller pieces (chunks), each a few paragraphs long with slight overlap for context continuity.
2. Embedding — Each chunk is converted into a mathematical representation (embedding) that captures its meaning, not just keywords.
3. Storage — Chunks and their embeddings are stored in a vector database.

During a call, when your agent needs information that might be in a document:

1. The agent formulates a search query based on what the caller is asking
2. The query is embedded using the same mathematical representation
3. The system finds the chunks most semantically similar to the query
4. The top results are returned to the agent, who reads them and answers the caller

This is called retrieval-augmented generation (RAG) — the agent's response is grounded in your actual documents, not general knowledge.

What Makes This Different from Keyword Search

Traditional search matches exact words. If your document says "cancellation policy" but the caller asks about "refund rules," keyword search fails.

Semantic search (what Allison uses) matches meaning. "Cancellation policy" and "refund rules" are semantically related, so the right document chunk surfaces even when the words don't match exactly.

What to Upload

Good document candidates:

• Employee handbooks — procedures, policies, reference material
• Service menus — detailed descriptions, pricing tiers, package breakdowns
• Insurance or coverage guides — what's covered, what isn't, co-pays, deductibles
• Procedure guides — step-by-step processes for common situations
• Product catalogs — specifications, comparisons, availability

Keep documents factual and current. The agent treats document content as authoritative — if an outdated document says you close at 5pm but your Quick Fact says 6pm, the agent may give conflicting information.

Documents vs Facts vs FAQs

Use documents when:

• The information is too detailed for a Quick Fact (which should be a single value)
• The content is too long for an FAQ (which should be 2-5 sentences)
• You already have the information in document form and don't want to manually re-enter it
• The knowledge is reference material the agent might need occasionally, not on every call

The agent searches documents only when it needs specific information it doesn't already have from facts and FAQs. Documents are the "deep reference library" — facts and FAQs are the "front-of-mind" knowledge.