Intelligence & Reports

Knowledge Base reports

Q: How do I check if my document uploaded correctly?

Open the Upload Status report. It shows counts at each stage (queued, processing, chunking, embedding, failed), a processing funnel, and a table of stuck documents with the stage they are stuck at and the error, so you can fix and re-upload.

Your uploaded documents are what the AI reads to answer customers. These reports show what's in the library, whether it processed correctly, and whether it's still being used.

These reports describe the current state of your library and do not change with the date filter.

Knowledge Base reports

Knowledge Library

What you see. Your documents with their type, age, chunk count, and retrieval count, grouped into freshness buckets (Current / Aging / Stale). Connected data sources appear here too, with sync status and last-sync time.

How to use it. Keep the library healthy. Documents in the Stale bucket that are still being retrieved are top priority to review or re-upload, because the AI is answering from possibly outdated content.

How it's calculated: Current = updated in the last 7 days, Aging = 8–30 days, Stale = older than 30 days. Retrieval count = how often each document has been pulled into an answer.

Upload Status

What you see. The processing pipeline: counts at each stage (queued, processing, chunking, embedding, failed), a funnel Uploaded → Processing → Chunked → Embedded, and a table of stuck documents with the stage they're stuck at, retry attempts, and the error.

How to use it. Check this right after uploading. If a document is stuck or failed, it isn't helping the AI yet, the error message tells you why (e.g. an unreadable file) so you can fix and re-upload.

How it's calculated: read directly from each document's processing status; "stuck" = failed, or retried more than three times.

KB Search Quality

What you see. How well the AI finds answers: total retrievals, average latency, average top score, low-confidence rate, the score distribution (five bands 0–1), average chunks retrieved vs sent to the AI, failed retrievals, zero-result queries, and the top documents being retrieved.

How to use it. A high low-confidence or zero-result rate means customers are asking things your library can't answer, a direct signal to add content. The top-documents list shows what's pulling its weight.

How it's calculated: scores are the AI's relevance ranking of each match (0–1). Low confidence = best match below 0.5; zero-result = a search that matched no document at all.

Common questions

What are freshness buckets in the Knowledge Library?

Documents are grouped by age: Current (updated in the last 7 days), Aging (8 to 30 days) and Stale (older than 30 days). Stale documents that are still being retrieved are top priority to review or re-upload, because the AI may be answering from outdated content.

How do I check if my document uploaded correctly?

Open the Upload Status report. It shows counts at each stage (queued, processing, chunking, embedding, failed), a processing funnel, and a table of stuck documents with the stage they are stuck at and the error, so you can fix and re-upload. See also check document status.

Do Knowledge Base reports change with the date filter?

No. Knowledge Base reports describe the current state of your library and ignore the date filter, so they always show how things stand right now rather than activity over a chosen period.