Failure Analysis
The specific reasons conversations are failing, with the exact exchange that shows it, so you know exactly what to fix.
Shown in Companion Dashboard only - this view isn't currently available on the org-wide Dashboard.
Failure Analysis is a searchable, sortable, filterable list of every detected failure, grouped by root cause and severity - not just a count of how often things go wrong, but why, with the actual exchange that shows the failure happening.
What you see
| Column | What it shows |
|---|---|
| Failure Reason | One of: knowledge gap, hallucination, misunderstood intent, incorrect info, missing product knowledge, capability gap, tone mismatch, policy block, unclear question, out of scope, or technical error. |
| Severity | Critical, High, Medium, Low, or None. |
| Created | When this failure was first hit, in a conversation from the selected date range. |
| Updated | When this failure was most recently hit, in a conversation from the selected date range. |
| Count | How many conversations hit this failure. |
Expand any row for a suggested fix specific to that failure reason, then a grid of the conversations that hit it - each card showing the quoted exchange where the failure happened. Click a card to open that conversation's full transcript in a new tab, or tap its info icon for a quick-look popover instead.
How to use this
- Sort by Severity first - Critical and High entries deserve attention before anything else, regardless of count.
- A high count of Knowledge Gap or Missing Product Knowledge failures means it's time to add or update training resources.
- Hallucination or Incorrect Information failures are the most urgent to fix - they erode trust fastest. Expand the row, read the quoted exchange, and trace it back to the specific source material to correct.
- Use this to explain movement in Resolution Rate and Response Quality - a dip in either usually shows up here as a spike in one specific failure reason.
Last updated on