> ## Documentation Index
> Fetch the complete documentation index at: https://docs.callkaro.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Inside an RCA Report
> What the report contains and what each root cause tells you
The report is written to be read by a business user first and an engineer second. It has four sections.
| Section | Contains |
| ----------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- |
| **KPI Metrics Missed** | Each metric that fell short, with its target, its actual value and the gap. |
| **Summary** | A short, plain-English account a non-technical reader can absorb in under a minute: what went wrong, why, and what to do next. |
| **Root Causes & Recommended Fixes** | The ranked findings. This is the substance of the report. |
| **Updates on Past Findings** | Status on issues raised in earlier reports — whether the fix appears to have been applied, and how any A/B rollout is performing. |
## What each root cause tells you
| Field | Meaning |
| ------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Cause** | The specific reason the metric missed. |
| **Category** | Which part of the system is responsible — see the table below. |
| **Evidence** | What in the day's calls supports this conclusion, with links to the actual audited calls. |
| **Confidence** | How certain the analysis is that this cause is real: High, Medium or Low. |
| **Severity** | How much it costs you if left unfixed, independent of confidence: Very High, High, Medium, Low, Very Low or Negligible. |
| **Prompt Section** | Exactly where in the agent's configuration the cause lives — system prompt, role, goal, instructions, call flow, guardrails, rebuttals, or a specific capability node. Marked *not applicable* when the cause is not a prompt issue. |
| **Prompt Excerpt** | The exact current text that is causing the problem, or a note that no instruction covers the case. |
| **Recommended Fix** | The concrete action. For prompt issues this is the literal wording to add or the literal replacement text — not a description of a change, the change itself. For non-prompt issues it is a specific operational action someone can execute. |
| **Metric Impact** | Which metric fixing this would move, and by roughly how many percentage points. |
| **Recurrence** | Whether this is **new**, or **recurring** from an earlier report — with the date it was first observed. |
## Root cause categories
| Category | Covers |
| -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Agent behaviour** | The conversation itself — weak rebuttals, poor closing, wrong call-flow steps. |
| **Function call or integration** | The agent's tool usage or an external integration misfiring — for example a CRM update failing silently. |
| **Post-call measurement** | The underlying data is being recorded incorrectly, so part of the gap is an instrumentation artefact rather than a real outcome. |
| **System / technical** | Infrastructure, telephony or runtime failures — drops, speech-to-text failures, timeouts. |
| **KPI target unrealistic** | The target you set looks inconsistent with what this agent has ever achieved. |
| **External lead quality** | Nothing in the agent, the systems or the measurement explains the gap — the leads or campaign source do. Only used when a concrete, executable action is attached. |
| **Other** | Anything not covered above, explained in the cause itself. |
**A deliberate design rule.** "We need more data" is never reported as a root cause. If the day's evidence was too thin to conclude anything, the report says so explicitly rather than padding the findings list.
## Notifications
When a report is generated, it is rendered as a PDF and emailed to the agent's owner automatically. You do not need to open the dashboard to know a metric missed — the report arrives with the metrics missed, the summary and the full findings, and the dashboard is there when you want to act on them.