> ## Documentation Index
> Fetch the complete documentation index at: https://docs.retellai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Review QA results

> Review AI QA results with Retell AI Conductor. Inspect cohort trends and individual evaluations, compare findings with call transcripts, and identify fixes.

Conductor can inspect AI QA cohorts, aggregate results, and individual call evaluations. Ask it to explain a trend or examine the calls behind a low score.

## When to use it

Use Conductor to investigate a drop in resolution rate, understand why a call failed a criterion, or compare a QA finding with the actual conversation.

For example: "Which calls in this cohort failed the appointment-confirmation criterion, and what happened in them?"

## Investigate QA results

<Steps>
  <Step title="Choose a cohort">
    Open a cohort and Conductor, or use **Add context** (+) → **AI QA** to attach one. You can also ask Conductor to find a cohort by name.
  </Step>

  <Step title="Start with the overall results">
    Ask about aggregate results and trends, then ask for the individual scored calls behind them.
  </Step>

  <Step title="Check an evaluation">
    Ask why a call received its score and have Conductor compare the evaluation with the transcript before proposing an agent change.
  </Step>
</Steps>

Conductor can return a navigation card with **View** to open the cohort and continue the conversation there. See [Open pages from replies](/conductor/overview#open-pages-from-conductor-replies).

## Why can one call have different QA results?

A call can belong to several cohorts with different evaluation criteria. Specify the cohort when asking Conductor to explain a score.

## Change the agent or the evaluation criteria

If the transcript shows an agent problem, ask Conductor to propose a fix and [review the changes](/conductor/build-and-refine#review-and-apply-changes).

Conductor’s QA tools inspect results; they do not create cohorts or edit their criteria. If the criteria do not reflect your intended outcome, update them using [Define resolution criteria](/ai-qa/define-resolution-criteria). See [AI QA overview](/ai-qa/overview) for setup and sampling.