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

# AI CSR call quality is low

> Diagnose persistently low Maya call quality scores using the Review page, including FAQ gaps, hallucination patterns, and transfer rule mismatches.

# AI CSR call quality is low

Maya's average quality score on the dashboard is below 4 stars, or multiple calls are appearing in the Call Review bucket with 1–2 star scores.

## Why this happens

* **FAQ gaps.** Maya doesn't have an answer for common caller questions, so she improvises or deflects. This often triggers Hallucination or Source failures.
* **Incorrect or outdated FAQs.** An existing FAQ contains wrong information (old pricing, discontinued services), causing Maya to give callers inaccurate answers.
* **Transfer rules not matching caller intent.** Calls that callers expect to escalate to a human are being handled by Maya when they shouldn't be, or vice versa. This surfaces as Escalation or Task failures.
* **Greeting or voice mismatch.** The current greeting is unclear or the selected voice persona doesn't fit your customer base, leading to confused or short calls that score low on Length or Politeness.

## Try this first

1. Open **AI CSR** in the left navigation, then click **Open review bucket** from the Action Required section (or navigate to `/receptionist/review`).
2. Look at the failing-axis tabs — **Hallucination**, **Source**, **Politeness**, **Length**, **Escalation**, **Task**. Note which axis has the highest call count. That is your primary problem area.
3. **For Hallucination or Source failures:**
   * Go to **AI CSR > Maya Asks** (`/receptionist/asks`). Filter by **Asked ≥3 times** to find the highest-frequency unanswered questions.
   * Click **Add as FAQ** for each relevant question, write a clear accurate answer, and save. Maya will start citing the new FAQ immediately.
   * Also open **AI CSR > Settings > FAQs** (`/receptionist/settings/faq`) and check the Active FAQ list for outdated entries. Edit or delete any that no longer reflect your services or pricing.
4. **For Escalation or Task failures:**
   * Go to **AI CSR > Settings** (`/receptionist/settings`) and review the **Transfer rules** section.
   * Check which custom rules are active. If a rule is too broad or too narrow, edit it so only the intended calls trigger a transfer.
   * Ensure the always-on safety rules (caller asks for a human, after-hours, existing customer with active job) are toggled correctly for your business hours.
5. **For Politeness or Length failures:**
   * Review the **Greeting** field in Maya Settings and update it if it's too long, too formal, or ambiguous.
   * Listen to a sample of the affected calls using the recording link on the call detail page to understand what's going wrong in the conversation.
6. After making changes, return to the AI CSR dashboard and monitor the **Quality Trend** chart over the next few days to confirm scores are improving.

## If that didn't work

* If Hallucination failures persist after adding FAQs, check whether the FAQ answers are specific enough. Vague answers give Maya room to improvise. Rewrite them with concrete, factual statements.
* If transfer-rule changes don't reduce Escalation failures, consider using the **Trust & Safety** plain-English rules field to add additional constraints (for example, "Never book a job if the caller mentions insurance or a lawsuit — transfer to Mike instead").
* If quality remains low across all axes despite FAQ and rule updates, contact Servinix support with a list of the affected call IDs so the team can review Maya's scoring model for that call type.

## Related

* [Review call quality](/ai-assistant/using-the-assistant/review-call-quality)
* [Review Maya Asks](/ai-assistant/using-the-assistant/review-maya-asks)
* [Configure FAQs for the AI CSR](/ai-assistant/setup/configure-faq-for-ai-csr)
* [Set transfer rules for Maya](/ai-assistant/setup/set-transfer-rules)
* [Configure Maya settings](/ai-assistant/setup/configure-maya-settings)