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

# Testing Your Agent

> How to properly test your agent before going live — checklists, scenarios, and red flags.

## Why Testing Matters

<Warning>
  **The risks of skipping testing:**

  * Agent gives incorrect information to real customers — damages trust
  * Edge cases cause confusion or repetitive loops
  * Functions fail silently — customers get no answer
  * You discover problems after launch, not before

  **Rule of thumb:** Run at least 20 test conversations before going live.
</Warning>

***

## Three Levels of Testing

### Level 1: Browser Test (Fastest)

Test directly in the Studio — no phone number required.

<Steps>
  <Step title="Open the Studio">
    Go to [studio.talkifai.dev/agents](https://studio.talkifai.dev/agents) and open your agent.
  </Step>

  <Step title="Click Try Now">
    Find the **Try Now** button on the agent card. You'll be redirected to the demo page at `/demo`.
  </Step>

  <Step title="Start Voice Test">
    Click **Start Voice Test** button. The browser will ask for microphone permission — click Allow.
  </Step>

  <Step title="Have a conversation">
    Behave like a real user. Watch the live transcription to see what the agent hears and understands. Take notes on anything that feels off.
  </Step>
</Steps>

<Info>
  **Note:** You're testing YOUR agent, not a demo agent. All conversations use your configured agent settings.

  **Text agents:** If your agent uses Text architecture, use the **Chat** tab to test text-based conversations instead of voice.

  **Session Duration:** Demo sessions last up to 30 minutes. The timer will show remaining time.
</Info>

<Tip>
  Use the mute button to temporarily silence your microphone without ending the call.
</Tip>

### Level 2: Phone Test

Call your agent on a real phone to verify carrier setup, audio quality, and latency.

1. Assign a phone number to the agent — [Telephony Guide →](/platform/telephony)
2. Call from your own phone
3. Verify: audio quality, response latency, call routing

### Level 3: Team Testing

Before going to production, have 2–3 team members test independently.

* Give them specific scenarios to test
* Collect structured feedback
* Fix issues before real customers encounter them

***

## Complete Testing Checklist

### ✅ Basic Functionality

<Check>Agent answers the call correctly?</Check>
<Check>Introduction is correct (name, company)?</Check>
<Check>Voice sounds right (speed, tone, quality)?</Check>
<Check>Language is correct?</Check>
<Check>Agent understands its role?</Check>

### ✅ Conversation Quality

<Check>Handles normal questions accurately?</Check>
<Check>Remembers context within the call (e.g., uses name after asking)?</Check>
<Check>Responses sound natural — not robotic?</Check>
<Check>Responses are concise — not overly long?</Check>
<Check>No unnecessary repetition?</Check>

### ✅ Functions (If Configured)

<Check>Functions are called at the right time?</Check>
<Check>Function results are used correctly in the response?</Check>
<Check>Function failures are handled gracefully?</Check>
<Check>User is informed when the agent is fetching data?</Check>

### ✅ Edge Cases

<Check>Handles unknown questions gracefully?</Check>
<Check>Redirects when user goes off-topic?</Check>
<Check>Stays calm with an angry or rude user?</Check>
<Check>Handles very fast speech?</Check>
<Check>Handles silence of several seconds?</Check>
<Check>Works with background noise?</Check>

### ✅ Escalation & End

<Check>Escalation triggers correctly when expected?</Check>
<Check>End call works properly?</Check>
<Check>Transfer works (if configured)?</Check>

***

## Test Scenario Library

Copy and run these scenarios during testing:

### Customer Support Scenarios

```
Scenario 1 — Happy Path:
"I'd like to track my order. The order ID is ORD-12345."

Scenario 2 — Information Not Available:
"Can I get an invoice from January 2022?"

Scenario 3 — Angry Customer:
"This is the third time I'm calling about the same issue and nothing has been fixed!"

Scenario 4 — Escalation Request:
"I need to speak with a manager immediately."

Scenario 5 — Off-Topic:
"By the way, what city are you based in?"
```

### Appointment Booking Scenarios

```
Scenario 1 — Simple Booking:
"I'd like to book an appointment for tomorrow."

Scenario 2 — Fully Booked:
"Is there anything available today?" (test when slots are full)

Scenario 3 — Reschedule:
"I need to change my existing appointment."

Scenario 4 — Wrong Context:
"I'd like to book an appointment for my dog."
(For a human clinic — does it handle gracefully?)
```

### Sales Scenarios

```
Scenario 1 — Qualified Lead:
"Yes, I'm very interested. My budget is around $500/month."

Scenario 2 — Not Interested:
"No thanks, I'm not looking for anything right now."

Scenario 3 — Price Negotiator:
"Can you just do it for $50?"

Scenario 4 — Competitor Mention:
"I'm already using [Competitor Name]."
```

***

## Red Flags to Watch For

<AccordionGroup>
  <Accordion title="🔴 Agent loses context mid-conversation">
    **Symptom:** Asks the customer's name twice, or doesn't use the name after learning it.

    **Fix:** Add to system prompt: "Use the customer's name throughout the conversation once you know it."
  </Accordion>

  <Accordion title="🔴 Responses are too long">
    **Symptom:** Agent speaks 5–6 sentences per response.

    **Fix:** Add to prompt: "Every response must be a maximum of 2 sentences. Never exceed 30 words."
  </Accordion>

  <Accordion title="🔴 Confused by edge cases">
    **Symptom:** When asked something out of scope, the agent repeats itself or gives a strange answer.

    **Fix:** Add an explicit edge case handler to the prompt.
  </Accordion>

  <Accordion title="🔴 Functions called at wrong time">
    **Symptom:** Order status function fires before the user asks for it.

    **Fix:** Improve the function description — be very specific about when it should be triggered.
  </Accordion>

  <Accordion title="🔴 Unnatural-sounding numbers or abbreviations">
    **Symptom:** "Order I.D. one two three four" sounds choppy.

    **Fix:** Add to prompt: "Spell out all numbers and abbreviations naturally — say 'order number one-two-three-four' not 'order ID 1234'."
  </Accordion>
</AccordionGroup>

***

## Performance Benchmarks

After going live, monitor these metrics in [Analytics →](/platform/analytics):

| Metric                         | Target            | If Below Target                               |
| ------------------------------ | ----------------- | --------------------------------------------- |
| **Call completion rate**       | > 80%             | Agent is frustrating — review the prompt      |
| **Function call success rate** | > 95%             | Check API/webhook connection                  |
| **Escalation rate**            | \< 20%            | Too many escalations = prompt is unclear      |
| **Average call duration**      | Use-case specific | Too short = unhelpful; too long = inefficient |

***

## Production Ready Checklist

Only go live when all of these are checked:

<Check>20+ test conversations completed</Check>
<Check>All basic scenarios pass</Check>
<Check>All edge cases handled correctly</Check>
<Check>All functions working properly</Check>
<Check>Voice sounds natural and appropriate</Check>
<Check>Escalation path works correctly</Check>
<Check>At least 2 team members have tested independently</Check>
<Check>Agent is Activated in the Studio</Check>

***

## Next Steps

<CardGroup cols={2}>
  <Card title="Connect a Phone Number" icon="phone" href="/platform/telephony">
    Set up real calling after testing is complete.
  </Card>

  <Card title="Monitor with Analytics" icon="chart-bar" href="/platform/analytics">
    Track live calls and continuously improve.
  </Card>
</CardGroup>