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

# Test your AI agent before going live

> Simulate real customer conversations to validate your BluBash AI agent's responses, verify skill execution, and catch issues before the agent goes live.

Testing your AI agent before activating it is an essential step. The test environment lets you simulate real customer conversations, verify that skills trigger correctly, and catch any issues before they affect your actual customers. You can test as many times as you need — test conversations do not appear in your history and have no impact on live data.

## Why testing matters

* **Validate training** — Confirm the agent learned the right information.
* **Test skills** — Verify that automatic actions execute as configured.
* **Catch problems early** — Find issues before they reach real customers.
* **Refine behavior** — Adjust responses, tone, and skill conditions.
* **Ensure quality** — Deliver a great experience from the first real interaction.

## How to open the test window

<Steps>
  <Step title="Open AI Agents">
    In the left sidebar, click **AI Agents**.
  </Step>

  <Step title="Select your agent">
    Click on the agent you want to test.
  </Step>

  <Step title="Click Test Agent">
    Click the **Test Agent** button in the top-right corner of the screen.
  </Step>
</Steps>

A chat window will open where you can send messages and receive the agent's responses in real time.

## How to run a thorough test

### 1. Simulate real customer messages

Type messages as if you were a real customer contacting your business. Start with the most common inquiries your team handles.

**Example messages to try:**

```
Hi, I'd like to know more about your products
```

```
How much does the premium plan cost?
```

```
I need help with my order
```

```
I want to speak with a human agent
```

***

### 2. Evaluate the responses

Analyze each response the agent gives:

* Is the response accurate and relevant?
* Does the tone match the configured style?
* Is the information current and correct?
* Is the reply clear and easy to understand?
* Is the agent polite and professional?

***

### 3. Verify skills are triggered

Check that the skills you configured execute correctly during the test.

**How to identify skill execution:** When a skill runs, it appears as a **blue box** in the chat. It shows which action was taken. These boxes are only visible to you in the test view — your customers will not see them.

**Skills to verify:**

* **Transfers** — Send a message like *"I want to speak with sales"* and confirm the agent routes to the correct team.
* **Labels** — Say something like *"I'm interested in the premium plan"* and verify the label is applied.
* **Notes** — Provide relevant customer information and check that a note is created with the correct content.

***

### 4. Test specific scenarios

Work through different situations that might occur in real conversations.

**Scenario 1: Simple question**

```
Customer: What are your business hours?
Expected: Agent replies with the correct hours
```

**Scenario 2: Purchase intent**

```
Customer: I want to buy the premium plan
Expected: Agent qualifies the lead and transfers to the sales team
```

**Scenario 3: Customer with a problem**

```
Customer: My product is not working
Expected: Agent collects relevant information and transfers to support
```

**Scenario 4: Request for a human agent**

```
Customer: I want to speak with a person
Expected: Agent transfers immediately
```

**Scenario 5: Out-of-scope question**

```
Customer: What's the weather like today?
Expected: Agent redirects the conversation to relevant topics
```

## What to look for

### Response quality

| Criteria         | What to check                             |
| ---------------- | ----------------------------------------- |
| **Accuracy**     | Are the facts correct?                    |
| **Relevance**    | Does the response address what was asked? |
| **Completeness** | Is the answer sufficiently detailed?      |
| **Clarity**      | Is it easy to understand?                 |

### Agent behavior

| Criteria            | What to check                                        |
| ------------------- | ---------------------------------------------------- |
| **Tone**            | Does it match the configured style?                  |
| **Empathy**         | Does the agent acknowledge the customer's situation? |
| **Professionalism** | Does it maintain an appropriate posture?             |
| **Naturalness**     | Do the responses feel human?                         |

## Fixing issues found during testing

If you identify a problem, return to the appropriate tab and make adjustments before testing again.

<Steps>
  <Step title="Add missing training">
    Go to the **Training** tab and add the information the agent lacked.
  </Step>

  <Step title="Refine skills">
    Go to the **Skills** tab, adjust trigger conditions, and fix any actions that did not work as expected.
  </Step>

  <Step title="Update settings">
    Go to the **Settings** tab and adjust tone, emoji use, response splitting, or topic restrictions.
  </Step>

  <Step title="Update the profile">
    Go to the **Profile** tab to refine the agent's behavior description or purpose.
  </Step>

  <Step title="Test again">
    Repeat the test after each adjustment. Keep refining until the results are satisfactory.
  </Step>
</Steps>

<Tip>
  A well-tested agent delivers a better customer experience and reduces the need for human interventions. Invest time in testing before activating — it pays off quickly.
</Tip>