# Using AI Insights

## How to benefit from AI Insights

Once you have completed the setup and Insights are active, the AI Insights dashboard gives you a comprehensive view of your chat data. 

* **Measure the ROI of your chat agent/chatbot.** Automatically calculate cost savings by tracking "Solved" vs. "Unsolved" cases based on real user outcomes.
* **Identify friction.** Use the Escalation Rate and Sentiment analysis to pinpoint exactly where users struggle or require human intervention.
* **Discover trending topics.** Monitor recurring issues in real-time to prioritize knowledge base updates or service improvements.
* **Performance audit.** Dive into into specific categories (e.g., "Billing" vs. "Technical Support") to compare satisfaction rates across different parts of your business.
* **Validate AI reasoning.** View the specific logic behind every identified topic or sentiment to ensure your chat agent's/bot's behavior aligns with your service standards.

## Breakdown of the AI Insights dashboard overview and metrics <a href="#dashboard-overview" id="dashboard-overview"></a>

### **Total chats** <a href="#id-1-total-chats" id="id-1-total-chats"></a>

Displays the total number of chats analyzed in the selected period, with a comparison to the previous period to help you track trends in your chat volume.

{% hint style="info" %}
We only process chats containing analyzable data, so you may see a difference between your total history and your analyzed results. This selection is based on criteria like message count, which we refine periodically.
{% endhint %}

<details>

<summary>Criteria for if a chat should be analysed</summary>

When all the following criteria has been met the chat will be analyzed:

* The chat contains at least three messages
* The chat includes at least one bot or agent message
* The chat includes at least one user message of type `Text`

</details>

### **Cost savings** <a href="#cost-savings" id="cost-savings"></a>

Cost savings are calculated by multiplying the number of solved cases by the cost per case.

By clicking the cog wheel icon in the Cost Savings box, you can access configuration options, including:

* Manually setting the cost of a case
* Changing the currency for the metric

### **Resolution rate** <a href="#resolution-rate" id="resolution-rate"></a>

How it’s calculated: Each conversation is classified based on whether a clear outcome was reached in the chat. If the issue is explicitly resolved, it’s marked as “Solved” Otherwise, it’s “Unsolved”.

* **Solved**: A conversation is Solved if the bot provides a clear solution or actionable next step and the user accepts it. Crucially, if the bot delivers an answer and the user simply stops responding without objecting, it is still categorized as a success.
* **Unsolved:** A conversation is Unsolved if the bot fails to offer a resolution or if the user explicitly states the issue persists. Furthermore, if a user leaves the chat while the bot is still troubleshooting or asking questions, the interaction is considered a failure regardless of the bot's intent.

**Resolution Over Time:** Shows how resolution types have changed across recent months. This helps you monitor improvements or spot periods needing attention.

### **Sentiment** <a href="#id-3-sentiment" id="id-3-sentiment"></a>

Displays the distribution of Positive, Negative, and Neutral conversations, helping you gauge overall customer satisfaction and changes in sentiment over time.

How it’s calculated: Only user messages are analyzed. Positive sentiment is flagged if the user expresses gratitude or satisfaction, especially at the end of the chat. Clear complaints or frustration are classified as negative. All other cases, or a mix, are considered neutral. Note: Bot messages are always ignored.

### **Most common topics** <a href="#id-4-most-common-topics" id="id-4-most-common-topics"></a>

How it’s calculated: AI Insights reviews each conversation and assigns up to two relevant topics that reflect the user's main issues. The first button the user clicks is prioritized, and explicit statements from the user determine which topics are set on the conversation. If no clear issue is mentioned, no topic will be set. Available topics are customized for each customer’s needs on the Config page.

### Suggested Topics

This metric uses autonomous AI analysis to identify emerging trends and themes that fall outside of your pre-defined list. The AI independently parses the natural language of the conversation to generate and apply descriptive tags. This is a generative process, meaning the AI is not limited to a list; it creates tags based on the actual context of the dialogue to complement your primary metrics.

### **Escalation rate** <a href="#id-4-most-common-topics-1" id="id-4-most-common-topics-1"></a>

The Escalation Rate measures the efficiency of your automation by tracking how many interactions required human intervention. This metric helps identify automation rates or scenarios where users prefer human assistance.

The "Escalated chats" card provides a breakdown of human interventions into two specific categories:

* **Total**: The aggregate number of chats that were escalated during the selected period.
* **Handovers**: The count of conversations successfully transferred to a live agent (e.g., via Live Chat).
* **Tickets created**: The count of conversations that resulted in a support ticket being logged for asynchronous follow-up.

> Note: The percentage indicators (e.g., `↓ -100%`) represent the change in volume compared to the previous time period.

### **Configuration: Tracking ticket creation**

While "Handovers" are usually tracked automatically by the chat system, Tickets created require specific configuration to be counted in the analytics.

For a ticket to register in the "Escalated chats" statistics, you must explicitly flag the interaction when a ticket is successfully generated. This is done by adding a `side_effect` to the bot's logic in the success scenario.

Add the following JSON syntax to the Ticket Success action/node in your scenario:

```json
{
  "side_effects": {
    "set_ticket_created": true
  }
}
```

<details>

<summary>How to implement ticket creation tracking</summary>

1. Go to the scenario **builder**.
2. Open a **scenario** triggered after a ticket is created.
3. Add an **intermediate card**.
4. Click on the **Show code** button located in the top-right corner.
5. Paste the code snippet provided above into the editor.

<figure><img src="/files/XBNkpSpijPGVxSVsP31Y" alt=""><figcaption></figcaption></figure>

</details>

### Language <a href="#id-4-most-common-topics-2" id="id-4-most-common-topics-2"></a>

Each chat is analysed to determine the primary language based on usage frequency. This metric provides a clear overview of the languages your customers use when interacting with your AI agent. This metric is available in the AI Insights dashboard as well as within individual chat logs.

### Topic overview <a href="#id-4-most-common-topics-2" id="id-4-most-common-topics-2"></a>

This section allows for a granular "drilldown" into specific categories to identify performance gaps or trends. While the global dashboard shows the big picture, the Topic Overview lets you filter data by a single subject (e.g., "Maintenance" or "Rental Agreements") using the dropdown menu on the left.

Once a topic is selected, the data reflects only conversations associated with that specific subject, subject to your global date and filter settings.

**Topic-Specific Metrics:** Three key cards display performance specific to the selected topic:

* **Total Chats:** The volume of conversations tagged with this topic during the selected period, including a comparison to the previous period.
* **Satisfaction Rate:** The percentage of chats within this topic that ended with positive user sentiment.
* **Escalation Rate:** The percentage of chats regarding this topic that were handed over to a human agent (if applicable) or where the bot could not resolve the issue.

### Chat history table <a href="#chat-history-table" id="chat-history-table"></a>

Below the main charts, you’ll find a detailed table of individual chat records, showing:

* **Username**: The participant in the conversation.
* **Resolution**: The outcome status (e.g., Solved, Incomplete, Unsolved, Unsolved / No Info).
* **Sentiment**: The tone of the conversation (Positive, Neutral, Negative).
* **Topics**: The main topics identified in the chat.

### Viewing chat details and insight reasoning <a href="#viewing-chat-details-and-insight-reasoning" id="viewing-chat-details-and-insight-reasoning"></a>

Clicking on any chat entry in the chat history table will open a detailed modal, allowing you to:

* **View the full conversation transcript:** Read the entire chat to understand the context.
* **See detailed insight breakdowns:** For each identified topic, sentiment, or resolution, you can see the AI’s reasoning, including the relevant parts of the conversation used for analysis.
* **Refresh insights:** If needed, you can refresh the insights to reanalyze the conversation with updated settings or data. If a chat has no Insights yet, you can use this button to manually trigger the chat for analysis.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.ebbot.ai/ebbot-docs/core-capabilities/ai-insights-for-chat/using-ai-insights.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.