Topic Analysis – Blip | Blip Help

Improve automatic resolution in conversational flows.

Index:

Introduction
How Topics Are Generated
Accessing Topic Analysis
Topic Filters
Using Topic Analysis
Topic Details
Exporting Reports
Contacts
Conversation History
Additional Information

Introduction

Managing all the conversations that happen every day in an Intelligent Contact can be complex, especially when you combine AI Agents, deterministic flows (Builder), and human service (Desk). Manually reviewing each interaction is not feasible, and the variety of available reports can make decision-making harder.

With that in mind, we created the Topics feature, a tool that automatically identifies the main themes users talk about during conversations with your Intelligent Contact. The analysis is based on a statistical sample of conversations that reliably represents all users and considers every type of automation: Agent, Builder, and Desk with human agents.

With this feature, you can understand which subjects are most frequent, where automatic resolution is failing, and when it is necessary to hand off to humans. This allows you to act more strategically to optimize your brand’s conversational experience, using a continuous improvement approach.

How Topics Are Generated

The Topics Analysis uses clustering algorithms with Machine Learning to identify recurring themes in conversations. The analysis takes into account:

messages sent by the user;
responses generated by automations (AI Agents, Builder flows, human agents);
the context of the conversation within an inactivity window of up to 4 hours.

The themes are grouped by similarity, which provides a clear view of the main subjects that emerge over time. A single conversation may cover more than one topic.

To generate these insights, the Topics Analysis does not process 100% of all conversations, but rather a representative sample.

We use statistical sampling techniques that split the conversations into two main groups (with handoff to a human agent and without handoff) and select an appropriate number of dialogues in each group.

This ensures that the results reflect the real behavior of the whole base, allowing comparisons such as:

differences in sentiment between conversations with and without handoff;
most frequent topics in each type of service.

In simple terms: you don’t need to review every single conversation to get a reliable view of the main topics.

Note: A conversation is defined as a continuous session that remains active as long as there are message exchanges between the user and the bot. The conversation is considered finished when there is a period of 4 hours with no interaction from either side.

Accessing Topic Analysis

To access the Topic Analysis for conversations, follow the path:

Navigation: Conversations > Topics

Topic Filters

Filter Topics by Date

Click the date dropdown menu.
Set the desired date range or manually enter the start and end dates.
Click Apply.

This filter lets you customize the analysis period using a calendar selector. By default, it is set to the last 7 days.

The filter uses the start date of the interaction (whether initiated by the user or the bot) to include the entire conversation in the analysis, from the first message to the last, even if it spans multiple messages or several days, until it is closed by the 4-hour inactivity window.

Filter Topics by Period

Click the period dropdown menu.
Select a predefined range (e.g., Last 7 days, Last 15 days, or Last 30 days).

By default, the filter is always set to the last 7 days.

Using Topic Analysis

On the main Topics screen, you will see a list of the most frequent subjects found in your conversations. This list is ranked based on two main aspects:

Conversation volume: How many sessions had that topic identified.
Human escalation: How often those sessions needed human support.

With this view, you can quickly spot bottlenecks and improvement opportunities in any part of your operation: knowledge base, Builder flows or human service performance.

Key Definitions

Topics: Topics are automatically generated labels that represent what was discussed in the conversation. They are created by a Machine Learning model based on a statistically representative sample of conversations.

A single conversation can have one or several topics, depending on what the user and the bot talked about during the session.

Example of Application

If the topic “Delivery time” appears with high volume and handoffs, this may indicate:

Lack of content in the Agent automation: The knowledge base or flow needs to be updated to include this type of answer.

Complex topic: It may be better handled by humans. Adjust the redirection rule by intent and keywords.

Adjustment to the guidelines: The prompt needs to be revised to better handle this topic.

Sorting the List

You can organize the list of topics according to the available columns, making data analysis easier. Follow the steps below to sort the list:

Identify the column you want to sort: The columns can include information such as conversation volume, human handoffs, among others.

Click the column header: Each column header allows you to sort the list; just click the name of the desired column.

Adjust the sorting as needed: When you click a column header once, the list will be sorted in ascending order (A–Z or smaller to larger). Clicking again, the list will be sorted in descending order (Z–A or larger to smaller).

Browse through the list pages: If there are more items than can be displayed on a single page, use the pagination controls at the bottom to navigate between the pages of the list.

Topic Details

When you click the “Open topic” button next to any topic in the list, you will be redirected to the detail page for that topic. On this page, you can:

Topic summary: Shows, in a sentence of up to 500 characters, what users talk about the most regarding this topic. When there is not enough information in the dialogue to generate this summary, the field does not appear on the screen.

Related metrics:

Conversations about the topic: Total volume of conversations that addressed this topic in the selected period, based on a representative sample of conversations.

Containment: Percentage of conversations in the sample, in the selected period, about this topic that were resolved only by automation, without human support.

Human support: Percentage of conversations in the sample, in the selected period, about this topic that were routed to human support.

Contacts

You can view the list of contacts who, within the analyzed sample, talked about the selected topic in the filtered period.

Only the sessions for which it was possible to calculate some analysis value (for example, sentiment) are displayed. Conversations about the topic that do not have this type of information do not appear in the contacts list.

For each session, the following information is available:

Channel: Indicates on which channel the conversation took place (e.g., WhatsApp, Blip Chat, Telegram, etc.)

Session date: Start date of the conversation (first interaction from the user or the bot) in which this topic was identified.

Support: Indicates whether the conversation was resolved only by automation (without human support) or whether it was handed off to human support.

Sentiment: Identifies the sentiment in the contact during the session, with the classifications: Positive, Negative and Neutral.

Contacts are displayed in a list ordered by the most recent conversation session date, where each session is closed after 4 hours without activity.

Contact Filters
The contact list can be refined with specific filters for:

Channel type: Allows you to segment contacts based on the communication channel (e.g.: WhatsApp, Blip Chat, Telegram, Messenger, etc.).

Sentiment: Filters contacts according to the predominant sentiment identified during the session (e.g.: positive, neutral, and negative).

Type of support: Filters by the type of automation used in the conversation, allowing you to choose between conversations handled only by automation or those that were routed to human support.

Sorting: Adjusts the display order according to specific criteria (e.g.: most recent, least recent).

Conversation History

To deepen the analysis of a topic, you can click “Open conversation” next to the contact in the list. When you do this, a side tab will open with the full history of the conversation, already positioned on the first message of the session in which that topic was identified. From there, you can browse through the conversation to better understand the context in which the topic appeared.

Navigating Conversation History
The history allows you to understand the full context of the interaction between the contact and your company at the moment when the topic was addressed.

You can scroll the mouse up or down to view earlier or later interactions in that same session.

If they exist, it will also be possible to access previous conversations from this same contact with your brand, forming an infinite history of the customer’s interactions. The “Summary” tab is related to the conversation session in which that topic occurred.

Interaction Identification

In the conversation history, it is possible to identify the origin of each interaction, allowing you to distinguish between the types of actions performed. The interactions are categorized as follows:

IA Automated Skills interactions:

Represent responses generated by the Artificial Intelligence agent.

They are identified by the “Source” button.

Star icon

Flow Automated Skills interactions:

Refer to deterministic actions configured in the conversation flows.

Robot icon

Human Automated Skills interactions:

Indicate actions performed by human agents.

These interactions occur after the handoff of the conversation and the generation of a support ticket.

Agent icon

Conversation Summary

This feature provides a complete analysis of the context of each interaction, offering a detailed view of the contact’s journey with your brand, from automated responses to human actions.

Summary

Next to the conversation history, a summary is displayed about the conversation session currently being viewed. If you navigate to a previous or later session, the summary is automatically updated to reflect the information corresponding to the new session.

Sentiment

In the summary, it is possible to view the contact’s sentiment during that conversation session. This information helps you understand how the customer was feeling about the conversational experience with your brand at that moment.

The sentiments are classified into three categories: positive, negative and neutral, according to the criteria below:

Positive: Indicates that the contact had a satisfactory or positive experience. Examples of expressions:

Words of Thanks and Gratitude:
"Thank you very much", "I appreciate the help".

Direct Compliments:
"Excellent service", "You are great".

Comments Highlighting Positive Aspects:
"Quick response", "Efficient support".

General Satisfaction Expressions:
"I am very satisfied", "That was great".

Detailed Positive Feedback:
The solution was quick and effective, I am extremely satisfied with the support.

Negative: Indicates that the contact had an unsatisfactory or frustrating experience. Examples of expressions:

Use of Offensive or Vulgar Language:
"This is really bad", "What a mess".

Phrases that Indicate Clear Dissatisfaction:
"This is unacceptable", "I am very disappointed".

Expressions of Frustration:
"This is very frustrating", "I can’t stand this anymore".

Specific Complaints:
"The service was terrible", "I didn’t like the support".

Detailed Negative Feedback:
"The solution was slow and ineffective, I am extremely dissatisfied with the support".

Neutral: Indicates that the contact did not show clear emotions of satisfaction or dissatisfaction. Examples of expressions:

Objective or Factual Communication:
"I understand", "Right", "Ok".

Simple Questions or Requests:
"What is the next step?", "How do I access it?".

General Comments without Evident Emotions:
"I received the information", "Everything is in order".

Contact details

In the Contact Details tab, you can view information such as name, phone number, e-mail, user ID and the date of the first contact.

If the contact is not found, the name will be replaced by a system identification key.

Additional Information

Report data is updated with a delay of D-1 (e.g.: a report exported on 10/17/2024 will contain data up to 10/16/2024).

The reports’ time zone follows the one configured in your profile.