Blip Copilot

Index:

1. Introduction
2. How does Blip Copilot work?
   1. Lead Score
   2. Smart Summaries
   3. Response Suggestions
3. Activation
   1. Release of Blip Copilot in the contract
   2. Agent Changes
      1. Lead Score
      2. Smart Summaries and Response Suggestions
4. Configuration
   1.    Response Suggestions
   2. Best Practices
   3. Knowledge Base
   4. Sending the knowledge base
5. Initial Messages
6. Advanced Settings
7. Blip Copilot Reports
8. Usage Demonstration

 

Introduction

BLip Copilot is a generative artificial intelligence assistant (LLMs) that aids attendants in customer service or sales operations by prioritizing tickets, providing consumer context, and delivering complete and precise responses based on the conversation context.

How does Blip Copilot work?

Blip Copilot has two functionalities, each with a specific purpose, namely:

• Lead Score functionality;
• Smart Summaries Functionality;
• Response Suggestions Functionality;

Below we will see how each of them works.

 

Lead Score

The lead score feature was created to assist the attendant in prioritizing service requests according to operational needs.

To prioritize tickets, keywords and tags are provided to the AI, each acting at a different layer as outlined below:

1st Layer | Keywords: These are words that the brand recognizes as signaling high or low priority in a service request. For example, the word "Tax" in financial operations typically indicates that the person is knowledgeable about investments, as they inquire about brokerage fees and other fees that financial institutions charge. Therefore, for the brand, it's an indicator that this person is more ready to make a purchase. Thus, the brand informs the AI that when it encounters the keyword "tax," it should set the ticket as high priority. The same applies to words indicating low priority.

2nd Layer | Tags: Similar to keywords, tags are terms that the brand informs the AI about. However, instead of signaling high or low priority, the tag informs the AI that the context of the service request classified with a specific tag is of high or low priority. With this information, the AI accesses the conversation history classified with the provided tags and learns which contexts within the conversation it should seek to define the ticket's priority. In other words, for the second layer, the AI is fed back each time a ticket is classified with a high or low priority tag, thereby increasing the AI's ability to deliver increasingly accurate prioritizations.

 

Smart Summaries

The smart summaries functionality was created to provide quick contexts about the consumer to the agent.
To deliver these contexts, the generative AI analyzes the conversation history of this consumer with the intelligent contact (pre-overflow and human service) and generates a summarized context, delivering it to the agent as soon as he opens the ticket of this consumer.
The summary brings information such as:

• Data provided by the customer to the bot or in previous conversations;
• Reason for the current contact;
• Number of calls from this consumer in the last 30 days;
• Conversation sentiment.

The summary is automatically generated for every ticket opened by the agent with the functionality enabled.

Response Suggestions

The response suggestions functionality was created to help the agent respond to the consumer in complex contexts and subjects that are unknown or new to him.
To deliver response suggestions, the AI analyzes the conversation context and the content of the previously configured knowledge base with the company's context and then delivers to the agent who requested the response suggestion two options so that he can choose the one that makes more sense with the moment.
The response suggestion functionality delivers responses on demand, that is, for it to generate suggestions, it is necessary for the agent to request a response by clicking on the button available on their Blip desk.
Before sending the response, the agent can make edits and customizations as deemed necessary.

 

Activation

Release of Blip Copilot in the contract

As it is a product sold separately, to activate it, it is necessary to contract it. After contracting, fill out the activation form available here.

This form must be filled out by Blip personnel (KAMs, CSMs, AMs, etc.) responsible for the client's account.

In this form, it will be necessary to fill in the following information:

• Company
• Deal gain date
• Company ID on the platform
• Company ID in Hubspot
• Hubspot Deal Link
• Responsible KAM/AM
• Responsible CSM
• Usage context
• Company's responsible person's email
• Emails of service managers
• Functionality to be released
• Emails of the agents who will use the functionality

Agent Changes

Lead Score

The modification of lead score attendants will occur upon request. If you already have everything enabled and need to include or remove any lead score attendant, fill out this form.

In this form, it will be necessary to fill in the following information:

• Company
• Company ID in Hubspot
• Hubspot Deal Link
• Functionalities to be changed
• Emails of the agents to add or remove from the functionalities

Smart Summaries and Response Suggestions

In the Smart Summaries and Response Suggestions features, the manager or person with access to the attendant settings in the Portal can enable or disable attendants as needed, without having to request it from Blip.

To do this, follow the steps below:

1. Access your portal.
2. Select the contract that contains your chatbot(s).
3. Then choose the chatbot for which you want to configure.
4. Access the "Atendimento" (Assistance) menu.
5. Go to the "Atendentes" (Attendants) side menu.
6. Select the attendants you want to change permissions for.
7. Then, on the right side, click on the person icon with the key.
8. The permissions screen will be displayed. Adjust the permissions for Response Suggestions and Smart Summaries as needed.
9. Then click back on the icon next to the word "permissions."
10. And that's it, your attendant settings have been adjusted.

 

Configuration

Response Suggestions

Only the response suggestion functionality needs configuration, and it must be set up by a BOT, meaning each BOT must undergo the following configuration process:

1. Access your portal;
2. Select the contract that contains your chatbot(s);
3. Then choose the chatbot for which you want to perform the configuration;
4. Next, access the "Service" menu;
5. Then select the Blip Copilot menu;
6. On this screen, click on the selector (switch) to enable Blip Copilot on this bot;
7. Once enabled, fill in the form data according to the table below.

 

8. After filling in the client information, save it.

 

Best Practices

The fields in the form are text fields, so pay attention to the correct spelling of the responses. In the "Company Name" field, enter only the company name with the initial letters capitalized. In the "Demand(s) for assistance in Blip Desk" field, the response needs to be objective and direct.

If it is necessary to add new instructions in the "Additional Guidelines" field, here are some recommended best practices:

• Instructions need to be clear and concise: Formulate instructions in a straightforward manner, avoiding ambiguities and unnecessary complexities.
• Limit the scope of responses: If necessary, instruct the copilot that the response suggestion needs to be restricted to a specific domain, topic, or context to avoid out-of-scope responses.
• Avoid inserting long instructions: The copilot tends to work better with direct and short instructions, so avoid including unnecessary information.

After filling in the fields, click the "Save" button to save the changes.

Knowledge Base

For the Blip Copilot to be able to suggest accurate responses to the attendants, it is essential that the knowledge base is well-fed with relevant information. This base serves as a data source for the copilot to formulate response suggestions for human attendants to address the demands that arrive at the Blip Desk.

Therefore, it is important to include detailed information about products and services that are handled by the attendants, such as benefits, prices, technical specifications, and other relevant information.

The base needs to be sent in .txt or .tsv format, and the document structure needs to be organized so that the CONTEXT NAME (always in uppercase letters) | Texts representing such context are on the same line.

It is important that the assembly and categorization of the base are done by someone who knows well what the main demands that human attendants receive in the overflow are.

Below, we have inserted two examples of how the base should be structured.

Note that in the 'Plain Text Model' it follows the pattern: CONTEXT NAME (always in uppercase letters) | And texts representing such context on the same line. The base in question has 6 lines and each line represents 1 context.

In the 'Spreadsheet Model', it will follow the same writing pattern but separated into two columns, where in the left column, the CONTEXT NAME (always in uppercase letters) should be filled, and in the right column, texts representing such context on the same line.

 

Access the .txt model of the knowledge base here.

 

 

Access the .tsv model of the knowledge base here..

 

Sending the knowledge base

After ensuring that the base has all the necessary information, upload the file in the Knowledge Base section. Select the .txt or .tsv file containing the knowledge base and then click on "Upload".

It is important to note that to add, modify, or remove any information from the knowledge base, you need to make these changes locally in the file and then upload that file. This new file will be the new knowledge base, as the previous .txt or .tsv file will be replaced. Furthermore, the content of the base must be text-only, as the copilot cannot extract information from images, links, videos, and the like.

 

 

Initial Messages

The next step is to configure the initial messages, which are sent by the human agent to the end consumer after the transfer. By default, two messages are added, but it is possible to modify or add more.

To configure, access the "Initial Messages" section

The minimum number of initial messages is 1, and the maximum recommended is up to 3 messages.

To add a new message, fill in the field with the message and then click on "Add Message".

To delete, simply click on the trash can icon located next to the message.

To save the changes, after making adjustments to the initial messages, click on "Save".

Note: Initial messages will be suggested when the conversation in the Desk has only messages from the customer.

 

Advanced Settings

To access the advanced settings, simply select the "Advanced Settings" tab.

In the advanced settings, you can adjust the temperature of the generative AI.

The temperature controls the degree of accuracy of the responses suggested by the copilot. A low value indicates that the responses generated by the model will be more precise and predictable, staying closer to the information contained in the knowledge base. A high value indicates that the copilot will be more likely to generate creative and unpredictable responses, potentially deviating from the scope of the information in the base.

This value can be set between 0.1 and 0.5.

WARNING: The default value is 0.2, only change this setting if you are certain it will not interfere with the quality of the responses suggested by the Copilot.

 

If you've made it this far, you've completed the Blip Copilot setup and can now begin using it in your operation.

 

Blip Copilot Reports

To access the metrics page, go to the left-hand side menu and click on "Export Reports"

On this page, you can download Blip Copilot metrics, which include suggested and selected responses, as well as a summary of activities that occurred within a period of up to 60 days prior to the current date.

To download the usage metrics, simply select a time interval (up to a maximum of 60 days prior to the current date) and click on "Generate Metrics". Afterwards, a .xlsx file containing all the information will be downloaded to your machine.

 

 

Usage Demonstration

Below, see how the Blip Copilot works in practice.

 

For more information, access the discussion on the topic inour community or the videos on our channel. 😃