How does AI work in Virtual PBX scenarios?
Posted: Sun Jan 26, 2025 8:34 am
The capabilities of virtual PBX have long gone beyond automatic greetings and call routing. Today, artificial intelligence-based technologies are integrated into call processing scenarios that improve operator efficiency and reduce customer wait times.
In this article, we will look at how AI works and what benefits it brings to businesses using Virtual PBX services.
What is AI Call Handling?
AI call handling is represented by the "AI" element in Virtual PBX scenarios. It is designed to answer customer questions using artificial intelligence.
There are several AI models to choose from:
OpenAI
AI Response Request. Turbo Model 128k Tokens input — 0.0008/token output — 0.003/token input — 0.0004/token output — 0.0015/token
AI Response Request. Model "4+" 128 thousand tokens input — 0.0008/token output — 0.003/token input — 0.0004/token output — 0.0015/token
AI Response Request. Model "4-mini" input — 0.00005/token output — 0.00018/token input — 0.000025/token output — 0.00009/current
GigaChat
Suitable for solving simple tasks that require high speed. Cheap and less demanding on hardware resources.
YandexGPT-4
This model is capable of generating text, answering questions, creating content, and performing complex natural language processing tasks with high accuracy.
How to analyze calls using AI?
In the Virtual PBX scenario, the "AI" element is linked to the "Question-Answer" element.
To start using artificial intelligence to handle calls, you need to place the "AI" element on the diagram and configure the following parameters.
Waiting time
Here you set a time interval from 5 to 15 seconds, after which it is ghana phone number data considered that the AI has not found an answer or its answer is not recognized.
AI Model
A model suitable for the intended use is selected.
Temperature
You can set the temperature in the range from 0 to 1, the recommended value is 0.7. In this case, zero corresponds to the most probable answers and speech constructions with a minimum of creativity, and one - to more creative and unpredictable answers.
Number of tokens
Tokens are units that measure the amount of text in a given context (more on that below) and AI responses. Each token is approximately 2-3 characters. By setting the number of tokens, you can regulate the "talkativeness" of the AI.
Message
In this field, you need to enter the variable {{stt_answer}} so that the AI takes the question it needs to answer from the "Question-Answer" block.
Context
The length of the context is limited by the allowed number of tokens. The context contains the logic by which the AI will process the information received from the client.
For example, the context could indicate that the AI is a secretary who must identify the topic of a customer's call, briefly list the main topics of calls, and what requests are typically received within each topic.
At the same time, AI can be configured to conduct a dialogue with the client, taking into account that the context of all questions asked is the same.
Next, you need to build the entire scenario of the Virtual PBX using AI, more on that in the next section.
In this article, we will look at how AI works and what benefits it brings to businesses using Virtual PBX services.
What is AI Call Handling?
AI call handling is represented by the "AI" element in Virtual PBX scenarios. It is designed to answer customer questions using artificial intelligence.
There are several AI models to choose from:
OpenAI
AI Response Request. Turbo Model 128k Tokens input — 0.0008/token output — 0.003/token input — 0.0004/token output — 0.0015/token
AI Response Request. Model "4+" 128 thousand tokens input — 0.0008/token output — 0.003/token input — 0.0004/token output — 0.0015/token
AI Response Request. Model "4-mini" input — 0.00005/token output — 0.00018/token input — 0.000025/token output — 0.00009/current
GigaChat
Suitable for solving simple tasks that require high speed. Cheap and less demanding on hardware resources.
YandexGPT-4
This model is capable of generating text, answering questions, creating content, and performing complex natural language processing tasks with high accuracy.
How to analyze calls using AI?
In the Virtual PBX scenario, the "AI" element is linked to the "Question-Answer" element.
To start using artificial intelligence to handle calls, you need to place the "AI" element on the diagram and configure the following parameters.
Waiting time
Here you set a time interval from 5 to 15 seconds, after which it is ghana phone number data considered that the AI has not found an answer or its answer is not recognized.
AI Model
A model suitable for the intended use is selected.
Temperature
You can set the temperature in the range from 0 to 1, the recommended value is 0.7. In this case, zero corresponds to the most probable answers and speech constructions with a minimum of creativity, and one - to more creative and unpredictable answers.
Number of tokens
Tokens are units that measure the amount of text in a given context (more on that below) and AI responses. Each token is approximately 2-3 characters. By setting the number of tokens, you can regulate the "talkativeness" of the AI.
Message
In this field, you need to enter the variable {{stt_answer}} so that the AI takes the question it needs to answer from the "Question-Answer" block.
Context
The length of the context is limited by the allowed number of tokens. The context contains the logic by which the AI will process the information received from the client.
For example, the context could indicate that the AI is a secretary who must identify the topic of a customer's call, briefly list the main topics of calls, and what requests are typically received within each topic.
At the same time, AI can be configured to conduct a dialogue with the client, taking into account that the context of all questions asked is the same.
Next, you need to build the entire scenario of the Virtual PBX using AI, more on that in the next section.