Smart Conversations AI Engine
Agent-Based AI Framework Powering Smart Conversations 🌟
Introducing Smart Conversations: Vonage’s most advanced AI engine yet 🚀
Powered by cutting-edge NLU and autonomous agents, Vonage's Smart Conversations AI engine enables virtual assistants to understand context, make decisions, and take action, turning static scripts into intelligent, goal-driven conversations.
Did you know? đź’ˇ
While Vonage still supports the Traditional NLU engine, we recommend using either Hybrid NLU or Smart Conversations for more advanced, flexible, and context-aware experiences. These newer engines are designed to better handle complex user intents and dynamic interactions. đź’ˇ
Learn more here.
To fully leverage Smart Conversations, it’s essential to understand how AI Agents operate — how they perceive, reason, and act autonomously.
How an AI Agent works
Overview
đź’ˇ Smart Conversations uses two templatized AI Agents within each Virtual Agent, ensuring all interactions between the VA and end users are routed through one of these agents. This approach enhances both the VA development process and the user experience during conversations.
The two templatized AI Agents are available in the form of two Smart Nodes:
Smart Classification Node: Identifies the intent or the topic that the end user wants to discuss.
Smart Capture Node: Collects the parameter details needed to fulfill the end user's intent.
Why use Smart Conversations? âś…
Use Smart Conversations when:
You want to reduce VA design time by using LLM prompts for defining intents and parameters.
Allow for greater adaptability if end users may give multiple pieces of information at once, or change direction mid-flow.
You are building for voice and need robust handling of free-form input.
Pro Tip 🔥
If your current flows rely heavily on Collect Input and complex NLU mapping, switching to Smart Conversations can significantly reduce complexity and maintenance.
How do Smart Conversations work?
đź’ˇ Smart Conversations relies on two specialized AI agents - called Smart Nodes - that structure how conversations are handled:
Smart Classification acts as the central hub of the assistant. It identifies the user’s intent at each stage of the conversation.
Smart Capture functions as the branches. Each node gathers the specific information (parameters) needed to fulfill a given intent.
How This Works in Your Conversation Flow 🚀
In a typical flow, one Smart Classification node is connected to multiple Smart Capture nodes, each associated with a different intent.
These nodes work together in a continuous conversation loop: classification leads to capture, which may return to classification if a new intent is detected.
This architecture ensures that every user input is evaluated and routed appropriately, enabling fluid, goal-driven interactions without the need for traditional handoffs or scripted transitions.
The loop continues until the assistant either reaches the end of the interaction or detects an escalation scenario, such as repeated failure to collect input or a user request to speak to an agent.
Smart Classification node
The Smart Classification node detects the intent or topic the end user wants to discuss. It performs intent recognition using the Smart Intents you define in the Properties panel.
Key Behaviors:
Intent Recognition: Classifies user input against Smart Intents to identify the most relevant intent or topic.
Global Intent Support: Recognizes and handles global intents (e.g., “Cancel,” “Talk to agent”) when configured.
LLM-Powered Matching: Leverages prompt-based classification via large language models (LLMs), requiring minimal training data.
Outcome Handling: Returns one of three outcomes—Success, Missed, or Failed—based on the classification result.
Intent Storage & Smart Fallbacks: Stores the detected intent for downstream use. If no match is found, it automatically generates a smart response to inform the user of the available intent options.
Isolated Testing: Includes a Save and Test feature for standalone testing of classification behavior during configuration.
Smart Classification only supports session parameters of the “string” type.
Smart Capture node
Collects the parameter details needed to fulfill the end user's intent.
Key Behaviors:
Intent Tracking: Monitors the current intent and detects context switches (e.g., when the user shifts to a different intent mid-conversation).
Multi-Parameter Capture: Collects multiple parameters in a single flow when needed to complete an intent.
Retry Handling: Supports parameter retries to ensure values are captured in the correct format or structure.
Reconfirmation Support: Reconfirms captured parameter values with the user—especially useful in voice channels where ASR errors are common.
Smart Responses: Automatically generates contextual responses when user input is incomplete or invalid, keeping conversations smooth and on track.
Proactive Escalation Detection: Identifies potential escalation scenarios mid-conversation, including the specific parameter causing the escalation.
How to get started
This guide provides step-by-step instructions for building, training, and optimizing Virtual Agents using the new Smart Conversations AI engine.
Choose the Right AI Engine
Upon creating an agent, you will see an AI Engine dropdown with three options:
Smart Conversations 🚀
Hybrid NLU
Traditional NLU
Once you select the Smart Conversations AI engine, you can provide an optional Agent Description (max limit of 2000 characters) to highlight the VA’s objectives, which will act as context and help improve VA performance. Don’t worry, you can edit the description of your agent later on as well.
Smart Conversations is not backward compatible with Hybrid NLU or Traditional NLU. This means if a VA is created with the Smart Conversations AI engine, then you cannot change it to Hybrid NLU/ Traditional NLU (by editing or duplicating the VA) and vice versa.
Add Intents and Parameters
Smart Conversations replaces a Hybrid NLU/ Traditional NLU setup with a simplified approach. Smart Intents and Smart Parameters are defined in the Properties panel of your AI Studio agent. Once created, they can be reused across the agent’s flow.
đź’ˇ Smart Intents
Smart Intents define the goals your assistant can recognize. Each intent includes:
A name (short and descriptive)
A natural-language description of the intent (similar to an LLM prompt)
Pro Tip 🔥
As a best practice, always add two intents (“End conversation” and “Talk to human agent”) in all VAs to improve the performance of the Smart Conversations AI engine.
đź’ˇ Smart Parameters
Smart Parameters describe the information your assistant needs to collect.
In the “Smart Parameters” modal, there will be three types of parameters: “Session”, “System”, and “Users”.
“Session parameters” (similar to “Custom parameters” in Hybrid NLU)
Need to be created by Studio users manually within a VA. These parameters are great for data used throughout a session.
Would retain their values for one session only (i.e after the session ends, the parameter values are forgotten).
Parameter name,
Parameter value
Booking_ID
“System parameters” (same as Hybrid NLU)
These parameters are automatically created for all VAs.
Would retain their values for one session only (i.e after the session ends, the parameter values are forgotten).
These parameters cannot be edited/ deleted by Studio users
Agent_ID, Caller_Number
“User parameters” (same as Hybrid NLU)
These parameters are great for storing data that doesn't change very often.
Example scenario: If an end user speaks with your VA on Monday and Wednesday, any end user parameters set on Monday will still be accessible on Wednesday.
These parameters follow a user across sessions with the same VA or different VA under a given Studio account.
Parameter name,
Parameter value
User_ID, User_Language_Preference
How to set up your smart parameters
✔️ Session Parameters
You define:
A name for the parameter
Type:
String,NumberorDateTimeDescription: explains the parameter role within the VA
Format: list of rules & criteria that the end user’s input must meet for the parameter to be considered valid
Value: User can specify a single or multiple pre-defined value(s) if required
Multivalue: Optional field which allows a parameter to store more than one value
Heads Up! ⚠️
Smart Parameters do not rely on the concept of “entities” which is critical in Hybrid NLU/ Traditional NLU AI engine
For the “Description” and “Format” fields, it is recommended to follow the best practices of LLM prompting.
✔️ System Parameters
System parameters exist by default for all VAs, but their values are automatically populated and retained for one session only. Below is the list of System parameters in the Smart Conversations AI engine.
✔️ User Parameters
User parameters follow an end user across sessions and agents running in a given AI Studio account. They are used for storing end-user data that does not change very often.
Build your flow using Smart Classification and Smart Capture nodes
When designing with Smart Conversations, all conversation logic is driven through Smart Classification and Smart Capture nodes.
Start the flow with a Smart Classification node to detect user intent. Use Smart Capture nodes to collect any required information linked to that intent.
Smart Capture supports multi-turn collection, re-confirmation, and can trigger escalation if input is not usable.
Handle all exit paths from these nodes—especially Intent Change, Escalation, and Failed.
Logs and Testing
Every conversation is logged and visible under Reports > Report type = Call Log > Choose Session. Logs show:
Input and detected intent
Captured parameters
Escalation reasons if capture fails
Use logs to improve coverage and adjust flows.
Pro Tip 🔥
Check the Smart Parameters tab to trace exactly how input was processed and where it failed or succeeded.
⚠️ Limitations
Available for voice agents only
A Smart Response is automatically generated by LLMs and cannot be manually defined by Studio users
Retry behavior and reconfirmation behavior are pre-defined and cannot be altered manually
Migration and Compatibility
When using Smart Conversations, AI engine selection becomes fixed. 🚨
Smart Conversations uses a different set of components and architecture than Hybrid NLU or Traditional NLU, which prevents cross-engine migration.
Once an agent is created with Smart Conversations, it cannot be switched to another NLU engine.
Agents built with the Hybrid NLU or Traditional NLU engines cannot be imported or duplicated into Smart Conversations.
Similarly, agents built with Smart Conversations cannot be converted to Hybrid NLU or Traditional NLU.
Last updated
Was this helpful?