Configure Genesys Agent Copilot natural language understanding
AI-generated summary
Genesys Cloud provides Natural Language Understanding (NLU) capabilities within Agent Copilot that enable creation of intents—descriptions of customer goals, activities, or tasks—paired with utterances representing the words or phrases customers use to express those goals. The NLU system functions as a classification model that assigns probability distributions across all intents, selecting the intent with the highest probability provided it meets a minimum 40% confidence threshold; inputs falling below this threshold trigger fallback or None intent responses.
Key configuration best practices include maintaining balanced training data across intents to prevent model bias, understanding word vector similarity to avoid misclassification of contextually similar terms, and reviewing full probability distributions during troubleshooting. The system supports AI-assisted generation of multiple intents and utterances based on general descriptions, with effectiveness improving as users provide greater contextual detail. Recommended implementation approaches include defining clear, non-overlapping intents with distinct purposes, starting with a core set of high-priority intents before expansion, and using consistent naming patterns. The default confidence threshold should remain unchanged unless specific justification exists, as raising it excessively may increase fallback responses while lowering it may increase incorrect matches. Users must click Save Copilot before navigating to other tabs to ensure work is preserved and the NLU model publishes successfully.
Training data should include 10–20 diverse examples per intent with varied phrasing, synonyms, different syntactic structures, and common typos, with a minimum of five utterances per intent and ideally 20 for optimal results. The Intent Miner tool enables organizations to build intents and utterances tailored to customer phrasing patterns by analyzing historical chat and voice transcripts across specified date ranges, supporting up to five concurrent mining sessions per organization with a maximum of 100 saved sessions and requiring a minimum of 50 conversations per session. Intent health monitoring provides performance guidance through status indicators—Very good, Good, Okay, Needs review, No data, Changes made, and New—based on NLU best practices and training data quality, with Genesys recommending review of intents marked Okay or Needs review. The Test NLU tool evaluates sample messages against existing intents and dialog acts, returning probability scores for classification accuracy, while the intent health column can be toggled on or off through Hide Health and Show Health options, with the Agent Copilot remaining health-aware regardless of column visibility.
With the help of Genesys AI, you can create intents to describe a goal, activity, or task that a customer wants to accomplish. Each intent must have utterances. Utterances are words or phrases that a customer uses to describe what they want to do.
Maintain balanced training data across intents. Disproportionate numbers of training examples per intent can introduce bias into your model. If one intent has a considerably larger amount of training examples than others, the model may favor that intent when uncertain. Ensure a relatively even distribution of training phrases across all intents. If you intentionally want to prioritize an intent, do so consciously and document the reasoning.
Understand word vector similarity. NLU systems use vector representations (embeddings) to understand word meanings based on context. Words with similar meanings or contexts, like cash back and refund, may be mapped closely in the model’s vector space. This can lead to misclassification if you do not clearly distinguish the intents they belong by context. Use diverse, representative training examples for each intent, emphasizing unique contextual clues that differentiate one intent from another.
NLU is a classification problem. NLU models classify input text by assigning probabilities to each intent. For N intents, the model returns a probability distribution over all N intents. The intent with the highest probability is selected, provided it meets the confidence threshold. Review full probability distributions when troubleshooting misclassifications, not just the top-scoring intent.
Understand confidence thresholds and intent fallbacks. The system typically requires a minimum confidence score to classify an input into a specific intent. The NLU model requires at least 40% (0.4) confidence to assign an intent. If the top intent falls below this threshold, the system returns a fallback or None intent. Leave the threshold at the default value unless you have a specific reason to adjust it. Raising it too high may lead to excessive fallback responses; lowering it may increase incorrect intent matches.
Warning: To ensure that you save your work and your NLU model publishes, you must click Save Copilot before you move to another tab.
Generate multiple intents and utterances with AI
You can generate a list of intents and utterances with the help of AI, based on existing intent and utterance descriptions. By adding a general intent description, AI generates a set of more concise intents, each with their corresponding utterances.
Note: This option is only available if you have not yet created any intents.
When you generate intents and utterances, the more context that you provide, the better. For example: The customer asks about a product’s availability in the online store.
Define clear, non-overlapping intents. To avoid ambiguity, ensure that each intent serves a distinct purpose. Overlapping intent definitions lead to confusion during training and lower classification accuracy. Establish clear boundaries between intents. If two intents are too similar, consider merging them or refining their definitions.
Start with a core set of intents. Begin with a small, high priority set of intents before you expand. A focused set helps you iterate faster and build a solid foundation. Develop a proof-of-concept model with essential intents. Expand only after the core experience is works reliably.
Use consistent naming patterns. Standardize the way that you name intents across your project. Consistency improves maintainability and helps teams understand intent purposes immediately.
On the NLU tab, click Generate intents. The Generate Intents dialog box opens.
In the Intent Description box, type a description of the intents that you want to create.
Optionally, under Industry, type the name of the industry for the bot.
Click Generate Intents & Utterances. Genesys Cloud generates your intents and utterances.
Note: The process to generate intents and utterances can take several minutes.
To see the generated utterances for an intent, click the intent.
To direct AI to generate new utterances, edit the intent description.
Generate one intent and the corresponding utterances with AI
If you already have one or more intents, you can generate an additional intent and the corresponding utterances.
On the NLU tab, click Intent.
In the Intent Name box, type a name.
In the Intent Description box, type a description for the intent, and then click Save.
AI generates the intent and the corresponding utterances, and Genesys Cloud navigates you to the list of generated utterances.
To delete an utterance, next to the utterance, click Delete .
To test your intents, in the upper right corner of the Intents list, click Test NLU.
Note: Enter any intent that you want to test. Test NLU allows you to test sample messages against existing intents and dialog acts. For more information, see Test the NLU classification of intents and slots.
Click Save Copilot.
Notes:
Genesys recommends that you create one intent per activity.
Add a minimum of five utterances to each intent. For best results, add 20 utterances to each intent.
Generate an AI-powered description for an intent
If an intent did not have a description before, then you can generate a description for the respective intent with the help of AI.
On the NLU tab, click Intents.
Select the intent that you want to generate a description for.
Click Generate description.
Note: Description generation can take a few minutes.
To the upper right of the NLU tab click Test NLU.
The NLU test dialog opens.
In the Type message here and press return key box, type your sample phrase or message and press Enter.
Note: To test NLU for supported languages, make sure that you switch from the default language to the supported language, and then repeat these steps.
The NLU test dialog returns probability scores for the phrase or message.
Avoid one-word inputs. One-word inputs are typically too ambiguous and lack the necessary context for accurate intent classification. NLU models rely on patterns and context. Without surrounding words, the model must guess based on limited information and may default to the most common or closely related intent. Use natural, full-sentence queries that reflect real user behavior.
Prepare high-quality training data using the following tips:
Develop a viable training set: 10–20 diverse examples per intent to start.
Vary phrasing: Include synonyms, different syntactic structures, and common typos.
Avoid overlap: To reduce confusion, make sure utterances clearly distinguish between intents.
Test coverage: Sample utterances from all intents manually to check NLU predictions.
Use the intent miner
To import a list of your intents with all the created utterances, use intent miner.
Create variations of the utterances and a wide range of semantics. The more varied the utterances, the more effective the intent.
Use intent miner to build intents and utterances tailored to your customers way of phrasing for better results in intent detection. Intent miner searches a historical set of conversation chat and voice transcripts between agents and customers for a specified date range. Next, it extracts the top set of intents and the analytics that surround those intents. Finally, it returns a list of intent utterances that the system associates with that intent.
The intents on the sample image have at least five utterances. Their slider indicates intent effectiveness.
Notes:
You can only run up to five mining sessions at a time.
You can save up to 100 intent miner sessions in a single org.
You must have a minimum of 50 conversations available for the mining session. If you cannot meet the minimum, consider expanding the date range to include more conversations.
Check intent health
Intent health helps guide you as you build out your intents and gives you the tools to improve the performance of your intents and training utterances easily and accurately. The Intent Health column displays one of the following heath statuses based on NLU best practices and training data.
Very good: The intent contains a good number of utterances, and no potentially confused utterances.
Good: The intent contains slightly fewer than optimal utterances or has a few potentially confused utterances.
Okay: The intent contains a low number of utterances or multiple potentially confused utterances.
Needs review: The intent contains fewer than five utterances or several potentially confused utterances.
No data: The intent contains no utterances.
Changes made: There are recent changes to this intent.
New: This is a new intent.
Intent status indicates which intents may need improvement. Genesys recommends that you review all intents, with a focus on intents with an Okay or Needs review status. For more information, see Work with Agent Copilot intent health. You do not need to address every issue for the Agent Copilot to work properly.
You can also choose to hide the intent health column; however, the Agent Copilot remains health-aware and computes any changes that you make toward the health algorithm.
To hide intent health, click Hide Health. The setting changes to Show Health.
To show intent health, click Show Health. The setting changes to Hide Health.
[NEXT] Was this article helpful?
Get user feedback about articles.
Configure Genesys Agent Copilot natural language understanding - Genesys Cloud Resource Center