Effective intent management is critical for ensuring accurate intent recognition and maintaining high system performance over time. This guide provides best practices for setting up, managing, and optimizing intents for AI intent routing to maximize success.
Intent groups organize related intents to improve accuracy. Start by identifying different user types or topics. For example, an airline might have separate groups for bookings, account info, and internal employee needs.
Group similar intents under each category and give the group a clear name. Each group should focus on a single topic or user type. By default, all intents go into one group unless specified. Update groups as your business needs change.
Learn more about how to create and manage an intent group.
An intent represents the customer's goal behind a question or comment. High-quality intents help you understand common questions without reading every message.
To create a high-quality intent, focus on the core goal behind different ways a customer might ask something. For example, “When does your London store close?”, “Is the London store open Thursday night?”, and “When does the London store open?” all point to the same intent: understanding London store hours.
Group similar questions under one intent with a clear name. Use intent groups to organize related intents and make the system more accurate.
Learn more about how to create and manage intents.
When you make changes to an intent group, you need to train it to create a new draft. Once you're happy with the results, you can publish the draft. Changes that require training include adding or deleting intents, adding training phrases, coaching an intent, or adding rules.
If the new version doesn't perform well, you can discard the draft and republish the previous version. This will remove all changes made in the draft, including any new training phrases.
Learn more about how to train and publish intent groups.
To improve self-service, leverage intent discovery to identify patterns in user queries that indicate the need for new intents or subflows. To access intent discovery, follow the steps below:
To improve accuracy, enable intent auto-coach to automate the generation of training phrases from real user queries. This ensures that your bot remains up-to-date with evolving user language and adapts to new patterns in user behavior. To enable intent auto-coach, follow the steps below:
.
Intent accuracy is assessed by evaluating the consistency of training phrases, comparing 80% of the available phrases against the remaining 20%. By raising accuracy to over 60%, we ensure that the machine learning model remains consistent and is exposed to a broader variety of training phrase examples. To view the intent model analysis, follow the steps below:
.For intents with an accuracy below 60%, review and refine the training phrases. Add more relevant examples and remove any that do not align with the intent's goal.
The goal is to continuously enhance intent detection by identifying and correcting incorrect intent matches. This can be achieved by moving inappropriate matches to the fallback intent. Additionally, improving intent recognition involves reviewing suggested queries and incorporating the most valuable ones as training phrases. To review the intent coach, follow the steps below:
