Train your virtual service agent to recognize an intent
Each intent represents a specific problem, question, or request that your virtual service agent can help to resolve for your customers. Read more about intents.
When a customer’s message is matched to an intent, the virtual service agent can start a conversation flow to get the customer the help they need — but first, the virtual service agent needs to be trained to recognize intents in customer messages.
To train the virtual service agent to recognize an intents, you need to provide it with training phrases. The virtual service agent analyzes these training phrases and then uses machine learning to recognize that intent in actual customer messages — even if their messages don’t exactly match one of your training phrases.
Add, edit, or delete training phrases from an intent
Next to your project’s name in the sidebar, select More actions (•••), then Project settings.
Select Channels & self service, then Virtual service agent.
Select Manage in Studio.
Select Intents, and then select the intent you want to edit.
Select Training.
Add, edit, or delete training phrases:
To add a new training phrase, type your phrase into the field under Training phrases, then press Enter.
To edit an existing training phrase, select the one you’d like to edit, make your changes, then hit Enter.
To delete a training phrase, select Delete ().
Select Save.
Tips for effective virtual service agent training
Use words and phrases most commonly used by your actual customers. Customers may ask for the same thing in very different ways — capture as many of these different ways of asking as you can! Read through past work items, and copy and paste the initial phrases your customers use when asking for help with the intent's main purpose.
Say the same thing in multiple ways, using a combination of statements and questions. For example, instead of just adding the training phrase “I can’t connect to the VPN”, try also adding “why isn’t my laptop connecting to the VPN?” and “how can I connect to the VPN?”. While the keywords in these statements are the same (connect and VPN), each phrase is slightly different, which helps the virtual service agent recognize what your customers are asking.
Make sure your training phrases don’t overlap with training phrases used by other intents. For more accurate intent recognition, avoid creating intents with training phrases that are too similar to each other.
The ideal number of training phrases for an intent depends on the intent and your virtual service agent setup. Depending on how many intents you have, and how much they might overlap with each other, the ideal number of training phrases is highly variable. For example, if you have an intent that’s very specific (like submitting expenses for reimbursement), you might only have 5–20 training phrases, while an intent that’s related to getting a software license might need more training phrases to cover all of the software programs your company uses. However, in this last example, you could improve your virtual service agent’s intent recognition by creating a separate intent for each software program.
Review and edit your training phrases as needed. As you watch the virtual service agent’s performance, you may want to come back to existing intents and edit your training phrases to make your virtual service agent’s intent recognition more effective over time.
Was this helpful?