Training Phrase
A Training Phrase (also called a training utterance) in Salesforce Einstein Bots is an example sentence or expression that an administrator provides to teach the bot's Natural Language Understanding (NLU) model how to recognize a specific intent.
Definition
A Training Phrase (also called a training utterance) in Salesforce Einstein Bots is an example sentence or expression that an administrator provides to teach the bot's Natural Language Understanding (NLU) model how to recognize a specific intent. For each intent, multiple training phrases are added to cover the different ways a customer might express the same request. For example, the 'Check Order Status' intent might include training phrases like 'Where is my order?', 'Track my shipment', 'I want to check my delivery status', and 'Order update.' The more diverse and representative the training phrases, the better the bot can accurately classify incoming messages.
In plain English
“A Training Phrase in Salesforce Einstein Bots is an example sentence you provide to teach the bot's NLP model what users might say for a specific intent. Like giving the bot examples of how people ask about order status so it recognizes similar questions.”
Worked example
When the Einstein Bot team at Greenmoor Software trains the bot to recognize the CheckOrderStatus intent, they add 35 Training Phrases: "where is my order," "track my shipment," "check delivery status," "when will my order arrive," "order tracking," and 30 others. The NLP model learns to classify all 35 (and semantically similar phrases) as the CheckOrderStatus intent. When a real user types "my package hasn't arrived yet" - a phrase the model has never seen but is semantically similar - it still classifies correctly. The richer the Training Phrase set, the more robust the bot is to natural-language variation.
Why Training Phrase matters
A Training Phrase (also called a training utterance) in Salesforce Einstein Bots is an example sentence or expression that an administrator provides to teach the bot's NLP model what users might say when expressing a specific intent. Multiple training phrases per intent help the model recognize variations.
The more diverse and representative training phrases are, the better the bot recognizes real user input. Mature bot programs invest in training phrase quality: providing many variations, including common misspellings and informal language, and updating phrases based on actual user interactions that the bot fails to recognize.
How organizations use Training Phrase
Provides diverse training phrases for each bot intent based on actual customer language.
Updates training phrases regularly based on unrecognized user inputs.
Invests in training phrase quality as foundational to bot accuracy.
Test your knowledge
Q1. What is a Training Phrase?
Q2. How many per intent?
Q3. How should you improve them?
Discussion
Loading discussion…