Customer Assist

You can create intents that use common phrases your customers use when submitting an inquiry. These phrases are words, or groups of words, that a conversational assistant listens to during a chat conversation. Using natural language processing, the assistant uses these phrases to identify your customer's needs and redirect them to a specific dialog so that they can get the information that they need. This article provides tips to take into consideration when creating phrases for your intent detection. 

In this article

• Have distinct examples across intents
• Use a wide vocabulary
• Make sure the assistant has the context it needs

Have distinct examples across intents

A common mistake when creating intents is to organize them into categories that are too granular. It's better to keep your intents more general to ensure variations are distinct. 

For example, let's say you've noticed that many customers contact you to inquire about topics such as, help with their password or email address. Instead of creating a different intent for each category and naming them Change password or Change address, you can instead create a more general My account category and save a variation of phrases for all of these inquiries.

Example variations of phrases that you can save under a general My account category are:

• I need to change my password
• I want to change my e-mail
• What is my password?
• Can you  reset the password?
• How do I recover my password?
• How do I update my account?
• Upgrade my account
• How can I edit my personal information?
• I cannot login into my account
• I forgot my password

These phrases provide a variety of ways to ask for information that can be grouped under account inquiries. When coming up with phrases for your intents, you should make sure that all of your intents are clearly defined and that a phrase can't be assigned to more than one intent category.

Use a wide vocabulary

A common mistake users make when creating intents is prioritizing the quantity of phrase examples, over the quality of the phrases themselves. It's much more important to make sure that the variations of your phrases cover a wide range of vocabulary and wording to ensure that the assistant can detect the various ways a customer might ask for the same thing. 

For example, let's say you want to create an intent that detects when a customer is inquiring about shipping costs. 

Example variations of phrases that you can save to capture the different ways a customer might ask for shipping cost information are:

•  How much does shipping usually cost?
•  How do you negotiate shipping costs?
•  What is the cost of shipping and handling?
•  How much does it cost to ship a package via USPS?
•  USPS shipping cost per pound?
•  What is the cheapest way to get the package delivered?
•  How do I choose a shipping strategy for an online store?
•  How much am I charged for shipping and handling?
•  How do you calculate shipping costs?
•  Can I reduce my shipping costs?

Make sure the assistant has the context it needs

The last tip to remember when creating an intent is that the assistant only understands what is said in the message and can't successfully understand a customer's intent if it needs to rely on external information to provide all of the context it needs. 

For example, let's say you are considering creating intents for Order delayed or Order on time because you've noticed customers regularly contact your organization for either of these reasons. If an agent is reviewing this inquiry, they are going to look at a variety of factors such as, whether the order is still being processed or tracking details to determine if the order is on time or delayed. Once they get this information, only then will they tag the conversation as Order delayed. 

The assistant, on the other hand, does not know the status from the original inquiry, even if the inquiry is "Is my order late?" In this scenario, creating an intent that directs the customers to a dialog apologizing for the late delivery could possibly be incorrect since more context is needed before making that conclusion. You should avoid creating intents that rely on additional information to correctly identify the customer's needs.