What's an intent?

An intent is a consumer request for action or information from your brand.

For some practice with intents, complete the Getting Started with Bot Building tutorial series.

Import a batch of intents

You can create a domain using an import file that contains a set of intents.

Add an intent

  1. Access Intent Manager.
  2. Locate the domain list in the Intent Manager dashboard.
  3. Select the name of the relevant domain, and then select Build from the menu that appears.
  4. On the Intents page, click Add Intent in the upper-right corner.
  5. Specify the following:

    • Intent Name: Enter the intent name. To name an intent, use a short phrase that describes the intent. Typically, an intent name has both a verb and a noun (e.g., "report login problem"). A domain can have dozens of intents, so using a standard naming convention is important for being able to easily sort and find intents.
    • Intent Display Name: Enter the display name.
    • Description: Enter a short phrase describing the intent. While this field is optional, it's often useful. Many intent names can be technical. A description adds clarity and is particularly helpful to a person not familiar with the domain.
    • Intent Type: Select either "Intent" or "Meta Intent."
    • Training: Enter as many training phrases as possible, keeping in mind these best practices. The NLU engine uses the training phrases in order to match a user's utterance with an intent. The more training phrases you include, the more likely it is that the NLU engine will accurately match the user's intent. Generally speaking, the phrases should be complete sentences (not long paragraphs, and not keywords). As an example, assume you have a "Check bill" intent. You might add the following training phrases, among others:
      • I want to check the status of my bill
      • Tell me what my bill is
      • I need to look into what's going with my bill
  6. Click Save.
  7. Train the domain so that the addition is reflected in a new model version.

Generate training phrases via Generative AI

Adding training phrases to an intent can sometimes be a challenge, so Intent Manager includes an AI-powered tool to help with this. Within an intent, describe the subject of the training phrases as the input, and the tool automatically generates additional phrases using Generative AI.

Use cases

  • You’ve mined your data for consumer messages that can be used as training phrases for an intent, but you only found a few that are suitable. So, you’re looking for more ways that a consumer could express the same intent.
  • You’re building out a LivePerson or third-party NLU domain that you’ve created from scratch.
  • You’ve converted a LivePerson (Legacy) NLU domain to the LivePerson engine, and now you need to increase the number of training phrases to meet the minimum requirements.

Important notes

This LLM-powered feature is only available 1) for English-language domains, and 2) if you log into Intent Manager through Conversational Cloud.

Before you begin

  1. You must activate this Generative AI feature.
  2. If the intent already has some existing training phrases, we recommend you review them before you generate phrases with our AI-powered tool. As per our best practices, ensure they don’t include sensitive info like consumer names or other identifiers. This is because existing training phrases are sent to the LLM (which exists outside of LivePerson’s servers) as examples.

Watch the 1-minute video


Generate training phrases with AI

  1. Access Intent Manager.
  2. Locate the domain list in the Intent Manager dashboard.
  3. Select the name of the relevant domain, and then select Build from the menu that appears.
  4. On the Intents page, select the intent.
  5. Above the existing training phrases, click Generate.

    The Generate button in the UI

  6. Enter the following info:

    The Generate phrases with AI dialog in which you enter info

    Topic: Required. The topic describes the subject of the training phrases (max 375 words). It helps the model understand what type of response is expected and guides its decision-making process. Remember to write the topic from the perspective of the consumer, not the agent.

    If the intent already has a description, we will use the intent description and trim it to generate a topic for you for the first time. You can also edit the topic. We’ll save the topic for reuse.

    Here are a few examples of good topics:

    • For a “cancel flight” intent, a good topic is, “Any request or intention to cancel a previously scheduled flight reservation. This can be due to a change in travel plans or unforeseen circumstances that make the trip impossible or impractical.”
    • For a “travel notification” intent, a good topic is, “Any request related to notifying the financial institution of a consumer's planned dates of travel, and might include the planned location/s of travel.”

    Guidance: Optional. Guidance is additional instructions to use when generating phrases (max 375 words). Add details that aren’t present in the topic, e.g., your brand’s name, product or service descriptions, phrases to include or exclude (examples: “don’t start with a greeting,” “reason for booking is a family emergency”), suggestions on tone (examples: “be polite,” “express frustration”), etc. Remember to write the guidance from the perspective of the consumer, not the agent.

    Here are a few examples of good guidance:

    • For a “cancel flight” intent, guidance might be, “Canceling a flight due to family emergency or work obligation. Customer is apologetic.
    • For a “purchase power tool” intent, guidance might be, “Company is Acme Tools. Don't start with a greeting. You want to purchase a power tool such as a miter saw, leaf blower, pressure washer, or a power drill, but don’t know what it is called. Describe in detail what it will be used for. Don't be generic.”
    • For a “return product” intent, guidance might be, “Reasons for returns could be manufacturer defect, item is not what was ordered, item is different than expected, too many were ordered, etc.”
  7. Click Generate to generate up to 30 phrases using AI.
  8. Review the generated phrases, and select the ones you want to add to the intent. Click Add to intent to add the phrases to the intent.

    The Generate phrases with AI dialog in which you select the phrases to add to the intent

  9. Back on the Intent Details page, you might want to edit the phrases to best fit the intent. Then click Save.

    The Intent Details page where you save the changes that you made to the intent

  10. Retrain the model.

Troubleshooting flows

If the AI is unable to generate enough phrases, the issues might be due to several reasons:

  • An unclear topic
  • A topic that conflicts with the guidance or an example phrase
  • And so on

To remedy, start small and gradually introduce more info:

  1. Remove the guidance. Tune the topic and retry.
  2. Add the guidance and retry.

    The Generate phrases with AI dialog showing an error where the tool was unable to generate phrases

    If the AI is unable to generate enough phrases and you don’t have any existing training phrases for the intent, the system will prompt you to input 3 training phrases as examples. These help the UI to further understand what kind of responses are expected and therefore improve the results.

    The Generate phrases with AI dialog showing three fields at the bottom for entering example phrases

    The training phrase examples are automatically added back to the intent, together with the AI-generated training phrases that you manually select, once you click Add to intent.

FAQs

Which LLM model are you using?

LivePerson is using one of the best and latest versions of OpenAI’s models. Advances in this area are happening quickly, so we’re continually evaluating the model we’re using to ensure it’s the best choice possible.

Currently, it’s not possible for you to select a particular model to use.

Will the LLM model generate inappropriate content? Is it safe?

With a well-defined topic, the model should generate appropriate content most of the time. Additionally, the guidance can help steer the content toward or away from certain types of phrasings. However, it is possible that the LLM will return something overtly or subtly inappropriate.

You always have an opportunity to review the generated phrases before adding them to your intent. This is an important step to ensure you only add to the model what you want.

Can the model generate more than 30 phrases at a time?

Our testing has found that the LLM consistently returns the best results with around 30 results at a time, so that number cannot be changed. However, you can repeat the process as many times as you like to get more training phrases.

Is my conversation data sent to OpenAI or Microsoft Azure?

No. We send example phrases taken from the intent’s existing training data, but we never send conversation data. As a best practice, avoid putting sensitive info like consumer names or other identifiers in your training data.

Find similar training phrases in your consumer messages

Adding training phrases to an intent can sometimes be a challenge, so Intent Manager includes a tool to help with this. Within an intent, provide just a single training phrase as the input, and the tool automatically finds additional phrases that are similar in meaning. The suggestions are based on actual utterances by your users.

Use cases

Important notes

  • This feature is only available 1) for English-language domains, and 2) if you log into Intent Manager through Conversational Cloud.
  • Expect the tool’s results to improve over time as more data is captured.
  • If you're a new customer, expect no results until suggestions can be offered based on utterances by your users. And here again, expect the tool's results to improve over time.

Find similar training phrases

  1. Access Intent Manager.
  2. Locate the domain list in the Intent Manager dashboard.
  3. Select the name of the relevant domain, and then select Build from the menu that appears.
  4. On the Intents page, select the intent.
  5. Beside an existing training phrase (you’ll need to add at least one to the intent), click Generate training phrases icon.

    Example training phrase, with the Generate training phrases icon beside it

  6. Review the phrases that were found, and select the ones you want to add to the domain. You might also want to refine the spelling or punctuation. The phrases are based on actual consumer utterances, so they might contain misspellings or poor grammar. (See the FAQ regarding typos and misspellings.)

    Find Similar Phrases window

  7. Click Add.
  8. Back on the Intents page, click Save to save the change.
  9. Retrain the domain.

Add or update training phrases in bulk

Use the Bulk Add or Bulk Edit features, respectively, to add or edit a set of training phrases in bulk.

Bulk Edit link on the Intent Details page

When you click the link, all of the training phrases are placed into a single field. This means you can make changes to all or some of them in bulk. For example, you can:

  • Insert a number of additional phrases at the top or bottom. This is handy if you have the phrases in a separate file. You can quickly copy them and paste them in.
  • Select a number of phrases and delete them all at once. This is faster than doing so one by one.

Example of all training phrases in a single field

To save your changes, click the Update Intent button. Once you save the changes, they can’t be reverted. To cancel, navigate away from the page without saving the changes.

Take care when using this feature to avoid making unintended changes:

  • To replace what exists, paste over it.
  • To add to what exists, paste the additions into (not over) the list.
  • Ensure that each phrase is on a different line.

Delete an intent

Deleting an intent is a non-recoverable action, so be certain about doing so before taking this action.

Ensure the intent isn't being used by any bots or knowledge bases before you delete it.

To delete an intent

  1. Access Intent Manager.
  2. Locate the domain list in the Intent Manager dashboard.
  3. Select the name of the relevant domain, and then select Build from the menu that appears.
  4. On the Intents page, select the intent.
  5. Click Three-dot icon (3-dot icon), and select Delete.
  6. In the confirmation dialog, click Yes.
  7. Train the domain so that the deletion is reflected in a new model version.