AI in playbooks allows Drift users to combine the intelligence of an AI model with the design capabilities of the playbook builder. This gives buyers a better experience while generating more efficient pipeline.

Billing Plan: 
Conversational AI in playbooks is available to FY22 Enterprise plan customers.
 

In this Article, we will cover:

• What is conversational AI in playbooks?
• What is the base model?
• How to create custom AI topics
• Use AI to generate topic examples
• Use AI to generate Bot Responses
• How to use conditional responses in custom AI topics
• How to train your custom AI topics with on-demand training
• How to configure a response to unrecognized responses
• Archiving AI Topics
• How to turn on AI in playbooks
• How to use AI topics to branch in the playbook builder
• How to review conversations with AI

What is AI in playbooks?

AI in playbooks allows Drift users to combine the intelligence of an AI model with the design capabilities of the playbook builder. This gives buyers a better experience while generating more efficient pipeline. Our AI base model is trained on two billion conversations to understand the most common things buyers say in chat, out-of-the-box. With customizable responses, common questions can be answered in your brand voice in any playbook, 24/7, saving the humans on your team for conversations with qualified buyers and customers. What’s more, you can use AI to listen across all your conversations to generate insights from the voice of the customer. If you want to go further, you can train custom AI topics to answer questions specific to your business.

What is the base model?

Drift’s out-of-the-box AI base model is pre-trained consisting of the most common topics, which are based on two billion conversations.  

• You can configure responses to these base model topics in your brand language. Note: Your bot will not respond to the topic until you add a response.
• When you turn AI on in a playbook, the bot will respond to an AI topic anytime it’s brought up in conversation. Your bot then continues down the path for that topic (see “How to use AI topics to branch in the playbook builder” section) or continues down the ‘All other responses’ path.
• Base topics are displayed with a topic description and three examples of the training data it is trained on.

Base model topics are marked with a lightning bolt icon under the ‘Examples’ column in ‘Library’ > ‘AI Topics.’ 

To configure responses to the base model topics, open a topic and complete the ‘How should your bot respond to this topic?’ section. Use the Group dropdown to add your topic to a group which will make it easier to bulk at topics to your Playbooks.

You can further customize base model topics by adding additional examples that reflect how your site visitors talk about this topic. 

How to create custom AI topics

Drift’s pre-trained topics may not cover all the requests from your buyers. There are two ways to enrich your AI topic library.

Option A: Users can create a new topic of their choice

Select ‘Create Topic’ under ‘Library’ > ‘AI Topics'. Complete required sections, including examples, and select ‘Save Topic'. 

Notes: 

• Your bot will not respond to the topic until you add a response.
• Examples, or utterances, are different ways a visitor might say the phrase. The more examples you add, the stronger the topic will be.  
• Custom topics behave the same as base model topics in the playbook builder. 

When creating topics, it is best to have at least ten examples. Our team recommends adding 10-50 examples to create a strong topic. We do not recommend repeating or single-word phrases as these will confuse the bot.

 

Use AI To Generate Topic Examples

You can now use the power of AI to help you generate examples for your topics!


Type out an example like you normally would and then click the "add and generate examples" button. Once the new examples have been generated, click the "add" button next to each example to add it to the topic.

Once you've added your new examples to your topic, you are able to have Drift's AI generate up to thirty examples or until your topic is "Strong". Once your topic is "Strong" you will no longer be able to generate AI examples for your topic.

Option B: User can utilize Drift AI conversation insight suggestions to create a new topic

Drift provides conversation insights into what your buyers are talking about in the “Suggestions” tab. These suggestions are pulled from conversations across the Drift app to get closer to the voice of your site visitor. Users can utilize this to create custom topics or improve existing ones.

The “Suggestions” tab shows groups of related examples that the bot was not confident in. You can select examples within the group to create a new topic or add them to an existing topic.

Additionally, examples that do not belong to a group can be added to existing custom AI topics, or used to create new custom AI topics.

New suggestions are generated every Sunday.

Notes: 

• Your bot will not respond to the topic until you add a response.
• Examples, or utterances, are different ways a visitor might say the phrase. The more examples you add, the stronger the topic will be.  
• Custom topics behave the same as base model topics in the playbook builder. 

Important Note: Changing the response to an AI topic will be reflected in your bot and playbook immediately. Updates to a topic’s examples will be reflected in the bot once the topic is trained (refer to section 'How to train your custom AI topics with On Demand Training' for details.)

Use AI To Generate Bot Responses

Note: AI Generated Bot Responses are available as a beta feature for customers on the Enterprise plan.

Instead of writing out your bot responses for each topic you create, you can now use the power of AI to generate bot responses. This feature can be found within each of your AI playbook topics. Click the playbook tab > library > AI topics > topics and click into the topic you wish to configure responses for.

To generate a reply, click the Generate response with AI button in the composer. This will generate a reply and will overwrite any existing reply that is currently in use. Once your response is generated, you can edit it however you like. When you're satisfied, click the update and close button in the upper right corner of the screen to save your topic.

Best Practices for Training Your Bot

When adding examples for custom AI topics, please use the following best practices:

• Do write examples as a single sentence.
• Do use examples from real conversations with site visitors.
• Don't repeat examples.
• Don't add the same example to multiple topics.
• Don't include deliberate misspellings.
• Don't worry about punctuation and capitalization.
• Avoid written examples that are long and complex.

How to use conditional responses in custom AI topics

You can use conditional responses for custom topics that require different responses based on keywords in the site visitor’s message. Conditional responses allow you to define a default response, keywords to listen for, and a specific response if the site visitor mentions those keywords.   

• For example, a ‘discount’ AI topic can require a different response for a healthcare discount vs. a student discount.  

When creating a custom topic, simply populate the default response and customized response for desired keywords as shown below.

How to train your custom AI topics with on-demand training

On Demand Training for AI topics provides the ability to retrain your model as needed so that your playbook can answer new trending buyer questions. 

AI Topics can be trained on demand as follows:

• When an AI topic is created or updated with new training examples, it will be displayed with a label of ‘unpublished.’  
• Once the user selects ‘Publish all changes,’ the new/updated AI topics will be displayed with ‘Changes in progress’ until training is complete within an hour.
• Once the changes are live, a ‘Published’ label will appear briefly.   
• Bulk changes (changes to more than one AI topic) will be published at the same time.
• Users can publish new changes while a job is currently running.
• Users can see whether a topic has been added/changed since they last published.  They see these as “Unpublished changes”.
• Users will know when their changes have been successfully published, or if there was an error publishing their changes.
• You do not need to publish changes to the topic name or response, those will be updated immediately.

How to configure a response to unrecognized responses

You can configure how your bot responds to novel utterances (otherwise known as unrecognized responses) by letting the site visitor know that their inquiry was not understood.  This is done in the AI Topics library, under the ‘not recognized’ topic.  Please note, this is a topic that cannot be left with a blank response nor can this topic be archived.



Novel response looping allows site visitors / buyers to repeat their inquiry (using different phrasing) to give the AI chatbot additional opportunities to classify their utterance and provide a response. The settings configuration in the AI Library allows the question to be repeated up to 4x when a novel utterance (unrecognized response) is encountered by the AI chatbot.
 

Handling Profanity

Profanity is handled via a locked base topic called ‘Profanity.’ Profanity is identified using keyword match across AI conversations. How the bot responds to profane language across all AI playbooks can be configured in the AI library. Once the profane topic response is generated in chat, the conversation will go down the 'all other AI topics' conversation branch.

Archiving AI Topics

Base topics and custom topics can be archived as needed. You can archive a topic by clicking the triple dots and selecting “Archive topic” from the menu.
 

If you are using this topic in any active playbooks, a warning will pop up, prompting you to remove the topic from the playbooks before archiving the topic.
 

Archiving a topic removes it from your model, meaning your bot won’t be trained to recognize and respond to the topic. You can restore an archived topic by changing your view to “Archived topics,” clicking on the triple dots on the topic you would like to restore, and selecting “Restore topic.”
 

Archiving is recommended for base topics and custom topics that are not relevant to site visitor conversations.  This ability to archive topics will enable stronger AI model performance by allowing for only appropriate topics from your AI library to be triggered.

How to turn on AI in playbooks

Topics from both the base and custom model will automatically cover “All other responses.” If the bot recognizes a topic from the site visitor’s message, it will reply with the response you added and continue down the “All other responses” path. You can also create a new path for an AI topic (see the last section).

To turn AI on in a playbook:

1. Users can switch the toggle on at the top of the ‘Bot Flow’ page in the playbook builder OR utilize our AI Playbook Template to get started.

AI Template

When creating a new playbook, you can leverage the AI playbook template to get started faster.

Important Note: AI responses will only apply to questions where a user can reply with  ‘open text.’ To turn on free text: Click into the node and enable visitors to respond ‘With buttons or text.’ You need to turn this setting on for every node you want free text and AI responses to be enabled.

How to use AI topics to branch in the playbook builder

Branching allows you to add and modify paths that a conversation can take based on the visitor's input. When you turn AI on in the playbook and add open text responses for the questions, you now have the option to add a new path based on AI topics. To add an AI topic, click the '+" within the new branch and select 'Add AI topic.' In the modal, select which topic you would like to branch on. If you have grouped your topics while creating you can use the drop down in the modal to quickly find the topic you're looking for as well as quickly add topics in bulk.

Note: If you add a topic group to a Playbook path and then later go back add a new topic to the topic group being used in the Playbook path, you will then need to delete and re-add the path in the Playbook because the newly added topic will not automatically be added into the existing topic group in the Playbook path.

‘All Other Topics’ and ‘Not Recognized’ are default conversation paths that will display in the playbook builder. AI topics that are classified but are not branched on will follow the ‘All Other Topics’ path.  Novel, or unrecognized utterances that the AI chatbot is unable to classify will follow the ‘Not Recognized’ path.

Not seeing any topics populate in the field shown above?  Make note: Topics will not appear in the playbook unless they have at least 10 training phrases.  See the How to train your custom AI topics with on-demand training section in this guide above for details on how to train your AI.



"Add AI Topic" will only be an available option when you turn AI on in the playbook and set the node to open text. If it is grayed out, the hover state explains which setting you need to turn on.

How to review conversations with AI

• Navigate to the 'Review' tab under the AI Library
• Select thumbs up or thumbs down to rate the AI classification of the message.  
  • You can click on the AI message to view the full conversation.
  • You can undo a rating if needed.
  • Rating results are displayed under the AI Topic library under the 'Performance' tab.

Alternatively, you can also view AI conversations from the Conversation Search menu in the Drift app.  From the conversation search results, filter for AI enabled playbooks, open conversations to review to see when AI detected a topic and how the bot responded.