Send to AI HubSpot Workflow Action

Send to AI HubSpot Workflow Action allows users to send custom prompts to be processed by a range of AI models, integrating data from CRM records and enabling bulk processing through workflows.

Purpose: The "Send to AI" action is designed to leverage AI capabilities to process custom prompts, utilizing data from HubSpot's CRM.
Supported AI Models: Includes OpenAI GPT-3.5, GPT-4, Cohere Command, Cohere Command with web browsing, and Anthropic Claude 2.

Requirements: Users need to have AIssistify account (app.aissistify.com) and their HubSpot portal connected.

Setting Up the "Send to AI" Action

Open or create Workflow in HubSpot, setup trigger and add an action, list of actions is on the left sidebar, find AIssistify app and Send to AI Action.

Choose AI Model:

• • Click on the dropdown menu labeled "Choose AI model" and select the AI model that best suits your task. Options include "OpenAI Basic (GPT-3.5)" for general requests, "OpenAI GPT-4 Turbo" for more complex and creative tasks, "CoHere Command," "CoHere WebBrowsing" for responses based on web data, "Google Gemini Pro" or one of the Anthropic models.

 

Set AI Creativity Level:

• • Next, decide how much creativity you want the AI to apply in its responses. The dropdown menu labeled "Set AI Creativity Level" allows you to choose between "Conservative," "Balanced (default)," "Creative," and "Inventive." Each level dictates the balance between accuracy and creative interpretation.
  • Technical details: AI creativity setting controls what is called Temperature of AI model. For most of the use cases in CRM automation please stick to the default setting or a Conservative one when you need precise answer. For creative tasks like generating sales email drafts or any other marketing/sales content you can try higher creativity levels.

 

Prompt for AI

• • In the "Prompt for AI" field, type your custom prompt that the AI will process. Utilize the personalization tokens to include dynamic content from your CRM, such as the company name or contact details.

 

Property to Include with Prompt

If you want to include additional CRM properties or a result of one of the previous workflow actions with your prompt, select the relevant property from the "Property to include with prompt" dropdown menu. This is mostly useful when creating chain of Workflow Actions.

For example you can use Cohere with webbrowsing model to search for last year revenue for a provided company name. And as a next action use GPT-3.5 to extract number from the response to save just the number in the Company (or other object property). 

PDF Attachment Feature

You can now include PDF attachments in your prompts:

1. Select Your PDF Source:
   - HubSpot Property: Attach PDF files stored in HubSpot properties
   - Public URL: Include PDFs from public URLs (e.g., Google Drive, Dropbox)

2. How It Works:
   - The PDF will be parsed, and its text content will be appended to the prompt.
   - Each page consumes approximately 400-500 tokens.
   - For longer documents, consider using models with larger context windows like Google Gemini or Claude Opus.

3. Use Cases:
   - Summarize and analyze file content
   - Extract data from PDFs
   - Ask questions about file content within your prompt

 

Split Response Feature

This new feature allows you to divide the AI's response into multiple structured outputs:

1. Enable Split Response:
   - Choose "Yes" to activate the split response feature.

2. Set Response Separator:
   - Specify a character or string to use as a separator (default is "#" if left blank).
   - The AI will use this separator to divide its response into parts.

3. Outputs:
   - Response Part 1 to Response Part 5: Contains the split parts of the AI's response.
   - If fewer than 5 parts are generated, the remaining outputs will be empty.

4. Use Cases:

Use single prompt to get multiple outputs. Ex. you can prompt AI to categorize customer request, identify its urgency and prepare draft of response and instruct it to separate category, urgency and response with #. This way you'll get three outputs: Response Part 1 will contain category, Response Part 2 urgency and Response Part 3 - draft of the reply. 

 

Save Result to Timeline?

Choose whether to save the AI-generated result as a Timeline Event by selecting "Yes (default)" or "No" from the "Save result to Timeline?" dropdown menu. Select "Yes" if you want the results to be recorded as an event for further tracking and visibility.

 

Click the "Save" button to apply the changes. Your "Send to AI" action is now customized and ready to be used within the workflow.

 

How to save AI response to a property?

After AI has generated a response to your prompt, you may want to save it to a property. To do that use standard HubSpot Workflow Action "Copy property value". 

In the field Property or value to copy from, choose Action outputs -> AI Response

And on the next step, choose Property where you want to save AI response. 

In the following example you can see that AI Response was saved to a custom property "Language".

Alternatively you can save result as a Note to the object timeline using 0CodeTools app from HubSpot Marketplace. Learn more here.

For paid subscriptions of AIssistify we offer extended free version of 0CodeTools with 500 free actions per month, please contact our support to get your extended free version.

Best Practices

1. Start with the "Balanced" creativity level and adjust as needed.
2. Use the "Conservative" setting for tasks requiring precise answers.
3. Leverage personalization tokens to make prompts dynamic and relevant.
4. When using PDF attachments, consider the document length and choose appropriate models.
5. For complex tasks, consider creating a chain of "Send to AI" actions, each focusing on a specific subtask.
6. When using the split response feature, provide clear instructions in your prompt for how the AI should structure its response.

Use Cases

• Formatting company names
• Identifying gender of contacts
• Categorizing and organizing CRM data
• Generating personalized sales emails
• Creating engaging email opening lines
• Crafting customer replies (Ticket-based)
• Enriching data with web-searched information
• Analyzing and summarizing PDF documents (e.g., RFPs, proposals, contracts)
• Extracting structured data from AI responses

Tips for Optimal Results

• Be specific in your prompts to get more accurate and relevant responses.
• Experiment with different AI models and creativity levels to find the best fit for each task.
• Use the AI Studio Playground to test and refine your prompts before implementing them in workflows.
• Monitor token usage, especially when using more advanced models or processing attachments.
• When using the split response feature, ensure your prompt clearly instructs the AI on how to format its response using the specified separator.