How to Run AI Agents Inside HubSpot Automations for Smarter Customer Interactions
Most sales teams waste hours manually reviewing customer histories before calls - only to miss key insights buried in old notes. HubSpot's new "Run Agent" workflow action lets you automate this research using AI agents like Claude and GPT, surfacing forgotten pain points and upsell opportunities automatically. Here's how to implement it without blowing your credit budget.
The Evolution of AI in HubSpot Workflows
For years, sales and marketing teams struggled with manual data review before customer interactions. Important insights remained buried in meeting notes, emails, and CRM fields - until HubSpot began integrating AI directly into workflows. The journey started with basic AI actions like summarizing records and researching companies, but the real transformation came with the ability to call custom LLMs.
The initial implementation required configuring agents through their individual interfaces, creating disjointed experiences. As one HubSpot power user noted at 2:15 in the video: "We'd have to jump into each agent's configure tab to run triggers - it felt like working with separate tools rather than one cohesive system." This friction limited adoption despite the clear potential for time savings.
The turning point: HubSpot's 2025 Q4 release introduced the "Run Agent" workflow action, finally allowing teams to call any marketplace or custom agent directly from automation workflows. This 10-credit action became the missing link between HubSpot's data and AI's analytical power.
The "Run Agent" Breakthrough
The "Run Agent" action represents a fundamental shift in how businesses can leverage AI within their CRM. Instead of requiring manual agent configuration, you can now:
- Trigger AI analysis based on any workflow enrollment criteria (task due dates, property changes, etc.)
- Chain multiple agent actions together with other workflow steps
- Store results directly in HubSpot as notes, properties, or tasks
As demonstrated at 4:30 in the tutorial, setting up an agent call takes just three steps in the workflow builder:
- Select your installed agent from the dropdown
- Map the required input tokens (like company ID or contact properties)
- Configure how to handle the output (typically adding a note or updating a field)
Pro Tip: Always add a subsequent workflow action to capture the agent's output. Unattended results provide no value - the power comes from surfacing insights where your team will see them.
Real-World Example: Automated Customer Reviews
Consider this common scenario: Your account managers should conduct quarterly business reviews with key clients, but preparation often gets rushed. Critical context from past conversations gets overlooked, and upsell opportunities go unnoticed.
The automated solution combines HubSpot tasks with AI analysis:
- Create repeating tasks for each customer (e.g., "QBR - Acme Corp") set to recur every 90 days
- Build a workflow triggered when these tasks become due
- Add a "Run Agent" action calling your customer insights agent
- Configure the agent to analyze all notes, emails, and meetings for that contact
- Add a workflow step to create a summary note with upsell recommendations
At 7:45 in the video, you'll see this in action - the agent surfaces forgotten pain points ("Client mentioned scaling challenges in April") and suggests relevant add-on services, all before the account manager even opens HubSpot.
Step-by-Step Setup Guide
Step 1: Install Your AI Agent
Browse HubSpot's marketplace for pre-built agents like Customer Upsell Analyzer or build your own custom agent tailored to your offerings. Ensure it's properly configured to access the necessary contact data.
Step 2: Create the Trigger
Set up your workflow enrollment criteria. For recurring reviews, we recommend task-based triggers with specific naming conventions (e.g., "QBR - *"). For reactive scenarios, use property changes or deal stage transitions.
Step 3: Add the Run Agent Action
In your workflow, add the "Run Agent" action and select your installed agent. Map the required inputs - typically the contact ID and associated company record. For custom agents, you may need to provide additional context tokens.
Step 4: Handle the Output
Add a subsequent action to capture the agent's insights. The most effective options are:
- Create note (for human-readable summaries)
- Update custom property (for structured data)
- Create task (for follow-up items)
Step 5: Test with a Small Segment
Before rolling out to all customers, test with a small group (5-10 contacts) to verify output quality and credit consumption. Monitor results for 1-2 cycles before expanding.
Implementation Time: A basic customer review automation takes about 2 hours to set up and test, while more complex multi-agent workflows might require 4-6 hours of configuration.
Credit Cost Management Strategies
At 12:30 in the discussion, the hosts address the elephant in the room: credit consumption. While each agent run costs about 10 credits (a fraction of manual research time), uncontrolled workflows could theoretically burn through allocations. Here's how to stay in control:
1. Implement Tiered Triggers
Not all contacts deserve the same level of AI analysis. Set up workflow filters to only enroll:
- Customers above $X in annual contract value
- Contacts with "Decision Maker" or "Champion" roles
- Accounts showing recent engagement signals
2. Add Manual Approval Steps
For high-value actions (like sending AI-generated proposals), insert a manual approval step in the workflow. This ensures human oversight before credit-intensive operations.
3. Monitor with Custom Reports
Create a dashboard tracking:
- Credits consumed per workflow
- Output quality ratings (add a simple 1-5 rating property to notes)
- Downstream impact on deal size or velocity
As noted at 14:50, "The key is measuring value, not just cost. If a 10-credit analysis helps close a $10k upsell, that's a 100,000% ROI on your AI investment."
AI Agent Best Practices
Through dozens of implementations, we've identified four rules for effective HubSpot AI workflows:
1. Focus on Augmentation, Not Replacement
The best workflows use AI to prepare human conversations, not replace them. For example, have agents summarize discussions but always include a "Verify with client" disclaimer.
2. Train Agents on Your Voice
Provide sample notes and emails so outputs match your brand's tone. One client reduced editing time by 70% after training their agent on past successful upsell narratives.
3. Build in Feedback Loops
Add a "Was this helpful?" button to AI-generated notes. This data trains future iterations and identifies areas needing improvement.
4. Start Small, Then Expand
Begin with one high-impact use case (like quarterly reviews), prove the value, then expand to other scenarios. This controlled approach builds confidence while managing costs.
Watch the Full Tutorial
See the complete walkthrough of setting up an AI-powered customer review workflow in HubSpot, including how to configure the agent inputs and handle outputs (jump to 5:15 for the key setup demonstration).
Key Takeaways
HubSpot's "Run Agent" workflow action fundamentally changes how teams can leverage AI in their CRM. By integrating agents like Claude and GPT directly into automations, you can:
In summary: Automate customer research to surface forgotten insights, standardize preparation for client interactions, and identify upsell opportunities consistently - all while spending just 10 credits per analysis instead of hours of manual work.
Frequently Asked Questions
Common questions about HubSpot AI agents
HubSpot workflows now support running marketplace agents like Claude, GPT-4, and Gemini, plus any custom agents you've configured. The "Run Agent" action lets you call these directly from workflow steps to analyze contacts, companies, or deals.
You can chain multiple agents together in a single workflow - for example, first analyzing a contact's history with one agent, then passing those insights to a second agent that generates meeting preparation notes.
- Marketplace agents are pre-configured for common use cases
- Custom agents can be tailored to your specific business needs
- Some agents specialize in particular data types (emails vs. call transcripts)
Most agent actions consume around 10 credits per execution. For example, analyzing a contact's history to generate upsell opportunities costs 10 credits - a worthwhile investment compared to manual research time.
Complex agents analyzing multiple record types (contacts + companies + deals) may cost slightly more, while simpler agents doing basic summarization might use fewer credits. HubSpot provides credit estimates before saving each workflow action.
- Basic analysis: 5-10 credits
- Complex multi-record analysis: 10-15 credits
- Custom agent configurations may vary
A common use case is setting up quarterly customer review tasks that automatically trigger an AI agent to analyze all notes, emails and meeting transcripts, then generate upsell recommendations and add them as a contact note before your call.
One client implemented this and reduced prep time from 2 hours per review to 15 minutes while increasing identified upsell opportunities by 40%. The agent surfaces forgotten pain points and suggests relevant add-ons based on the customer's history.
- Triggers: Recurring tasks or date-based workflows
- Analysis: Full contact history review
- Output: Structured note with recommendations
Yes, you can set workflow enrollment criteria just like any other HubSpot workflow. For example, only trigger AI analysis for contacts with "Customer" lifecycle stage and over $10k in annual contract value.
We recommend implementing tiered automation - simple summarization for all contacts, deeper analysis for high-value accounts, and manual review for strategic accounts where nuance matters most.
- Filter by lifecycle stage, deal size, or engagement
- Use lists to create target segments
- Combine with manual approval steps for critical accounts
HubSpot currently lacks credit consumption warnings, so we recommend testing workflows with small contact segments first. Set up workflow reporting to monitor credit usage and consider adding manual approval steps for large enrollments.
One effective strategy is creating a "credit budget" property on workflows that tracks expected monthly consumption based on enrollment counts. When this approaches your limit, pause new enrollments until the next billing cycle.
- Test with 5-10 contacts before full rollout
- Monitor credit usage in workflow reports
- Implement approval steps for large batches
Agents can analyze all standard and custom properties, plus associated records like notes, emails, meetings, and tasks. They can't access deleted items or data outside the contact's permission scope.
The most powerful agents combine multiple data sources - for example, analyzing meeting transcripts alongside email sentiment and support ticket history to identify at-risk customers before churn occurs.
- All active CRM data fields
- Associated activities and communications
- Permission-based access only
Accuracy depends on your prompt engineering and data quality. In our tests, properly configured agents identify relevant upsell opportunities with 80-90% accuracy when given clear instructions and sufficient contact history.
The key is providing enough examples of "good" outputs during agent configuration. We recommend saving 10-20 ideal customer notes as references for the AI to learn your preferred style and depth of analysis.
- Quality improves with more training examples
- Clear prompts yield better results
- Human review still recommended for critical decisions
GrowwStacks specializes in building custom HubSpot AI workflows that deliver real business value. We'll help you identify high-impact use cases, configure agents with optimal prompts, set up credit-efficient workflows, and train your team.
Our implementation process includes:
- Discovery session to map your ideal customer interactions
- Agent configuration with your brand voice and offerings
- Workflow design that balances automation with human oversight
- Training and documentation for your team
Book a free consultation to discuss your specific needs and see demo workflows in action.
Ready to Transform Your HubSpot with AI Agents?
Every day without AI-powered workflows means missed insights and wasted prep time. Our HubSpot automation experts will design custom AI workflows that surface the right opportunities at the right time - typically delivering 5-10x ROI on credit costs.