[CLOSED] AI and Automation in Help Desk AMA: Ask Us Anything [November 17-21]

[CJuliano]

Thank you all for joining us Nov 17-21 for this AMA. Check out some other help desk resources and keep your eyes out for Chapter 3: Reporting in the Help Desk.

Help Desk Chapter 1: Why and How to Migrate to HubSpot's Help Desk Workspace

• AMA CLOSED

Help Desk Chapter 2: AI and Automation

• AMA November 17-21

Curious how AI and automation can actually help your support team—not just add more features to ignore?

What This AMA Is About

AI in customer support can feel like a lot of hype and not much help. But when it's built right, it genuinely makes your team faster and your customers happier. During the week of November 17- 21 our Help Desk team is here to answer your questions about how AI and automation work in Help Desk, what they can do for your team, and how to get started without overwhelming anyone.

Meet Your Expert Panel

We've brought together the people who build and scale these features every day:

• Caitlin Juliano - Service Hub Product Marketer
• Rose Sail - Product Manager, Help Desk
• Sophie Merrow - Respond & Resolve Help Desk PM
• Jenn De Mello - Product Lead, Help Desk
• Marc Ollmann - Reporting & Insights Help Desk PM

Example Questions to Get You Started

Not sure what to ask? Here are some questions we'd love to answer:

• What's the difference between automation and AI in Help Desk?
• We have a small knowledge base. Is AI still useful for us?
• How do I know if AI is giving customers good answers?
• What should we set up first—AI features or automations?

How This AMA Works

Post your questions as comments anytime between now and November 21. Our team will check in regularly throughout November 17-21 to provide answers. We'll make sure every question gets a thoughtful response by the end of the week.

No question is too basic or too specific. Whether you're trying to understand what AI Agent can handle, figuring out your first automation, or wondering how to measure if any of this is actually working, we're here to give you clear, practical answers.

Drop your questions in the comments below. 

[CJuliano]

Drop your questions in the comments below. 

[Isabel_c]

Hi, been working on AI livechat and done a lot of different scenarios where I found that the AI is lacking.

Pre-hand off process

• A way for the AI to respond a follow up or reminder that if there is 10-min inactivity then the chat will auto-close
• Be able to create custom property under the conversation object
• AI agent to ask for first name and use that all throughout the conversation
• Ask for email address initially, so if the chat gets disconnected, the live agents can follow up. Skip option if they dont want to provide an email.

Hands off process

• Availability status should sync google/hubspot calendar availability so we dont autoassign the conversation to people who are actually engaged to a meeting
• If outside the business hours when Liveagents are not available, there should be a prompt asking the contact to create a ticket or close the conversation.

Post Hands off process

• Everything is fine while in conversation with Liveagent until the conversation extends to outside the business hours.
  • The agent's availability depends on business hours setup in the settings of hubspot, then the liveagent can become unavailable even while in conversation with the contact.
  • A way to send email reminder before the ticket closes and a button to go back to the webpage in that email where the conversation was created. This should be separate from the email transcript.
  • Detect inactivity and auto-close message reminder

When conversation is closed.

• A contact can still respond to the email transcript and that will create a never ending email exchange. Can we disable reponding to transcript?

[CJuliano]

Thank you for your patience here! 

• A way for the AI to respond a follow up or reminder that if there is 10-min inactivity then the chat will auto-close
• This is not possible right now, but our team is aware it’s something that would be beneficial to customers.
• AI agent to ask for first name and use that all throughout the conversation
• This should be possible with the CRM Data beta https://app.hubspot.com/l/product-updates/?rollout=265897
• If outside the business hours when Liveagents are not available, there should be a prompt asking the contact to create a ticket or close the conversation.
• The ability to create a ticket is on the roadmap
• Ask for email address initially, so if the chat gets disconnected, the live agents can follow up. Skip option if they dont want to provide an email.
  • There is a beta coming for this soon - would you like to be ungated for it? If so- send me an email cjuliano@hubspot.com

[CJuliano]

Hi @Isabel_c 

Thanks for taking the time to give detailed comments and feedback here. We love to see it!

I’m going to send this along to our product team and we can get back to you with what is on the roadmap.

[ppkazo]

Can AI help your business to  bring fresh leads, or customers ?
After working in this field for more than 5 years, I am starting a digital marketing business so as per late 2025 how is the scope going with the advance in AI and agentic tools. Now what should be the focus areas of services.  Can you share what is changing and how AI can ease out the things and can make me more time-efficient. 

[CJuliano]

Great question! There is a ton we could go into here but I'll hit some highlights. AI in customer service is definitely changing how businesses operate, and you're right to think about time efficiency.

Here's what I'm seeing with HubSpot specifically:

Yes, AI can help generate leads

HubSpot's Prospecting Agent is built for exactly this. It researches potential customers, finds their contact info, and even personalizes outreach at scale. So instead of spending hours manually building prospect lists, the AI does the research work while you focus on actually talking to qualified leads.

AI also helps you scale support without scaling headcount

On the service side, AI in the Help Desk workspace handles repetitive work—suggesting responses based on your knowledge base, summarizing long threads so you don't need to read 30 messages to get context. That frees up time for strategic work.

HubSpot's Customer Agent takes this further by handling entire customer conversations autonomously. It can answer questions, troubleshoot issues, and resolve common requests without any human involvement. You set the guardrails for what it can handle, and it escalates to your team when needed.

I have a blog post here on AI in the Help Desk with more awesome AI features. 

AI improves customer experience, which protects revenue

When customers get faster, more accurate answers (because AI is pulling from your actual help content), they're happier. Happy customers stick around longer and are more likely to upgrade or refer others. So AI in service protects and grows your existing customer base—often more valuable than new acquisition.

The Customer Agent piece is especially powerful here—customers can get help 24/7 without waiting for business hours, and your team isn't buried under simple "how do I reset my password" tickets.

[Premier_Will]

Hello! 

Can we deploy the Customer Agent to operate in only specific categories?

Essentially, we find that the Customer Agent can't handle particular action-based queries well, but can thrive in other areas - for example general enquiries or questions. We want to avoid the agent replying to the customer and then a User Agent having to deal with it anyway, causing excess for the customer.

Many thanks,
Will

  

[CJuliano]

Hi there!

Interesting! Thanks for sending this question. The customer agent should trigger a hand off to a human when it can’t answer the question, the visitor requests to speak to a human, or the agent is paused.

It sounds like customizing the handoff process could be useful here (Knowledge Base article). With this you could use specific action based words or phrases to hand off to your User Agents.

You could also use Workflows with the Rotate owner action on tickets, or the Assign conversation action for conversations on any channel.

For live chat or Facebook Messenger specifically, you could use Rule-Based Bots.

These options allow adding more controls on handoffs to Customer Agent. The product team is currently working on increasing visibility and adding templates for these tools to make setup easier.

[Premier_Will]

Thank you, most helpful!

[Monjogard]

Hello
I have two questions (or rather, one question and one suggestion)

• Will the customer agent train on existing tickets and responses from "real agents"? Or can it only use website pages/knowledgebase as source?
• While I appreciate the effort of scanning tickets for potential knowledgebase drafts, it would be much more efficient if you could select a closed ticket to generate a knowledgebase article from. Is this somehow possible today?

[CJuliano]

Hi @Monjogard  Thanks for these questions!
1. Today, Customer Agent only learns from Knowledge Base articles, uploaded files, and Short Answers that agents manually add in the Help Desk. It doesn't auto-train on ticket history because of confidentiality risks—not just PII (which is detectable), but contextual secrets like unannounced features mentioned to one customer, special deal terms, or confidential business info that LLMs can't distinguish as "private."

The KB Agent could be a good option to leverage and we have a feature where there is a Knowledge Gap where human agents can go to Help Desk and tell Customer Agent within the ticket the correct answer, which will show up as a recommended short answer to approve.

2. If I understand correctly, it seems as if you’d want a mechanism within the ticket object to generate a knowledge article. This is on our radar, yet not possible today!

Let me know if you have any follow up questions here!

[chighsmith]

Thank you @CJuliano and @mollmann for providing a space where our Top Experts, customers, Community members, partners, and others can come together to learn more and ask questions!

Tagging in some of our Top Contributors --@franksteiner79 @Josh @StjepanGrcic @BukunmiOdetayo @beeginman and @HubSpot_Corey: We would love for you to share your burning questions about AI and Automation within HelpDesk with our Product Team!

Don't hesitate to share it in your networks in case your colleagues have questions too!

Thank you and have a lovely day!

Cassie, Community Manager

[danmoyle]

Let's say I want to test Customer Agent without using credits. Is it possible?

[CJuliano]

Hi @danmoyle . Thanks for the question! You can preview and test the customer agent without using credits.

1. In your HubSpot account, navigate to Service > Customer Agent.
2. In the top right, click Test [agent name].
3. To select pre-populated questions to ask the customer agent, click Test questions, and select a question. You can also enter a custom question in the chat widget.
4. To test how the customer agent performs actions, click Actions and select an option. For example, you can see how the customer agent responds to a request to book a meeting.
   

There is more info about this in our Knowledge Base article.

[RReed13]

Most effective way to set Category from an incoming email ticket using AI

[mollmann]

Hi @RReed13, thanks for reching out. If you have SvH Ent you can turn on “category identification” in your Help Desk AI tool settings. The category for each ticket will be set automatically based on the first message of the ticket. Feel free to review and correct values the AI has set so that our system can learn. Let me know if you have any other questions.

[RReed13]

what about in service hub pro?

[CJuliano]

Hi @RReed13! At this time it's only available in Service Hub Enterprise.

[CUrsu1]

Interested to hear what the needs/questions are. Here are some examples of automations I built:
- use AI to categorize tickets - this was a required property to close the tickets, AI is analysing the emails and picks the relevant category
- use Ai to extract a order number from the subject line - customers reply to order notifications, the order number was a required field 
- close Customer agent tickets - we had tickets not closing because the person did not give a thumbs up or the support team wanted them closed if there is a thumbs up, faster than the 24h that they close automatically - we build a workflow to close after 2h