Kathleen Ross, Teachable’s Product Support Associate Director, could see exactly where new users were getting stuck in the product. 33% of new-user support conversations came from the Course Pricing page. 30% from Payment Settings. And over a third of the people writing in weren't angry, they were confused and lost.

The data Kathleen needed was there, but their AI support agent couldn’t see any of it. "Our support was 100% reactive," Kathleen says. "We only heard from creators once they were already frustrated enough to open a ticket."

That’s the limitation most AI agents share, felt by customer success (CS) teams everywhere. Agents have no access to page history, behavioral signals, or what the user has already tried to do, so they only engage users after they’ve already been asked a question. Even if a creator shows in-app frustration, like rage clicks, they have no idea because they didn’t fill out a ticket.

But a few months ago, Teachable found a way around this.

The problem every AI support agent shares

Teachable is a platform used by 150,000+ creators that build, sell, and deliver online courses for their 1+ million students.

Kathleen had already deployed Fin (Intercom's AI agent, known as "Sunny" for Teachable's users). It worked, but it had two limitations most support agents share:

  1. No behavioral context. When a user opened a support chat, the support agent didn’t know which page they had just left, how many times they had clicked the same broken button, or whether they had already read the help doc it was about to recommend. Every conversation and user interaction started from zero.
  2. No proactivity. Fin answered questions from users who know how to ask, but it couldn't reach the ones who got stuck or rage-clicked on pages and closed the tab.

The missing piece: product context

One of the biggest challenges that support teams face is translating what users do in-app into actual business impact. That’s why Pendo tracks what users do in-app, like page visits, feature usage, rage clicks, error clicks, and U-turns, so teams can find where users hit friction. And now, with Pendo Agent Toolkit, that usage data is fed into your AI agent in real-time for proactive customer support.

Together, they give CS agents the context and ability to reach out first. When Pendo detected friction on any of those pages, a webhook fired to Fin with page context attached. Fin opened a chat: "Looks like you hit some friction adjusting your settings. What were you trying to change?"

Fin's message names the problem, and that specificity led to a 200% increase in engagement. The results, six weeks in:

  • 86% of proactive conversations were resolved without escalation.
  • 67.2% resolution rate on Pendo-initiated conversations vs. 54% overall automation rate.
  • 91% never needed to reach a human teammate.
  • 4.10 out of 5 average CX score on conversations, Sunny closed without human support.
  • 38% open rate on proactive messages vs. 9–12% on their previous onboarding pop-up.

One creator was panicking about a domain stuck in “pending” status, and another had $864 in withheld funds, with no explanation. Both situations are time-sensitive, emotionally charged, and prone to ticket spirals. Yet because Pendo could show Fin what each user was doing in-app, Fin answered with specific details and explanations tailored to the user. In the end, both conversations closed with a CX score of 5 out of 5.

"Sometimes, users don't use the same language we do internally," Kathleen explained. "But because Pendo's on that page, Fin could deliver a very specific answer because the agent was actually seeing what the user was doing."

A tip from Teachable: What to do before you start

Set your audience rules before you turn on Pendo Agent Toolkit. When Teachable first activated the proactive trigger, they hadn't scoped it to new creators yet. The tool fired for everyone, including high-earning creators with years on the platform who didn't want bot outreach. Kathleen fixed it within a few hours, but who you send support to matters just as much as what you send.

A few things she advises:

  • Start narrow. Pick two or three high-friction pages and one audience segment, and confirm you’re getting clean data before you expand.
  • Think about cadence. In a product where users move around a lot, fine-tune how often the trigger fires.
  • Pitch it as a specific problem. When Kathleen made the case internally, she didn't lead with AI. "Instead of just doing the big 'AI can help' pitch," she says, "find a specific problem you want to solve. New creators are getting stuck. We need to reach out to them in a more proactive and specific way. That's what got the buy-in."
  • Setup is fast. It didn’t take Teachable more than an hour to get it going. Pierce Healy, Sr. Director of AI Products at Pendo, estimates that setup takes 30 to 60 minutes.

Beyond proactive support

The infrastructure Teachable built for support works just as well for retention. The same Pendo signals that trigger support conversations can also drive feature adoption, making customer support a proactive process: nudging creators toward key workflows, and catching drop-off before it becomes churn.

Teachable’s next trigger is Stripe onboarding: when a creator starts the setup and drops off, Sunny follows up. It’s the same Agent Toolkit, helping users out proactively during a different point in their journey.

Get started with Pendo Agent Toolkit

The Pendo Agent Toolkit is in beta. If you're running an AI support agent, like Fin, Sierra, Decagon, or anything else, and want to improve your customer support team’s impact, reach out to your account manager or join the beta.

This blog includes content from a Teachable webinar. Check it out here.