Teachable fed behavioral data from Pendo to Fin for proactive, real-time customer guidance.
67%
Of support tickets are resolved by Fin
200%
Increase in reply rate, from 2% to 6%
4.1/5
Average CX score on Fin-led conversations
Users were stuck and silent
Teachable's support team could see that new creators were getting stuck via product analytics: they'd reach out from the Course Pricing page, get confused on Payment Settings, and bounce off Course Setup before they'd built anything.
By the time a creator submitted a ticket, they'd already lost the moment where a quick answer could've kept them moving. Kathleen's team was only ever meeting creators during moments of frustration.
Their AI support agent (Intercom's Fin, branded "Sunny" for Teachable's end-users) could answer questions, but it was missing signals that revealed when creators were struggling. Pendo was already capturing those signals, but Fin couldn't act on them.
“Support was 100% reactive. We only heard from creators once they were already frustrated enough to open a ticket with us.”
Kathleen Ross, Product Support Associate Director @ Teachable
Teachable started by mapping a problem. Using Pendo data, Kathleen's team identified exactly where new creators hit walls: the Course Pricing page (33% of new-user support conversations originated here), the Payment Settings page (30%), and Course Setup. These were the predictable failure points in the launch journey.
They also layered on sentiment data: 39% percent of users writing to support were simply confused. That's a different problem than an upset customer, and it calls for a different response.
Teachable's
Advice
Before you build anything, pull your support data by page. Where are new users writing from most? What are they feeling when they write in? Confused is different from frustrated, and proactive guidance will land differently on each.
Teachable had already deployed Fin, and the agent was good. But like most AI support agents, it had a fundamental limitation: it could only respond when prompted, never to someone stuck and silent.
This is the quiet failure facing reactive AI support agents at every organization. Your agent resolves the conversations it gets, but it never sees the ones where users just quit. Teachable's insight was simple: an agent that waits to be asked will always be too late for users who stop trying.
“We knew we wanted to find and help users, because Pendo showed that users who got support were much more likely to publish a product. We just didn't know how to reach out.”
Kathleen Ross, Product Support Associate Director @ Teachable
When Teachable connected Pendo to Fin, they deliberately scoped the first rollout to three pages and one audience: new creators. Pendo tracks frustration signals, like rage clicks, error clicks, and U-turns, in real-time to show you exactly where users are getting stuck.
Teachable chose the three pages generating the most support volume and tied the proactive trigger to accounts in their first 60 days. That kept the signal-to-noise ratio high and made it easy to measure results from day one.
Teachable's
Advice
Pick two or three highest-friction, highest-stakes pages where getting stuck leads to long-term disengagement. Then, pair that with a clear, focused audience segment. New users have the most to gain from guidance, and you'll have a cleaner baseline to measure against.
The Pendo Agent Toolkit (ATK) is what makes this possible. It connects your Pendo workspace to any AI agent via webhooks, which push real-time signals, and an MCP data connector that lets agents pull session context on demand.
When Pendo detects friction on a tagged page, a webhook fires. The event goes to Fin with page context attached: what page the creator was on, what they were trying to do, what type of frustration signal triggered the alert. Fin then opens a proactive chat: "Looks like you hit some friction adjusting your settings. What were you trying to change?"
Because Pendo has context, Fin's message is specific and smart. Not "Hi, how can I help?" but a targeted diagnosis of exactly what the creator was struggling with and how to fix it.
When Teachable also enabled the Pendo data connector for all conversations, not just proactive ones, the impact was immediate: conversations where Fin had Pendo context resolved at 70% versus 62% without it. Escalations to a human dropped from 24% to 17%.
“I see you're struggling to set up a dashboard. That's because your settings aren't configured properly — here's how to fix it.”
Sunny (Fin), Teachable's AI support agent
Teachable found that the best support conversations they had were ones users didn't initiate. When Sunny reaches out proactively, they saw a 38% open rate on proactive messages, compared to 9–12% on their previous onboarding pop-up.
6% of recipients started an actual conversation in the first six weeks (versus 1–2% on the old pop-up), and 91% of those conversations never reached a human.
The sensitive cases stand out most. One creator was panicking about a domain stuck in "pending" status, which is often time-sensitive and stressful for creators. Sunny stepped in proactively, explained that SSL verification can take up to 30 minutes, and gave clear next steps before anyone's blood pressure peaked.
“Sometimes users don't use the same language we do internally. But because Pendo's on that page, it was able to deliver a very specific answer because it was actually seeing what the user was doing.”
Kathleen Ross, Product Support Associate Director @ Teachable
An 86% resolution rate is the headline, but the more important number is what happens to those who do escalate. 9% of proactive conversations reached a human teammate. Kathleen doesn't see this as a failure. If they need a teammate because they found something actually broken, that means they're getting the support they need, before frustration sets in.
The goal for CS agents everywhere shouldn't be zero escalations, but rather earlier intervention. Even when Sunny can't close the loop alone, proactive support changes the nature of every escalation that follows.
Teachable is already planning what's next: expanding to all pages, targeting existing creators for feature adoption, and adding workflow follow-up triggers for cases like Stripe onboarding.
After six weeks live with the Pendo Agent Toolkit powering their Fin agent, Teachable saw incredible growth:
Of proactive conversations resolved without escalation
These are conversations that started because Sunny reached out, not because a frustrated user created a ticket.
Resolution rate for Pendo-initiated conversations
That's a 13-point improvement, and a direct result of Fin knowing what page a creator was on and what they were trying to do, in-app.
Average CX score
For outreach users didn't ask for, a 4+ star rating means the messages felt helpful, not intrusive.
Open rate on proactive messages
Compared to an average 9–12% on previous onboarding pop-ups, this one named exactly what the creator was struggling with.
“Because Pendo has context, it was able to deliver a very specific answer because it was actually seeing what the user was doing.”
Kathleen Ross
Product Support Associate Director, Teachable
When your users are getting stuck and you're only finding out about it after the fact.
Teachable knew early-stage creators who got support on launch blockers were far more likely to publish a product. The goal was always proactive, they just needed behavioral context.
