The Challenge
HeyReach is a LinkedIn outreach automation tool. Sales teams and agencies use it to run outbound campaigns across multiple LinkedIn accounts at once and feed the results into their sales tools. When an account connection or campaign breaks, support questions are often urgent and technical.
Early on, HeyReach's founders showed up for their customers wherever they were: email, WhatsApp, Slack, Discord. But with no shared record, no one knew what a customer had already been told, and customers had to repeat themselves. After a launch on AppSumo, a marketplace where startups sell software deals to early adopters, their conversation volume spiked. The founders could no longer keep up answering customers themselves, and decided support needed a proper system.
The Solution
They decided to bring every channel into Intercom, Fin's AI-powered helpdesk, giving the team a single view of all customer interactions. Fin's Early Stage Program was what made it work at their size. The first attempt at self-serve was a decision tree in the chat: a menu of buttons, then another, then a recommended help center article. Many people couldn't find what they were looking for. "It took about five minutes to get through, and by then people were annoyed and just asked for a human," says Vick Mucunski, Chief Client Officer at HeyReach. Those requests landed on six agents working weekdays, so a queue built up, and anything that came in on a Saturday waited until Monday.
I saw the vision. We were going to be Fin first in every possible way. Every conversation now starts with Fin introducing itself and trying to solve the problem.
After about a year of routing conversations through decision-tree flows, the team added Fin, an AI customer agent, at the very end of each path, where it handled roughly a third of conversations. When Vick took over support, he saw Fin as the most underused asset in the stack. "I saw the vision. We were going to be Fin first in every possible way," he says. "Every conversation now starts with Fin introducing itself and trying to solve the problem."
Putting Fin first exposed how much the quality of its answers depended on the content behind it. "Stale information reflected badly on Fin," Vick says. "Then it hit me: this is exactly what customers see when they try to help themselves. It wasn't Fin, it was on us."
Vick knew the knowledge base needed rebuilding. But the agents were worried about AI replacing them. "Day one, they saw fewer tickets and started fearing for their jobs," he says. "That fear held Fin back: agents hesitated to work on the knowledge base, knowing every article made Fin better." By the second month, they could see the time Fin freed up going into more rewarding work, and knowledge base improvements became a weekly habit.
The Results
The switch to Fin-first paid off immediately. "We literally tripled a lot of our numbers overnight," Vick says. The content work kept the numbers climbing. Fin is now involved in 80% of HeyReach's conversations across chat, email and Slack, and resolves around 60% of those – around 1,800 resolutions a month.
Customers feel the difference in speed. A question that used to take two hours now takes about 15 minutes, and Fin covers the weekends, so nobody waits until Monday for an answer. It shows in the satisfaction scores: customers rate Fin positively more than 90% of the time, and the human team's scores sit in the high 90s.
Conversation volume has grown 58% this year, and the team is still six agents and one technical specialist. Instead of growing headcount, they've moved into higher-value work – several now specialise in technical support, integrations, and billing, taking on the nuanced conversations Fin hands over.
Support has also started feeding the product roadmap. With Fin handling the routine questions, the conversations that reach a human are the ones worth studying. Each week Vick reviews them by theme, and when the same problem keeps coming back, he takes it to the product team to fix it at the source.
Next, Vick wants Fin to resolve more on its own before handing over to an agent. His goal is to push resolution toward 70%. To teams weighing the move to AI-first support, his advice is simple: "Test it on a small segment of long-standing customers who'll tell you the truth. The moment you enable it everywhere, you've got it cracked."
