Building one of the most successful AI agents in customer service required Brian Donohue, VP of Product at Intercom, to break his team’s operating model.
Not tweak it. Break it.
Here’s the story of Fin, the product that reinvented Intercom.
Brian threw out their six‑week cycle—that sacred cadence teams thought was universal truth because Basecamp independently discovered the same rhythm. Teams stopped being stable triads. People stopped owning clear domains. Instead, they identified eight critical workstreams and assembled whoever each project needed—maybe three designers, maybe none.
The result? Fin charges $0.09 per resolution and has a 65 percent resolution rate, saving the stagnant company.
But the cost was real.
Two years ago, Intercom’s bot used seat‑based pricing—customers paid regardless of value. Some paid a substantial amount of money for products they never set up. Others extracted massive value while being undermonetized. The company needed to reinvent itself or die.
People lost clarity about what they’d work on for three months. Product managers couldn’t point to ownership areas. Donahue feels it himself—constantly figuring out his role and how to add value. Destabilization creates anxiety.
When he tried to explain the new operating model during our podcast, he admitted it doesn’t make coherent sense. The structure is incoherent.
And yet it works.
Here’s what I realized: You can’t framework your way to making big products.
We’ve trained ourselves to believe the right operating model unlocks innovation. Agile sprints. Squad structures. OKRs. Product trios. We aggregate feedback, build prioritization matrices, and remove judgment in favor of signals. The process becomes the product.
Intercom proved something different. It focused on what matters—only fixing things that clearly need fixing. If the most important work gets done, that’s sufficient.
Donohue found the irony brutal. Two years ago, he predicted AI would amplify product‑led growth. Instead, AI products need significant handholding—not because they’re bad, but because they represent genuine change.
Eventually you have to do the work. Get close to customers. Embrace the messiness. Technology is increasing our ability to do more in less time, disrupting role boundaries because everyone has access to the same knowledge—if they ask the right questions.
The frameworks don’t save you. The willingness to blow them up might.
Find Brian’s full episode on wayofproduct.com
Listen to The Way of Product: Apple Podcasts or Spotify
Show Notes:
In the latest episode of the Way of Product Podcast, I had the pleasure of chatting with Brian Donohue, Vice President of Product at Intercom. Brian has been a critical player at Intercom for over 11 years, where he has navigated the company’s growth and transformation, especially in the ever-evolving world of AI integration.
Brian shared his insights and experiences in transforming product development at Intercom, focusing on building a Fin AI agent that’s set to redefine customer support.
With over two years dedicated solely to AI, Brian discussed the company’s journey from its early machine learning beginnings to embracing large language models.
Connect with Brian on LinkedIn
Listen to The Way of Product: Apple Podcasts or Spotify
Actionable Takeaways
✅ Embrace the uncertainty and potential of AI-driven product innovation.
✅ Aligning incentives through outcome-based pricing instead of traditional SaaS models.
✅ Balancing traditional product management structures with innovative AI development approaches.
Time Stamps
04:15 AI Integration in Product Development
07:40 Architectural and Product Thinking in AI
11:05 Challenges and Innovations in AI Implementation
18:00 Continuous Improvement and Reassessment
24:50 Inherited Product Design Flaws
31:35 Technical Rigor and Product Validation
37:10 Evolving Product Management Practices
42:15 The Role of AI in Modern Product Development
49:20 Outcome-Based Pricing Explained
55:00 AI Transformation and R&D Services
59:30 Adapting Product Development to Customer Needs
66:45 Final Thoughts and Future Outlook
70:20 Connecting with Brian Donahue
