Intercom just hit a milestone that should make every product leader pay attention: their AI agent Fin has grown from $1M to $100M+ ARR, handling 80% of support volume and resolving over a million customer issues per week.
That's not a pilot program. That's a fundamental shift in how customer support works.
But here's what most coverage of this story misses: while everyone celebrates faster ticket resolution, almost nobody is asking the more important question—why are customers filing those tickets in the first place?
The Support Automation Gold Rush
The numbers are undeniably impressive. Fin's resolution rate climbed from 27% at launch to 67% today. Intercom charges $0.99 per resolved issue and backs it with a $1 million performance guarantee if resolution targets aren't met. This isn't vaporware—it's outcome-based pricing that actually works.
And Intercom isn't alone. OpenAI CEO Sam Altman recently declared that "customer support is doing great" with AI. Klarna's AI assistant handles the bulk of routine chats. Zendesk reports that 73% of agents say an AI copilot would help them do their jobs better.
The promise is clear: deflect more tickets, reduce handle time, cut costs. And in the narrow lanes of password resets, order tracking, and refund policies, AI delivers exactly that.
But there's a dangerous assumption baked into this entire approach: that the goal of customer support is to make support conversations more efficient.
The Metric That Actually Matters
Here's what Zendesk's own research reveals, tucked between the triumphant AI statistics: 63% of consumers say they will switch to a competitor after a single bad experience, even as AI becomes more prominent in frontline service.
Read that again. The tolerance for bad experiences is shrinking just as AI takes over more customer touchpoints.
This creates an uncomfortable reality for product teams: you can automate support all you want, but if customers keep hitting the same friction points, you're just getting really efficient at treating symptoms instead of curing the disease.
The real question isn't "how fast can we resolve tickets?" It's "why do these tickets exist at all?"
What 67% Resolution Rate Doesn't Tell You
When Fin resolves a ticket about a confusing pricing page, that's a win for support metrics. But it's a failure for the product team.
When AI handles the same onboarding question for the hundredth time, that's efficient. But it's also evidence of a product problem nobody's fixing.
When your chatbot smoothly explains a workaround for broken functionality, you've successfully deflected a ticket. You've also buried valuable feedback about what needs to change.
This is the trap: the better your AI support gets, the less pressure there is to actually fix underlying issues. Automation becomes a pressure valve that prevents problems from bubbling up to the people who could solve them permanently.
Intercom themselves acknowledge this tension. In their conversation about Fin's success, they noted that "the real moat is product feedback loops at scale"—the speed at which insights move from customer conversations into product improvements.
That's the insight everyone is ignoring while celebrating resolution rates.
The Hidden Cost of Deflection
Consider what happens in most organizations today:
- Customer hits a problem
- AI resolves the ticket
- Customer moves on
- Problem remains in the product
- Another customer hits the same problem
- Repeat forever
The support cost stays low. CSAT might even look good. But the product never improves, and customers who don't bother contacting support just churn silently.
This is what Mark Waks from Slalom meant when he told CMSWire that AI has delivered value "in the unglamorous places: triage, summarization, routing and knowledge retrieval"—but these aren't transformation, they're optimization of a fundamentally reactive process.
The companies that will win aren't the ones with the best ticket deflection. They're the ones who understand their customers deeply enough that fewer tickets get created in the first place.
From Reactive to Proactive: The Real AI Opportunity
Imagine a different approach:
Instead of just automating responses, what if you used AI to analyze patterns across every customer interaction—support tickets, sales calls, NPS comments, product reviews, Slack conversations, Gong recordings—and surfaced the underlying themes?
Instead of celebrating how efficiently you handled 1,000 tickets about confusing checkout flow, what if you identified that pattern before it hit 100 tickets and fixed the actual problem?
Instead of waiting for customers to complain, what if you could predict friction points from usage patterns and behavioral signals?
This is the shift from customer support automation to customer intelligence. And it's where the real competitive advantage lives.
The Voice of Customer Problem
The challenge is that customer feedback is fragmented across a dozen systems: Intercom for chat, Zendesk for tickets, Gong for sales calls, Slack for internal discussions, NPS surveys, app store reviews, Twitter mentions, and that random spreadsheet someone started three years ago.
Each system captures a slice of the truth. None of them show the full picture.
Product managers spend hours each week trying to synthesize this manually—reading through tickets, tagging feedback, trying to spot patterns. By the time they've processed last month's data, new problems have already emerged.
This is why so many product decisions still feel like educated guesses. Not because teams don't care about customer feedback, but because the operational cost of actually understanding it at scale is prohibitive.
What Modern Product Teams Need
The Intercom success story proves that AI can transform operational efficiency. The question is whether we apply that same capability to the more valuable problem: understanding customers deeply enough to build products that don't generate as many support tickets in the first place.
This means:
Pattern recognition across all feedback channels. Not just support tickets, but sales calls, user research, social mentions, and internal discussions. Connecting signals that no human could process manually.
Real-time insight synthesis. Knowing today—not next quarter—when a new pain point is emerging, when a feature request is gaining momentum, or when churn signals are appearing.
Evidence-based prioritization. Moving beyond gut feel and loudest-voice-in-the-room to decisions grounded in actual customer impact data.
Closing the feedback loop. Ensuring that insights from customer conversations actually reach the teams who can act on them, not just the teams who can respond to them.
The Uncomfortable Truth About Support Metrics
Here's a thought experiment: What if Intercom's AI agent became so good that it achieved 100% resolution rate? Every single support ticket resolved instantly, perfectly, automatically.
Would that mean the product is perfect? Obviously not. It would mean the product has automated its ability to handle complaints—which is valuable, but fundamentally different from not generating complaints.
The companies that thrive in the next decade won't be the ones with the most efficient support operations. They'll be the ones who understand why customers struggle, what they actually need, and how to build products that anticipate problems before they become tickets.
Intercom's $100M success is impressive. But it's a success at the response layer. The bigger opportunity—the one that creates lasting competitive advantage—is at the insight layer.
Making Customer Intelligence Operational
The gap between "we have customer feedback" and "we understand our customers" is where most product teams live. They're drowning in data but starving for insight.
Closing this gap requires:
-
Unified customer signal processing. Breaking down the silos between support, sales, research, and product to create a single source of truth about what customers are experiencing.
-
AI-powered pattern detection. Using the same kind of sophisticated AI that powers Fin's resolution—but pointed at understanding rather than responding.
-
Actionable insight delivery. Getting the right information to the right teams at the right time, in formats they can actually use for decision-making.
-
Impact tracking. Connecting insights to outcomes, so teams can see which customer problems, when fixed, actually move retention and revenue.
The Path Forward
Intercom deserves credit for proving that AI can fundamentally change customer support economics. But the companies that will build lasting advantages are the ones who realize that efficient support is the floor, not the ceiling.
The real question isn't "how many tickets can we deflect?" It's "how deeply do we understand what our customers need, and how fast can we deliver it?"
That's not a support problem. It's a product intelligence problem. And it's the one that determines whether you're building features customers actually want or just getting really good at explaining why the features you built don't work the way they expected.
The AI revolution in customer experience isn't about automating responses. It's about finally having the capability to truly understand customers at scale—and building products worthy of that understanding.
Ready to transform customer feedback into actionable product insights? See how Pelin helps product teams understand what customers actually need—not just how to respond when they complain.