Eoghan McCabe on Intercom's AI agent Fin: resolution-based pricing is the future, growth rate doubling three years running

Summary

Eoghan McCabe co-founded Intercom in 2011, building what started as a general-purpose customer messaging platform — a JavaScript snippet you dropped into your site that learned about your users and handled sales, support, and marketing in one place. Within two or three years, Intercom was the fastest-growing software company since Salesforce, until Slack beat them on that metric. Fourteen years later, McCabe is back as CEO, and the AI wave has given the company its second wind.

Resolution-based pricing

Intercom was the first company to charge per resolution rather than per seat, and McCabe says Salesforce called him to ask how it was working before adopting something similar. The model aligns incentives cleanly: Intercom only charges when Fin, its AI customer service agent, actually resolves a ticket. When Fin gets better, Intercom earns more. McCabe argues that revenue earned per customer is the real benchmark for AI performance — not benchmark scores — because it captures whether the model is doing genuine economic work.

Fin's performance edge

The gains Intercom has built aren't primarily from model-layer improvements. McCabe says most of the performance advantage comes from orchestration — roughly 200 A/B tests, proprietary RAG tooling, and using multiple models in combination. He believes the next leap for verticalised AI companies will come from tightly coupling their own models with their applications, rather than relying on off-the-shelf base models.

On the human-versus-AI question, Intercom's internal research found that only 65% of tickets human agents marked as resolved were actually resolved. McCabe's view is that well-applied AI already outperforms humans on availability, consistency, and accuracy — the problem is that most deployed AI agents are still cheap wrappers that drag down category perception.

Labor displacement — so far gradual

Fin is mostly handling the easier tickets and serving demand that was previously unmet, including free-tier customers who never had support before. McCabe says Intercom hasn't seen mass human agent layoffs. Buying decisions tend to come from CTOs and CEOs rather than support teams, who have an obvious vested interest in not adopting the technology. Longer term, he expects AI to displace the most repetitive, lowest-satisfaction work first — and notes that most customer support reps don't want those jobs permanently anyway.

Margins and model costs

When Intercom launched Fin roughly four months after ChatGPT, each resolution cost the company around $1.20 and they charged $0.99, deliberately running at a loss. McCabe says margins are now "very, very healthy" without disclosing specifics. His structural view is that companies building deep applications on top of base models will preserve SaaS-like margins — analogous to how SaaS companies ran 80% gross margins on AWS — while pure wrappers will be forced toward commodity pricing as model costs continue to fall.

Growth trajectory

McCabe says Intercom's annual growth rate will have doubled three years in a row by next year, at a scale he describes as "in the hundreds of millions." He frames this as directly attributable to the AI pivot, and says the company has barely scratched the surface of the eventual opportunity — which he defines as all of customer experience, sales, marketing, and success: a market with trillions of dollars of global salary spend behind it.

AI BDR and the broader wrapper market

McCabe is skeptical that AI sales outreach holds up as a sustainable category. His view is that any effective outreach tool gets adopted so fast through internet distribution that the edge disappears almost immediately — the same dynamic that has historically caused marketers to spam every channel into uselessness. On the broader wrapper question, he thinks the right framing is market size: if AI applications end up doing a meaningful share of all work, the total addressable market could dwarf the entire software industry. He's agnostic on whether that means 100x or 1,000x the current software TAM, but says the magnitude justifies the competition flooding into every category.