At the Berkeley AI Conference, an Intercom leader described something I've never heard a company at their scale admit publicly: the weekend after ChatGPT launched, their leadership team got together and decided to rip everything up. Not the roadmap. Not the Q1 plan. Everything. The roadmap, the company strategy, the metrics they tracked, the organizational structure, the company values. They essentially restarted the company.
This wasn't a startup. This was a business with hundreds of millions in revenue and an existing machine that was working. And they tore it down because their CEO looked out three to five years and concluded: the world is going to be the opposite of what it is today.
I've been in product for 15 years, and I've seen plenty of "pivots." Most of them are deck pivots. The strategy changes. The org chart stays. The roadmap gets a new theme. The actual day-to-day work shifts by maybe 20%. Intercom did something genuinely different, and the results tell the story.
The Startup-Within-a-Startup, Taken Seriously
The move Intercom made wasn't just creating an AI team. Lots of companies do that. They created a self-contained, self-functioning team that sat in a different part of the office. This team wasn't burdened by existing red tape. They had one job: build, build, build, and make the AI product successful.
I've worked in big companies where "innovation teams" get announced with fanfare and then die because they can't ship without going through the same approval process as everyone else. Intercom's version worked because they actually severed the connective tissue. The team had autonomy, not the org-chart kind where you still need four sign-offs, but the real kind where you're physically separated and operationally independent.
Once the product hit escape velocity, they merged it back in. That sequencing matters. Build first, integrate second. Most companies try to do both simultaneously and end up with a thing that's too slow to be a startup and too disconnected to be part of the core business.
The Pricing Model Nobody Was Ready For
The other move Intercom made that caught my attention was their shift to outcome-based pricing. Instead of charging per seat, they charge per resolution. Their AI agent, Finn, resolves customer support issues. The customer pays for resolutions.
The concept is elegant: Intercom's revenue is directly aligned with the value their AI delivers. More resolutions, more revenue. Better AI, better outcomes for everyone.
But the execution exposed a problem nobody talks about: buyers aren't ready for consumption pricing in categories where they've never seen it. The Intercom speaker was candid about this. Support leaders who've only ever bought seat-based products didn't know how to explain variable billing to their CFO. They'd see overages and panic, not because the product wasn't working, but because the billing model was unfamiliar.
This is a real product management lesson. The pricing model can be technically superior and still fail if customers don't have the mental model for it. Intercom's solution was education: transparently communicating what "resolution" means, showing customers the cause and effect between configuration changes and billing outcomes, and training account teams to walk customers through the new motion.
For PMs building AI products, this is worth internalizing. Your pricing innovation might be ahead of your customers' procurement process. That's not a reason to retreat to seat-based pricing. It's a reason to invest in education as a product feature.
The Six-Month Market Shift
Here's the stat that stuck with me most. The Intercom speaker said that one year ago, prospects would ask: "Why do I need an AI agent for customer service?" Today, that question never comes up. The question is now: "How do I know you have the best AI agent?"
In traditional SaaS, that kind of market evolution takes three years. In AI, it happened in six months. The implication for product teams is severe: if you're building your roadmap assuming the market will move at normal speed, you're already behind.
This matches what I'm seeing across the AI product landscape. The adoption curve isn't following the Crossing the Chasm model that most PMs were trained on. Early adopters aren't trickling in. They're flooding in, and the majority is right behind them. Box's CTO said the same thing at a different panel: once AI products start working well, the chasm crossing will happen faster than any previous technology wave.
What I'd Steal From This Playbook
I've shipped products inside companies ranging from scrappy startups to enterprises with thousands of employees. Here's what I'd take from Intercom's approach if I were leading an AI transformation today:
Make the bet over a weekend, not a quarter. The speed of Intercom's decision was the decision. If they'd spent three months doing an analysis of whether to pivot, the window would have closed. The leaders who move fastest in AI aren't the ones with the best data. They're the ones with the best judgment about when to act on incomplete information.
Physically separate the team. Not a Slack channel. Not a tag in Jira. A different part of the building. Different rituals. Different approval processes. The goal is to remove every friction point that exists in the main business. You can always add governance back later. You can't add speed back later.
Change the metrics, not just the roadmap. Intercom didn't just build an AI product. They changed what they measured. If you pivot to AI and keep tracking the same KPIs you tracked before, your team will optimize for the old world. New strategy requires new scorecards.
Invest in customer education as a first-class product feature. Outcome-based pricing, consumption models, resolution-based billing, these are better models for AI products. But your customers need help getting there. Build the education into the product experience, not just the sales deck.
Restructure your team every six months. Intercom's solutions leader said they've changed their team model three times in one year. That sounds chaotic, but it's actually responsive. The market is moving faster than annual planning cycles. Your org structure should move at least as fast as your roadmap.
The Uncomfortable Part
The part that made the room quiet was this: every speaker who'd done a major AI pivot described it as painful. Not painful in the abstract. Painful as in: people lost their roles, roadmaps they'd spent months building got thrown out, features they were proud of became irrelevant.
But every single one of them also said the same thing: the cost of not pivoting was higher. The companies that hesitated, that tried to run AI as a side project while protecting the core business, are the ones losing ground.
One speaker put it plainly: "If you're expecting things to stay the same, you're probably with a company that's not growing."
That's the real takeaway from Berkeley. The AI pivot isn't optional. And the companies that do it fastest, most completely, and with the most conviction are the ones pulling away.