How Intercom Reaccelerated Growth with Outcome-Based Pricing

“What’s your AI strategy?”

If that boardroom question makes you sweat, you're not alone. Many of us CFOs are still sorting through where AI actually fits into our finance orgs in a way that is practical, not just performative.

That’s why Brex created The CFO Guide to AI Strategy. It’s a straightforward look at how finance teams are using AI today to cut manual work, move faster, and make better decisions, without overcomplicating things. Get the guide:

I've been tracking Intercom's AI pivot for a while now. They did something that most SaaS companies are terrified to try: betting the entire company on a new pricing model while the old one is still working.

And the numbers are insane.

As my Mostly Growth co host Kyle Poyar covered, their AI agent, Fin, went from $1 million to $12 million ARR in a single year. That was a year ago. They're now at $400M ARR overall, with Fin about to pass $100M. Fin is on pace to be half of Intercom's total revenue early next year.

And to really make it something for the SaaS silver screen: their new customer NRR jumped from 112% to 146%.

For context, 112% is solid. It's "you're doing good. you have some embedded growth. keep it up". But 146%? You don't get to 146% NRR by simply raising prices. You get there by building something customers actually use more over time. A lot more.

Which is crazy, because Intercom was in rough shape before this.

Five consecutive quarters of declining net new ARR.

I've been on the other side of this, by the way, as a CFO paying for Intercom. If I’m being honest, it was expensive as hell on a per-seat basis, and all of that hit my gross margin. It started to feel less like a tool and more like a tax. And you never ever wanna be called a tax in SaaS. The value linkage was suspect, if not broke .

Then they went all-in on AI and re-designed the entire business model.

I sat down with Dan Griggs, Intercom's CFO, on Run the Numbers to unpack exactly how they pulled this off. Then I talked to their co-founder Des Traynor on Mostly Growth to get the product and strategy side.

What I found was a masterclass in how to price AI when you're selling outcomes, not seats. And how to think about your P&L as it morphs over time.

GPT 3.5 dropped in November 2022. By the following Monday, Intercom had made the call: all-in on AI.

That timeline is bananas, as Gwen Stefani would say.

Most companies would still be booking flights for a non descript conference room in Orlando to “discuss”. Intercom had their AI-ML team lead in a room with the founders within days, delivering a simple message: "Shit’s different this time."

Dan told me how it went down:

The three of them built conviction that this was going to change customer service forever. And once you believe that, the next step is obvious. Scary, but obvious.

It was existential. You can either disrupt yourself or wait for someone else to do it for you.

Within three to four months, they had the first version of Fin in beta. GA by May 2023. For a company doing hundreds of millions in revenue, that's a wild speed to market.

From a finance seat, this is the part that's hard to wrap your head around.

How do you greenlight a bet this big, this fast? There's no model for "what if LLMs change everything." There's no comp set. You're making decisions with massive resource implications based on conviction, not data.

Before tech, Dan worked in FP&A for an ice cream company (what a glow up, from frozen treats to AI). His old boss Tony had a saying for moments like this:

He'd be sitting in a forecast review, and someone would say, "We don't know what this new thing is going to do to the business." And Tony's response was always: "Well, you know it's not zero."

As finance people, we're trained to be precise. We don't like making stuff up. But in the face of real ambiguity, you have to put something down… a range, a scenario, a stake in the ground… and then refine it as you learn.

They’d do it live.

And they weren't kidding. Per Eoghan, Intercom moved nearly 80% of their R&D allocation to Fin while it was still a single-digit percentage of revenue. They grew the AI team from 6 to 60 people in three years, hiring PhD-level researchers. They bought a $1M .ai domain, built Fin its own brand and website, and drove 100% of paid marketing traffic to Fin, not Intercom. They were marketing the new business over the old one before it was even material to the P&L.

Not every software category is ripe for this kind of AI disruption. Dan made this very clear:

That distinction matters. Most SaaS helps people do their jobs faster. AI in customer service actually does the job.

The list of categories where that's true is still pretty short: customer service, coding, some creative work. That's kind of it for now. Sales hasn't had its moment yet. Finance hasn't either.

But customer service was the obvious first domino. And it wasn't just because AI could handle the conversations. It's because the outcome is measurable.

Dan put it simply:

Compare that to, say, a marketing agent that helps you write better emails. How do you measure "better"? Open rates? Revenue attribution six months later? Good luck getting a CFO to sign off on outcome-based pricing for that.

That dog don’t hunt.

Customer service gives you a clean scoreboard. Resolution or no resolution. That's what made the pricing model possible.

Once you accept that AI is doing the work, not helping humans do the work, the pricing model has to change.

Seats made no sense. There's no human sitting in a seat using Fin to do their job. Fin is doing the job.

Dan walked me through how they landed on it:

They considered a few options.

But they kept coming back to outcomes.

They landed on 99 cents per resolution.

First-in-market with that model, by the way. Everyone else was still fumbling around with seat-based AI add-ons or charging per API call. Intercom went straight to: we get paid when there's dope on the table.

That's a bold strategy, Cotton, let’s see how it plays out for him!

If Fin doesn't resolve the conversation, Intercom doesn't get paid. The incentives are fully aligned with the customer, which sounds nice in customer pitch, but it means your revenue is directly tied to how good your product actually is.

And here's the 180: remember how Intercom used to feel like a tax? Expensive per seat, value linkage broken, hitting your gross margin whether it performed or not?

At 99 cents a resolution, that math (and more importantly sentiment) inverts completely.

I've been on the hook for forecasting customer support costs before. Depending on the company, cost-per-ticket can range anywhere from $5-6 on the low end to $20+ for B2B SaaS. And that's not just wages; that's fully loaded. Training, tools, QA, management overhead, the works.

So when someone tells me I can resolve a conversation for 99 cents… Sign me up.

Even at the floor, say your support org is incredibly efficient and you're running at $5 a ticket, you're still getting 5x ROI on every resolved conversation. At $20 a ticket? That's 20x. The math is so lopsided it almost feels like a mistake. Highway robbery.

Dan confirmed they know exactly how good of a deal this is:

Intercom went from "expensive vendor we're stuck with" to "no-brainer line item."

So Intercom charges 99 cents per resolution. But what does it cost them to deliver that resolution?

This is where it gets into the inside baseball for the finance people tasked with estimating this for the first time in their careers. It's somewhat similar to guessing your GCP bill and somewhat different.

The key cost driver is LLM compute - tokens in, tokens out. And not all conversations are created equal.

Dan gave me a good example:

A password reset might take one ping to the model. A complex migration might take a dozen back-and-forths, more tokens, more compute, more cost.

So how do they manage the margin?

Resolution rate is the biggest lever.

When Fin first launched, it was resolving about 25% of conversations. Now they're north of 65-70%. That’s both a product win and a unit economics win, because every point of resolution improvement means you're generating revenue on conversations that used to cost you money with no return.

The other lever is model selection. Dan explained that Fin isn't one model, but many models working together:

An engineering friend once explained it to me like cloud environments: you don't spin up the expensive instance for a task that runs once a week. Same logic here. You don't throw the Ferrari V12 at a task that requires a Honda Civic 4 banger.

And they're not just using off-the-shelf models. Intercom is building their own:

That's a nice moat. They have years of customer service conversations to train on. A startup trying to compete doesn't have that institutional data.

The last piece: falling LLM costs. Yes, the cost of running these models keeps dropping. But Dan was honest that it's not as simple as "costs go down, margins go up."

In other words, some of the cost savings get reinvested into better performance (something I wrote about recently). You're not just pocketing the margin; you're using it to widen the resolution rate gap. It’s a virtuous loop.

Since, if you're pricing on outcomes, flows right back to revenue anyway.

That's essentially the pitch Intercom made to themselves. They knew the value they were delivering. They knew customers were paying $5, $10, $20 per ticket. And they priced at 99 cents anyway.

I asked Dan if he ever thought about how much money they were leaving on the table.

If you believe AI is going to transform customer service, then the game isn't maximizing revenue per customer right now. It's building an on-ramp so steep and frictionless that everyone slides right in.

But underpricing Fin was only half of it. Intercom also actively destroyed roughly $60M in existing ARR by restructuring their legacy pricing, reducing prices on everything that wasn't the helpdesk-plus-agent combo. They didn't just leave money on the table with the new product… They went so far as to light old money on fire to force the migration.

There was also a trust problem to overcome. Remember, this was 2023. The memes about hallucinating chatbots were everywhere. The idea of letting an AI talk directly to your customers, without a human in the loop, felt risky as hell.

At 99 cents per resolution, the risk calculus changes completely. It's not "do I bet my customer experience on this new AI thing?" It's "for less than a dollar a conversation, why wouldn't I try it?"

Low price. Low risk. High volume. Land the customers now, expand later.

It's a classic land-and-expand play, except the expansion happens automatically as customers throw more conversation volume at Fin. You don't need an upsell motion when usage-based pricing does the work for you.

Here's the question every CFO asks when they see a new product eating into the old one: are we cannibalizing ourselves?

Dan admitted he worried about it too.

The fear makes sense on paper. If Fin resolves 70% of conversations automatically, don't you need fewer humans in seats? And if you need fewer humans, don't customers buy fewer seats? And if they buy fewer seats, doesn't your core business shrink?

Here's what actually happened: not much.

Some customers reduced offshore seats over time. But most didn't cut heads, but redeployed them. Customer support reps got shifted to high-touch accounts, or moved into customer success, or handled the complex stuff that Fin couldn't resolve.

So instead of "we need fewer people," it's "we can handle way more volume with the same people." That's a more nuanced story.

And even if cannibalization does show up eventually? The math still works.

Dan broke it down:

Said differently, a seat might cost the customer $100/month. If that human was handling 300 conversations, and now Fin handles them at 99 cents each, that's $297 in Fin revenue versus $100 in seat revenue.

This isn't the snake eating its own tail. It's the caterpillar becoming a butterfly (also, I can’t believe I just made that lame analogy). The old business doesn’t so much get cannibalized, but transforms into something with better economics.

Intercom's playbook isn't going to work for everyone. Customer service had a unique set of conditions:

But there are a few things worth stealing.

The Intercom team went from "this technology is different" to "first beta in market" in three to four months. That doesn't happen if you're waiting for perfect data. Sometimes you have to put something down, know it's not zero, and refine as you learn.

But this also requires brutal honesty about where you actually are. Intercom had five consecutive quarters of declining net new ARR. The product was stale. The pricing model was broken. When you're staring at metrics like that, you have to call it what it is, not rationalize it away.

And you have to know which decisions are one-way doors and which are two-way. A two-way door you can walk back through if it doesn't work. A one-way door locks behind you. Intercom's bet on AI wasn't technically irreversible, but waiting six more months might have been. Sometimes the cost of hesitation is higher than the cost of being wrong.

"Did it resolve or not" is binary. “Did the AI make your marketing better" is not. Before you chase this model, figure out if you have a scoreboard that others will buy into.

Intercom priced Fin at 99 cents when they could have charged multiples of that. But they were optimizing for adoption, not extraction. If you believe you're early to a category-defining shift, land grab now and expand later.

The fear is that the new thing kills the old thing. What actually happens is more nuanced. Customers absorb volume, redeploy resources, and the overall revenue per customer often goes up, not down.

Multi-model architectures, variable token costs, resolution rates that move every quarter… none of this fits neatly into a traditional SaaS model. Get comfortable with ranges and scenarios. Refine as you go.

Intercom used to feel like a tax on my P&L. Now it's a case study I was pumped to write about. Funny how that works.

You gotta salute those who are bold enough to burn the ships and bet on themselves.

No doubt there will be more SaaS companies making the same scary, and exciting, calculus over the coming months.

Wishing you an outcome pricing based strategy that is measurable,

CJ