A New Day for Finance: How Fin (formerly Intercom) Is Rebuilding FP&A for Outcome-Based AI

EDITOR’S NOTE: Intercom rebranded to Fin on May 12, 2026. This post refers to the company as Fin, its new name. Due to the fact that the live conversation took place prior to the rebrand, the livestream replay linked at the bottom, as well as Chargebee’s prior interview with Fin’s director of product for monetisation, Aisling O’Reilly, refer to the company as Intercom throughout.

At Beelieve ’26, Chargebee COO Jeff Sant sat down with Tommy Bettles, VP of FP&A at Fin (formerly Intercom), to get an inside look at what it takes to run finance inside an AI-forward company.

Fin serves more than 30,000 customers, and its AI agent charges per resolution. That means revenue depends on a chain of variables and product behaviors: how many customer conversations take place, how often Fin gets involved, how often it successfully resolves the issue, and what each attempt costs.

Tommy called it “the formula.” Then he made the shift clear: 

“The formula of what I walked through is not something that existed, I don’t know, 15 months ago. It’s a whole new day.”

That set the tone for the conversation. Outcome-based pricing changes far more than just packaging. Tommy explored its impact on how FP&A forecasts revenue, manages margin, partners with product, and works day to day.

Outcome-based pricing changes what ARR means

In a traditional SaaS model, ARR is relatively predictable. Seats expand, contract, renew, or churn. The math can still get complicated, but the basic unit is familiar. But Tommy pointed out an important missing element from this model: it often does not align exactly with customer value.

Fin’s outcome-based pricing model is different:

“For the first time in my career, there’s [now] a more direct connection between the revenue, ARR, cash we’re forecasting and the value customers are receiving. The more successful customers are, the more revenue we have — which wasn’t necessarily the case before. It’s really customer friendly and really easy for sales people to articulate.”

Outcome-based pricing can absolutely simplify the sale. But that simplicity on the front end creates real complexity on the back end. FP&A can no longer forecast from bookings alone — they have to model the chain that produces the outcome, staying tight on how customers use and derive value from the product:

“We’re just having to introduce a lot more complexity to the way we forecast: bringing in machine learning, using new types of ML models that I never thought we would rely on within FP&A.”

Customer-friendly pricing creates accounting complexity

Jeff raised a point many teams feel only after launch: flexible pricing can be great for customers, but painful for accounting.

Tommy shared the example of “Fin buckets,” which work like prepaid buckets of usage. The idea is customer-friendly because support volume is rarely linear. A business may have seasonal spikes around tax day, holiday volume, or launch periods, and fixed commitments can feel risky.

Buckets help customers buy in a way that matches their usage patterns. But they also create backend complexity.

“The accounting for… Fin buckets is real fun. It’s a real delight.”

Behind that tongue-in-cheek comment is the real work of tracking usage and burndown, managing deferred revenue, handling revenue recognition, and aligning billing systems with product usage.

“There is not this [implication that] we are not going to move forward as a business and not make this pricing decision because accounting finds it difficult,” Tommy said.

 But finance has to be involved early enough to understand the downstream implications and fit the model into the right systems.

This is where monetization infrastructure matters. Pricing flexibility creates value only when the business can operationalize it without turning every new model into a manual finance project.

Launching below cost can be a strategic choice

Fin launched its AI agent at margin-negative pricing:

“When we first launched the product, it was margin negative, which is not a sustainable business.”

That could sound reckless in a traditional finance review. But in an early AI market, Fin was making a deliberate adoption bet. Customers were still skeptical of AI, and the market was competitive. Charging a price that fully reflected the early cost structure would have slowed trust before the product had a chance to prove itself.

The plan was not to stay margin-negative, but rather to earn adoption first, then improve unit economics through specific operating levers: falling LLM costs, better model routing, higher resolution rates, and proprietary model development. 

“It’s important to say we are continuing to be on that journey. It’s a never-ending, always-optimizing journey.”

Resolution rate is a margin lever

Resolution rate sits at the center of Fin’s model because it connects product performance directly to revenue and cost.

Tommy described the dynamic like this:

“How many times does the LLM do the task and succeed when they do it? Then we don’t have this waste, right? And so when we started, our resolution rate was in the 30s… and that’s why one of the reasons it was margin negative… and we’re now in the high 60s, which has had a tremendous amount [of impact].”

That progression is the economics of the model.

At low resolution rates, the system still incurs cost on every interaction, but a meaningful share of those interactions never convert into billable outcomes. Said another way, the product is running, but the business isn’t capturing enough of its work as revenue. 

In a separate interview for the Thank You For Vibe Pricing podcast, Fin’s head of pricing, Aisling O’Reilly, confirmed the stakes here:

“Our costs scale with how often Fin is in a conversation, not how often it resolves it. So we’re very incentivized, and we’ve taken on all of the LLM cost risk by having an outcome-based model.”

As resolution rates improve, the same system starts to behave differently. More of those interactions turn into successful outcomes that can be charged for, and fewer attempts end in failure. It’s one of the few levers entirely within their control, and Tommy named it as one of the most direct paths to better unit economics.

For FP&A, that makes resolution rate a core financial driver. It determines how much of Fin’s cost base turns into revenue, and how efficiently their infrastructure is being monetized.

Cost is now a product design input

Tommy described a structural shift inside Fin: product managers now see what it costs to run the product.

That visibility changes how product decisions get made, especially when it comes to model selection.

“If it’s a really high value feature… we’ll pay more for that. But if it’s a little widget… and it [costs] a fortune, then maybe we can not use the latest and greatest.”

He simplified that with an analogy: 

“[If] it’s a Honda Civic problem, we don’t need to give it the Ferrari.”

This new cost awareness has also shifted FP&A’s role upstream:

“My job… is to build a culture in which we’re getting finance more embedded within the product organization.”

That means surfacing cost data early (instead of reporting margin after the fact), helping teams understand tradeoffs, and flagging where cost and value may be misaligned.

Vertical models are part of the margin strategy

Fin also released Apex, its own model trained on 15 years of customer data. Tommy explained why that matters:

“While these frontier models are unbelievable, they’re not specific… We can make fantastic models for our use case that perhaps an Anthropic or OpenAI are just not necessarily ever going to go focus on.”

While building a proprietary model is a significant R&D investment, the bet is about both quality and cost. Better performance drives higher resolution rates. Lower reliance on third-party models reduces cost per interaction. And a model trained on Intercom’s specific use case — 30,000 customers, 15 years of data — can outperform a general-purpose frontier model on that task in ways that become a durable point of differentiation

Pricing decisions now move at product speed

With faster product development, more features, and more packaging decisions, pricing is no longer a quarterly exercise at Fin.

Tommy described how the team manages that pace:

“It’s the one-way door, two-way door framework, which I’m sure most people are familiar with: two-way door, you make a decision, you can walk it back. Those are the types of things you move fast, you learn, you iterate, and if you make a mistake, you just undo it and move forward. [For those] we try to have a very helping hand. It’s with those big, one-way door decisions that… my job is to pull the reins back and [say], ‘Okay, come on guys. Can we just take a beat before we put this out into market?’ But given how competitive and dynamic it is, the business does not have months for us to give them feedback.”

That changes FP&A’s operating model. Finance has to respond at product speed, not planning cycle speed.

FP&A is being rebuilt by AI, too

The session ended with Tommy sharing a second transformation: beyond forecasting AI revenue finance teams are rebuilding their own work with AI.

Tommy shared that he ran a Claude Code hackathon with his FP&A team. After five days, the team had built outputs he described as “mic drop crazy.”

He also referenced a colleague in data science who saw a major shift in the shape of the role. Routine reporting and ad hoc analysis, which used to consume much of the day, started moving toward zero. The deeper research and insight work became the center of the job.

Tommy’s message to his team was direct: ‘We’re famous for starting in spreadsheets or Google Sheets. We’re done with that. You start in Claude Code. That’s where your work begins every single day.’

Then, he added:

“Those words would not have left my mouth a month ago.”

What finance teams should take from Fin’s experience

Tommy’s story points to a new operating model for AI finance teams:

1. Outcome-based pricing reduces friction for customers, but it increases complexity for finance. The team has to understand the full chain behind revenue: opportunities, involvement, resolution, cost, and margin.
2. AI margins can improve through deliberate operating levers. Model routing, falling LLM costs, resolution rate improvement, and vertical models all affect the path from adoption to viable unit economics.
3. Product and finance now share responsibility for margin. When product managers can see cost, it shapes decisions before they’re made — not after.
4. FP&A itself has to change. The same technology finance is trying to forecast is becoming part of how the finance team works.

The FP&A teams that move upstream—closer to product, closer to usage, closer to cost, and closer to AI-enabled workflows—will be best equipped to lead the businesses they’re forecasting.

Watch the full When FP&A Meets AI session on demand