What Miro and Fin Learned Running Pricing Like a Product

At Beelieve ’26 in London, Vinay Seshadri, Sr. Director of Product Management at Chargebee, sat down with two operators who have been running pricing as a core product discipline at scale for years: Yuliya Malysh, Head of Self-Serve Business and Growth at Miro, and Aisling O’Reilly, Director of Product, Monetisation at Fin (formerly Intercom, which Salesforce announced it was acquiring for approximately $3.6 billion in June 2026).

The session covered what that practice looks like under pressure: how product teams should read signals to understand when a pricing model isn’t working, why some of the best pricing decisions deliberately leave short-term revenue on the table, how AI changes the cost calculus in product-led environments, and what it means that software agents are now beginning to buy software on their own.

We’ve been following both companies’ pricing journeys for well over a year. At Beelieve San Francisco, we explored how Miro used pricing to drive collaboration and how Fin rebuilt its finance organization around outcome-based pricing. This conversation connected those stories, showing what it actually takes to run pricing as an ongoing product discipline.

Across each of these conversations, three themes kept resurfacing: pricing starts with a hypothesis, customer feedback tells you what to change, and long-term trust is built through the operational experience of pricing (not just the pricing model itself).

Pricing as a Product Practice

Yuliya opened with a framing that product managers will immediately recognize. Pricing, she argued, follows the same loop as any other product function: you form a hypothesis, observe how customers respond, and change course when the data tells you to:

“You have users, you have a solution for those users, you try to deliver value, you have hypotheses, you test and iterate to find the ideal solution, and you keep improving.

“Last year we launched three new products on top of Miro, and we applied a product mindset to how we were launching them: trying to find product-market fit, model-market fit, testing in the field and in a scaled self-serve motion. Within a few months, we got signals that invalidated our initial hypothesis. So we made a pivot.

“That’s why pricing is product to me — because it keeps evolving, and it should stay close to the buyer, to the customer, and to how you deliver value.”

Yuliya described pricing as a process of forming hypotheses and iterating. Aisling then shared a story that showed exactly what looks like when a hypothesis fails.

What Fin Learned When Customers Pushed Back

Aisling’s first story cut to the core of what outcome-based pricing demands from a team.

Fin, the company’s AI agent for customer support, charges only when a conversation is successfully resolved. Today, that approach is widely praised. But initially, it wasn’t:

“We get a lot of praise for it now, but when we first launched it, it was a bit of a disaster. Customers kept telling us they didn’t know how to predict it, that they were finding it really hard to make the business case internally. 

“The obvious question was: should we change the pricing model?”

Fin didn’t jump to that conclusion:

“We actually sat down with customers — got both the champion and the decision-maker in the room — and asked them how exactly they think about budgeting, ROI, and forecasting spend. Through that process, we realized it actually wasn’t the pricing model that was wrong. It was the tooling we provided around it.”

The team built calculators. They introduced crediting mechanics and longer free trials. They gave customers the visibility and predictability they needed to trust a model they couldn’t yet fully see. Aisling notes the model now gets a lot of praise — a contrast to its rocky start.

Together, Yuliya’s and Aisling’s stories point toward the same conclusion: treating pricing like a product is as much about diagnosing customer behavior before deciding what to change, as it is about shipping pricing changes themselves.

Where You Sit in the Org Shapes the Bets You Can Take

Both speakers have unusual org chart positions that give them significant surface area.

At Fin, the pricing function reports to the CTO, Darragh Curran. Aisling gets asked why constantly:

“In pricing, you are so often limited by the systems and technology that underpin your billing infrastructure. The fact that pricing rolls into Darragh has unlocked us in a number of places to move more quickly. But the other reason is a people thing. Darragh is a huge risk taker. He puts customers first, always, and he is not afraid to say we are going to make less money in the near term in order to keep things simple or make long-term progress.”

That last part carries weight at this particular moment. When AI is rewriting the assumptions behind pricing models, you need leadership that can tolerate short-term revenue loss for long-term structural health. That’s not a universal trait, and where pricing sits in the org often determines whether that tolerance exists.

At Miro, Yuliya’s team reports to the Chief Product and Technology Officer and spans both monetization and adoption. That dual remit is the point: the team doesn’t have to negotiate between driving revenue and driving usage. It holds both:

“The only person I have to debate on whether we want to drive more revenue or drive more adoption is myself and the leadership team. That helps a lot. It means we focus on delivering value first, before pulling on monetization. Having all the resources to execute on that — engineers, designers, the ability to iterate on pricing pages and product experiences — accelerates everything.”

The practical result is that Miro can sequence decisions in the right order: figure out where product is delivering real value, then decide when and where to introduce a sales motion. Having engineers, designers, and the ability to iterate on pricing pages and product experiences in a single function compresses that cycle significantly.

The $50 Million Bet That Paid Off

The most striking story of the session was Aisling’s account of Fin’s pricing simplification, a decision to walk away from tens of millions in ARR to fix a model that customers had come to resent.

Through a series of individually rational pricing decisions, Fin’s model had accumulated so much complexity that customers actively mocked it. A dedicated Slack channel was created just to collect the memes.

The team decided to reset.

“We estimated we were going to lose about $50 million of revenue for a $400 million business. That is a scary number… But what we learned was that the $50 million we originally estimated turned out to be $20 million. 

“When you tell customers you’re going to give them money back, sometimes that builds trust, and they start trying new products or redeploying that budget with you elsewhere.

“Three years on, we think that net loss is down to about $5 million.

“What felt like a really scary prospect at the time turned out to be one of the best bets we made as a business for our growth.”

Quantifying the operational friction and brand damage the old model was causing, and testing a migration with a cohort before committing, is how Fin built the internal case and managed the execution.

Fin for Sales: When Customers Define the Outcome

The resolution-based pricing model for Fin worked because two conditions held: the agent was fully autonomous, and the outcome (whether a support conversation was resolved) was binary and consistent across customers.

When Fin expanded its AI agent into sales qualification, both conditions changed. Fin for Sales is priced at $10 per qualified lead (at the time of writing). But what counts as a qualified lead varies significantly by customer. Fin built tooling that allows each customer to define qualification criteria for themselves. They’re only charged when Fin meets their own standard.

It’s an unusual bet. The model requires customers to have enough self-knowledge to define their criteria accurately, and it requires Fin to trust that those definitions will be applied consistently. Aisling acknowledged the risk directly:

“Customers may use it in unexpected ways. But we’re taking the bet that by aligning closely with customer outcomes and value, it will help accelerate growth.”

The principle is consistent with how Fin has approached pricing throughout: when the outcome metric isn’t universal, let the customer define it. That alignment is what makes outcome-based pricing legible to buyers who would otherwise find it too unpredictable to commit to.

AI Changes the Cost Calculus in PLG

The second half of the discussion shifted toward AI.

For years, product-led growth operated on a relatively simple assumption: another free user cost very little. AI has changed that equation. Every interaction now carries real compute cost.

Miro has operated one of the most studied self-serve motions in enterprise software for years. Yuliya described how Miro thinks about that trade-off:

“Previously, adding new customers and users didn’t meaningfully add to infrastructure costs. With AI, every new user carries some cost risk… Right now I’d frame the cost of AI as a new customer acquisition cost. We don’t spend heavily on paid marketing, so that’s actually another lens to look at it through — rather than viewing it purely as a cost to reduce, we’re treating it as an investment in driving adoption and delivering value to the user base.”

The priority is still to understand what high-frequency AI use cases actually look like before optimizing the cost structure around them.

That same thinking extends into enterprise monetization.

For enterprise customers, the credits model Miro has built adds another layer of complexity. Different AI actions consume different numbers of credits based on computational weight: arranging sticky notes is lightweight, while generating a working prototype is substantially heavier. At enterprise scale, that distinction compounds quickly, and admins need visibility into who is consuming credits, on what capabilities, and at what volume before they’ll trust the model enough to expand it organizationally.

“To actually sell AI and deliver trust, you have to invest heavily in usage insights, analytics, consumption data, and controls.”

How Miro and Fin Approach Pricing Research

Vinay asked both speakers directly: in the world of AI, what quantitative and qualitative methods still inform how you approach pricing?

Yuliya’s answer was grounded. Miro runs MaxDiff analysis and choice-based conjoint studies to get structured input on customer preferences, but treats them as inputs, not conclusions. The team forms hypotheses first, then uses research to validate or refute them rather than generate ideas from scratch.

“Rather than doing open-ended research, we go in with something specific to validate. That’s how we find direction rather than just asking people what they think.”

Aisling agreed on method and added one thing that’s new: cost disparity between customer cohorts is now a real variable in pricing decisions. In a seat-based world, the heaviest users don’t cost dramatically more to serve than the lightest. AI usage changes that relationship materially, which affects how teams model unit economics and segment customers for pricing purposes.

“The fundamentals of monetization have not changed. It’s still about your customer. 

“What is different now, is that we have a very real constraint of cost… and new monetization mechanisms, with credits being the emerging one for AI.”

The core work of understanding how customers experience value, what they’ll pay, and how they make decisions internally runs on the same methods it always has. The environment around those decisions is moving faster, with fewer established reference points.

Aisling and Yuliya were consistent here. AI hasn’t replaced pricing research; it has simply made judgment more important.

Agents Are Already Buying Software

The session closed on a question that opened onto a new frontier: as AI agents increasingly make purchasing decisions on behalf of humans, how does that change how you think about pricing?

Aisling noted that Fin already has over a hundred agents that have bought and set up  accounts autonomously. Two implications stood out:

“It used to always be about keeping pricing simple enough for someone to scan your pricing page.

“An agent doesn’t care about simplicity. If anything, they appreciate complexity because it better aligns willingness to pay.

“We also have an MD file that’s a little easier for agents to parse.”

Yuliya pushed the question further by suggesting that humans still shape agents’ behavior:

“There’s still a person behind the agent who created the instructions.

“I think some things will still matter, like what the company stands for and what values it holds… I think brand still matters.”

Rather than replacing traditional buying behavior, agentic commerce may simply introduce another buyer persona — one that values transparency, machine-readable pricing, and explicit commercial logic.

What It Takes to Operate Pricing as a Product

Across the full conversation, a few operational principles held throughout every example:

Diagnose before redesigning. When pricing isn’t working, the problem may be the execution layer: the calculators, the visibility tools, the trial mechanics, rather than the model itself. Fin’s outcome-based pricing survived intact because the team asked customers what was actually hard before deciding what to change.

Cost visibility is part of the pricing product. Dashboards, usage analytics, crediting mechanics, and spend-forecasting tools determine whether complex pricing models are legible and trustworthy to buyers. Miro is building this now for AI at the enterprise level. Fin built it early for Fin and credits it as a primary driver of adoption.

Leave short-term revenue on the table when the model is wrong. Fin’s $50 million simplification bet returned a fraction of the projected loss and unlocked years of cleaner retention and expansion. The ability to make that bet depended on leadership that understood pricing complexity has real, measurable costs in operations, sales friction, and brand damage that sit outside the revenue line.

Match research to hypotheses, not the other way around. Structured research like conjoint analysis is most valuable when used to validate a specific direction. Both speakers run discovery with a working hypothesis already in place.

Build for transparency. Pricing that is clear, published, and predictable before a sales call reduces acquisition friction for human buyers and agents alike. Fin maintains an MD file formatted for machine parsing alongside its public pricing page.

From pricing simplification to enterprise controls to agentic buying, the panelists kept arriving at the same place: pricing is becoming more integrated due to AI. The pricing model, the product experience, the finance systems, the customer research, the organizational structure, and the tooling are increasingly one system. That’s what it means to run pricing like a product. Watch the full session.

Explore Chargebee’s Pricing Labs for frameworks on building pricing as a product.