How Not To Screw Up Your Pricing: Lessons from 200 Pricing Transformations

Pricing has become the hardest product problem in AI.

For years, SaaS companies could get away with pricing that was merely “good enough”. Healthy margins hid a number of mistakes. But AI has changed that. Compute costs are real, heavy users are expensive, and pricing reveals the deeper question of whether your business model is even right for your business.

That’s why Ulrik Lehrskov-Schmidt, who has led more than 200 B2B SaaS and AI pricing transformations, says companies need to stop thinking about pricing as a number and start thinking about it as a structure. 

Ulrik wrote the book on pricing — literally. It’s called The Pricing Roadmap, and he runs a specialist advisory called Willingness to Pay. We brought him to Beelieve ’26 in San Francisco to talk about how companies break their pricing and what they should do about it. His core argument is simple and uncomfortable: many companies think they have good pricing because they’ve never experienced what actually good pricing feels like. 

As soon as Ulrik finished on the Beelieve stage, I sat down with him to unpack the ideas behind his session. Here’s what came out of that conversation (we also turned it into an on-demand webinar for you).

The money is in the structure (not the number)

Most companies treat pricing as the final step of product development: build the product, launch it, and then pick a price. 

Ulrik sees it completely differently.

“Pricing is only very far down the process. It’s not a number. It’s a structure. It’s a structure that articulates the value you deliver, to whom, and it’s a structure your organization has to align around. The money is in the structure.”

That leads to one of his simplest (and most useful) pricing rules:

“Don’t price the product. Price the customer.” 

That distinction matters, he said, because the same product can deliver wildly different value to different customers. Most SaaS companies price somewhere in the middle of their value distribution — they pick a number that works for the average customer and ignore the fact that some customers extract ten times that value. In traditional software, that was a defensible choice. Delivery costs were near zero, so leaving money on the table with high-value customers was annoying but survivable.

AI changed that equation.

The double hockey stick problem

Ulrik named the structural problem that’s forcing every AI company to rethink pricing from scratch “the double hockey stick”.

SaaS companies could historically price along the flat end of that curve. Once the product was built, adding a customer cost almost nothing — no incremental compute, no variable infrastructure. Economies of scale meant pricing could stay flat without compressing margins.

With AI, compute costs are real, and specific to each and every user. The customer getting ten times the value is also generating ten times the cost. If you still have a flat pricing model, you now have two hockey sticks working against each other: the value hockey stick, where a few customers extract most of the value, and the cost hockey stick, where those same customers are pulling real money out of your P&L.

As Ulrik shared:

“We have a client right now whose most loss-making customer is costing them close to 400 times as much as they take in revenue from that customer.”

The result is companies scrambling to answer two questions simultaneously: can we reduce the cost to build and deliver our product, and can we price it in a way that captures the value? Most are trying to answer both at once, under time pressure, without a clear framework.

One of the more interesting counterpoints to this argument at Beelieve San Francisco came from Kunal Agarwal, CFO of Gorgias. In his keynote, he laid out his framework for operationalizing usage-based pricing for the company’s AI agent and noted that his team originally tried to do exactly what Ulrik suggested: price per customer, based on the value their product delivered. They found that early customers pushed back, though — since their customers were new to adopting AI, they wanted as much pricing predictability and as little complexity or nuance as possible. While this anecdote appears at first glance to contradict Ulrik’s point, it actually underscores his broader statement that pricing is a system, not a single destination.

Which pricing model is right?

Usage-based, outcome-based, credit-based, hybrid — the market has cycled through pricing models fast enough that founders are starting to treat the choice as a strategic crisis. Ulrik’s take is refreshingly unsatisfying: 

“All of the pricing models are right for someone. Which spice is best? Curry? Thai basil? It depends what you’re making.”

His point is that pricing models are not universally good or bad. They work or fail in specific contexts, for specific customer profiles and buying behaviors. The companies that become evangelists for usage-based pricing are usually the ones for whom it happened to unlock growth — and that experience makes them overconfident that it will work for everyone else.

Credit systems are having their moment now, and Ulrik thinks it is deserved for a specific reason: they give AI companies a workable structure when the value they deliver is still being defined. Credits solve several AI-specific problems at once: they allow for complex product structures, capture prepaid cash upfront, and price based on usage without requiring the company to have fully mapped every value outcome. For AI products that can do many things and are still discovering which ones customers will pay most for, that flexibility is genuinely useful.

Read the Definitive Guide to Prepaid Credits →

The hard part isn’t picking a model. It’s building a framework to evaluate which model fits your product, your customers, and your go-to-market motion, and then having the discipline to test it and change it if it doesn’t work.

Rapid pricing changes aren’t a sign of failure

Many founders worry that changing pricing too often signals weakness. Ulrik argued the opposite:

“Between zero and roughly three to five million ARR, you’re in what we call the ‘hustle phase’. You basically sell anything in any way customers want to buy it. Then you figure out what works.” What’s more surprising is that he believes companies like Salesforce and OpenAI, which have made ten or more major pricing changes in the past twelve months, are behaving in exactly the same way. His read is that they’re treating pricing the way an early-stage company should: as a discovery process. They have a new product category; they don’t yet know what the right model is, and they have the resources to absorb the cost of being wrong and trying again.

“The winner is going to be the one who finds the right model. Think deeply. Test it. If it doesn’t work, try again.”

Every function can break pricing

Every pricing model has a failure point. In our conversation, Ulrik mapped exactly where those failures live.

He framed the org chart as a value stream: 

• Product is the value architect: the function that designs what value is delivered, to whom, and how it is packaged and priced.
• Sales is the builder: the function that executes that blueprint in the market.
• Finance is the structural engineer: the function that ensures the numbers hold, the billing infrastructure supports the model, and margin integrity is maintained.

Each function can independently break an otherwise sound pricing model.

• Product breaks it by building poor packaging — a pricing architecture so misaligned with actual customer value that no amount of sales skill can save it. 
• Sales breaks it by making so many bespoke exceptions and discounts that the model that actually gets executed in contracts bears no resemblance to the one that was designed.
• Finance breaks it by demanding that every feature, every customer, and every interaction be individually profitable — which creates complexity, forces add-on proliferation, and generates exactly the wrong incentives upstream.

The failure mode Ulrik sees most often isn’t that one function is incompetent. It’s that the three functions are not coordinating. Each makes locally rational decisions that collectively produce a broken pricing model.

The fix isn’t complicated, though it is hard: get the senior product, sales, and finance leaders in a room, agree on what value you deliver and to whom, and lock in a shared model before each function goes off to execute it independently. The coordination is what makes the architecture hold.

The oxygen mask principle

Ulrik closed with a reframe that shifts how his clients think about the purpose of pricing.

Most founders treat pricing as something that happens between them and their customer — a negotiation, a positioning decision, a number on a page. Ulrik’s view is that it’s also a resource allocation decision that directly determines how much you can invest in delivering value.

Price at $100, and you have $100 to spend on product, sales, and margin. Price at $200, and you have doubled the budget for everything. The companies that learn to charge for the value they deliver are also the ones that can afford to build the product, hire the team, and invest in the customer experience that justifies the price. It’s a compounding loop.

“Pricing becomes a perpetual game. If you manage to charge a lot of money, that’s what enables you to give a lot of value back.

“It’s the oxygen mask principle. Put it on yourself first so you can breathe, and then you can help someone else. That someone else is your customer.”

Ulrik covered even more in his Beelieve session, including his framework for introducing AI pricing into mature SaaS businesses, his “Petri dish” approach to pricing experimentation, and how philosophy unexpectedly shaped his approach to commercial strategy. Watch his session here.

Beelieve 2026 brought together 500+ leaders shaping the future of AI-ready companies and business models in San Francisco.  All sessions are available as on-demand replays. Watch here.