Durable monetization: How four AI founders solved the pricing puzzle

Ask four successful AI founders how they priced their products, and you'll get four different answers. Not because some got it right and others got it wrong, but because there's no single winning formula for pricing AI products. Pricing reflects each company's customer value, defensibility, delivery economics, and proof of impact. Over the past year, AI companies have experimented with usage-, workflow-, outcome-, and hybrid pricing models (see The AI Pricing and Monetization Playbook for a detailed breakdown). Increasingly, businesses gain the most revenue momentum by pricing on outcomes, but the cost-benefit tradeoff depends entirely on the ideal customers they serve.

A bespoke approach is often the right entry point for devising your AI-native monetization strategy. So the fundamental question to ask is, “How does a business manufacture pricing power given its unique variables (industry, ICP, GTM motion)?

In our latest pricing installment in the Building AI Differently series, we share new observations on the core drivers of pricing power, then look inside four portfolio companies—Recall.ai, Strella, Graph AI, and Ada—to see how those principles played out in practice.

Key takeaways on building durable AI pricing models

1. Durable AI pricing ties revenue to something a customer can independently verify, not to a claim. Recall.ai charges by infrastructure consumed, Graph AI by workflow automated, Ada by conversations purchased. Each model gives the customer a number they can check against their own dashboard, which is what survives a renewal conversation.
2. Hard ROI commands premium pricing; soft ROI compresses it. A copilot that surfaces the right answer still leaves the outcome in the customer's hands, so the value is real but unprovable. An agentic product that closes the loop end-to-end removes the question, "Would this have happened anyway?" Proof of impact is what earns pricing power.
3. When AI costs are low relative to the labor being replaced, founders gain pricing flexibility, not pricing pressure. Strella's AI moderation costs "pale in comparison" to paying a human research participant, according to CEO Lydia Hylton. Because AI wasn't the dominant cost driver, Strella optimized for adoption with seat-based pricing instead of protecting margin with usage fees.
4. Price against the labor category you're replacing, not the software category you're entering. Graph AI benchmarks pricing against the combined cost of software and human review, not against other pharmacovigilance platforms, since its product automates 75% of a workflow that was previously billed by headcount. The unit of pricing follows the unit of value created, not the technology sold.
5. The right pricing model matches how customers want to buy, even when a different model captures more value. Ada started with pure outcome-based pricing per resolution, then shifted to a hybrid model once enterprise buyers wanted predictable annual budgets over per-conversation billing. The underlying value story didn't change, but the packaging had to.

Three drivers of pricing power for AI products

Adam Fisher, Partner at Bessemer, explains that solving the AI pricing puzzle ultimately comes down to three variables:

1. Customer value

The most fundamental question in pricing is simple: How much value are you creating for your customer?

This can be measured in a variety of ways, including:

• Increased productivity
• Cost savings
• Revenue uplift
• ROI metrics

2. Fungibility

The next factor is how easily your product can be replaced, or in other words, how defensible it is. The more differentiated your offering—whether through proprietary data, domain expertise, model performance, workflow design, or deep integration into customer systems—the greater your pricing power.

If customers have few viable alternatives, or replacing your product with human labor isn't practical, you can generally command higher prices.

3. Delivery economics (i.e. cost of goods sold or COGS)

Unlike traditional SaaS, where the cost of serving one more customer is virtually zero, AI products incur real costs with every query. Those costs vary widely. Some AI companies face significant training and inference costs that scale with usage. Others rely heavily on human services, implementation, or domain experts to deliver their product. Understanding your delivery economics is essential to designing a pricing model that can scale profitably.

The power of monetizing proof

While every AI pricing decision involves the same three variables, their interaction determines not just how much you can charge, but how you should charge. The variable behind how much pricing power a company commands is proof of the customer value created.

Two products can create identical customer value and still command entirely different prices if one can prove it closed the loop and the other can only claim it helped.

When we revisit the AI Value Framework (first shared in our AI pricing playbook), we see this dynamic play out.

The AI Value Framework as an ROI guide

For example, a copilot that surfaces the right answer still leaves the outcome in the customer's hands: The value is real, but it's soft, and soft ROI compresses willingness to pay because customers wonder, "Would this outcome have happened anyway?" An agentic product that completes the workflow end-to-end removes that question. The result is measurable, attributable, and much harder to negotiate down.

Most companies spent initial purchase cycles in adoption mode where teams bought on potential with minimal price scrutiny. As contracts hit renewal, buyers are re-underwriting them against delivered outcomes, not original promise. Products built on hard ROI, such as Sweep's workflow completion and Sett.ai's closed-loop delivery, renew on the strength of their own evidence.

Products still living in soft ROI territory, however sophisticated the underlying model, will face the harder conversation: prove the value or reprice accordingly.

However, service replacement plays by a different rule entirely. It isn't sold on revenue uplift or efficiency framing, but rather total cost of ownership. The companies that win here are the ones that force an honest total cost ownership (TCO) comparison rather than letting the customer compare sticker price to sticker price.

Enterprises routinely undercount what their legacy approach actually costs; founders who can make that gap visible are the ones who capture the replacement budget. Recall.ai, Strella, Graph AI, and Ada made different choices about usage, workflow, outcome, and hybrid pricing. But each was solving the same underlying problem: how to make their value creation legible enough, and their delivery economics durable enough, to price with confidence rather than guesswork.

How leaders from four AI companies landed on the right pricing model for their business

Here’s an inside look at how each company arrived at its pricing model and the key lesson AI founders can apply to their own businesses.

1. Recall.ai: Align your pricing with both customer success and your own economics

Recall.ai built a usage-based pricing model that naturally aligns customer value with delivery economics, charging customers in proportion to both the value they receive and the infrastructure costs the company incurs.

For Recall.ai, pricing was a natural extension of the business it had built. Recall.ai provides the infrastructure layer that powers AI products built on conversation data. Customers use the platform to capture audio, video, transcripts, and metadata from meetings, which means the more successful their products become, the more conversation data they process.

That growth benefits both the customer and Recall.ai. As co-founder and CEO David Gu explains, “What we say to our customers is if your product becomes very successful and you’re now processing a lot of data, Recall.ai is going to cost more. But if you’re building internally, testing with your first pilot group of customers, and there’s not a lot of volume, then Recall.ai will be less expensive.” In other words, pricing scales alongside customer success.

This approach also reflects Recall.ai's delivery economics. Since infrastructure providers incur real costs as customers consume more resources, a usage-based model ensures those costs scale alongside revenue, creating what David describes as an "alignment of success" between Recall.ai and its customers.

That alignment extends beyond economics. Because customers already purchase cloud infrastructure, APIs, and AI models on a consumption basis, usage-based pricing feels familiar. It avoids the friction of seat-based licensing, eliminates the need for complex pricing tiers, and allows customers to predict costs based on expected volume while giving Recall.ai room to grow as customers expand.Recall.ai neatly illustrates the interaction between the three drivers of AI pricing power: Customer value scales with every hour of conversation data customers process.

Delivery economics scale alongside it, as Recall.ai incurs additional infrastructure costs with higher usage. And because the company has built specialized infrastructure that developers can't easily replicate, it has the defensibility to monetize that value through a consumption-based model.

For a deeper look at how Recall.ai turned one of the most painful infrastructure challenges in AI into the foundation for thousands of AI products, read Recall.ai: Building the infrastructure behind AI tools and products for video conference recording.

2. Strella: Remove barriers to adoption before optimizing for revenue

Strella chose seat-based pricing to remove barriers to adoption, demonstrating that when customer value is high and delivery economics are favorable, founders have the flexibility to optimize for adoption rather than maximize revenue per use.

Founders often think about pricing in terms of maximizing revenue. But the team at Strella approached it differently, asking not how to capture the most value, but instead what pricing model would drive the most product adoption. The AI-powered customer research platform automates qualitative interviews end-to-end, from moderating conversations to synthesizing findings into usable insights.

While charging per interview or per research project might have captured more value in the short term, Strella deliberately chose a seat-based pricing model. "We wanted to remove every barrier to running a Strella interview that you possibly could have," explains CEO Lydia Hylton. 

Charging per interview or project would require teams to seek budget approval every time they wanted to conduct research, which would slow down the process and potentially deter customers from using the platform. A seat-based model eliminated that friction, allowing customers to negotiate spend once, then run research as often as they wanted. 

The reason this pricing model works comes back to the interaction between customer value, fungibility, and delivery economics. Strella's customer value is clear: Projects that once took weeks can now be completed in days at a fraction of the cost. At the same time, access to frontier models is increasingly commoditized. Strella's differentiation lies in its research-specific product and workflow, not the underlying model. Its AI moderator is designed to conduct natural, adaptive interviews and solve research-specific challenges that general-purpose models don't.

Critically, Strella's delivery economics gave the team flexibility. Unlike many AI companies facing significant inference costs, Lydia shares that the team’s "cost per interview pales in comparison to the cost of paying the person for their time who's participating in the interview." Because AI wasn't the dominant cost driver, the company wasn't forced into a usage-based pricing model to protect its margins.

Ultimately, Strella optimized pricing for long-term growth rather than short-term value capture. Predictable revenue was also a welcome byproduct of the pricing model. As Lydia puts it, "Predictable revenue is worth more than unpredictable revenue."

For a detailed case study on how Strella built and launched an AI-powered qualitative research platform, read Strella: Transforming qualitative research from a bottleneck into an AI superpower.

3. Graph AI: Price against the labor you're replacing, not the software category you're entering

Graph AI shows how companies with massive, measurable customer value and strong domain-specific defensibility can price against the labor they replace rather than traditional software benchmarks.

When Graph AI entered the pharmacovigilance market, pricing was largely driven by headcount. Software vendors charged for their platforms, while the substantial human effort required to complete the workflow was billed separately. Graph AI saw an opportunity to fundamentally change that equation.

"A lot of these services are delivered by people and less by software," says CEO Raghavendra Parvataraju. "Today, customers are forced into a model where 75% of their spend goes toward human labor and only 25% toward the software that's supposed to help them. We're inverting that ratio. We've built a platform where 75% of the heavy lifting is handled by software, and only 25% requires human oversight."

That inversion reflects the shape of the labor being replaced. Legacy pharmacovigilance labor sits at two extremes: a large, high-volume front end (e.g., sourcing, intake, data entry, triage), staffed by pharmacology and BPO resources at relatively low cost per head, and a much smaller back end of quality and medical reviewers whose expertise carries a far higher cost per hour. 

Graph AI automates the high-volume front end outright, while bringing AI-assisted analysis to the reviewers themselves, spending their expertise on judgment rather than preparation. The result is software-grade gross margins that widen as more of the workflow is automated. Because Graph AI front-loads its heaviest processing early in a case's lifecycle, incremental cost falls as the case progresses. This means that the platform becomes more profitable per case at volume, rather than carrying flat costs throughout.

Graph AI uses an outcome-based model. Because the customer value it delivers extends beyond software alone, the company's pricing is benchmarked against the combined cost of the software and labor it replaces. In other words, it prices the value it creates, not simply the technology it sells.

Rather than anchoring prices to software seats or licenses, Graph AI prices according to how much of the workflow it automates. As the product automates more of the pharmacovigilance workflow—from processing individual safety cases to generating reports and eventually handling larger portions of the review process—each new capability creates additional customer value and justifies the next pricing tier. Put simply, the unit of pricing follows the unit of value.

That staged approach also creates a pricing model that evolves alongside the product itself. Each new module automates another layer of human work, so every pricing tier reflects a new category of value delivered. 

The model also offers customers predictability. Because pharmacovigilance volumes are relatively stable and forecastable, per-case pricing is easier to budget for than seat-based licensing. By tying revenue directly to the work the platform processes, Graph AI aligns its commercial model with both the value it creates and the way customers plan their spending.

For an in-depth case study on how Graph AI transformed a services business into an AI-native platform for life sciences, read Graph AI: A service firm turned AI-native solution for pharma and life sciences.