From a recent live webinar conversation 

The cost structure of modern software is shifting far faster than the commercial models underneath it. This conversation captures the ways we can understand how engineering, finance, and go-to-market teams can create a resilient monetization strategy.

Below is a lightly edited A&A with Ville Lehto, VP of Strategy at Aiven, pulled directly from the live discussion during a recent webinar. It covers who should own pricing at early-stage companies, how to handle the reseller ecosystem, and how to communicate spiky bills to customers.

Q: How should early-stage companies think about building a pricing function from scratch?

At a company’s earliest stages, pricing shouldn't be outsourced to a consultant or delegated away from the core team. Because AI usage and model costs are so fluid, pricing decisions require deep product intuition and the authority to make fast, instinct-driven pivots.

"The first thing I would say is one of the founders needs to build this expertise. If you aren't an expert on this, then it's going to be really hard to build a great product." —Scott Woody, Head of Product, Stripe Revenue Suite
‍Hear Scott’s live answer

In an established, static market, hiring an outside firm works well. But in the AI space, the number one cost driver—the models themselves—is incredibly dynamic. Founders are best positioned to align those shifting costs with how users perceive value.

"What you need at that stage is founder instincts, and then you need iteration speed. You won’t get either of those if you outsource this, by any means."
—Ville Lehto, VP of Strategy, Aiven
Hear Ville’s live answer

Both Scott and Ville emphasized that pricing must be treated as a computational, iterative process rather than a “set-it-and-forget-it” milestone.

Q: How are cloud marketplaces evolving as infrastructure ships directly into customer tenants?

Cloud marketplaces already drive a massive share of revenue for infrastructure-as-a-service vendors, and that trend is only going to accelerate in the AI era.

Scott highlighted that emerging AI platforms (like vibe-coding tools) are beginning to act as demand aggregators. They hold large customer budgets and are allowing those budgets to be spent across third-party applications listed in their own ecosystems.

"I expect that the reseller marketplace is actually going to become a pretty big trend for these companies selling AI services."
—Scott Woody
Hear Scott’s live answer

Because autonomous agents won't be loyal to a single cloud provider the way human buyers often are, they will seek out and consume compute wherever budgets exist.

Q: How do you explain a usage bill that changes every month to customers—or even to your own board?

When bills are highly volatile, the solution isn't better customer support. It's better product design. If a customer is surprised by an invoice at the end of the month, the product has failed to properly communicate value and forecast spend along the way.

"I always approach this as a product problem... can you show the user before they run something on your platform that, hey, by the way, this will cost this much."
—Ville Lehto
Hear Ville’s live answer

Scott agreed, noting that pricing and cost telemetry need to be treated with the same level of care as user onboarding flows.

"If you're not also leading with value and making that value and cost and all of that stuff exposed to your user, you're building a product that is designed to give surprise bills at the end of the month."
—Scott Woody
Hear Scott’s live answer

The most successful companies invest heavily in tools that translate raw usage into understandable, value-driven metrics before the bill ever arrives.

Q: How should infrastructure companies price inference? Should it be a pass-through or a margin layer?

The answer depends on how a company positions its core value. If inference is just a raw utility to facilitate a downstream action, treating it as a cost of doing business makes sense.

"If you have outcome-based pricing and you have this specific action that you can quantify, then inference is probably not something that you should be just giving away."
—Ville Lehto
Hear Ville’s live answer

On the other hand, Scott warned against relying too heavily on pass-through pricing if you’re building distinct value on top of the raw models.

"I would encourage most people who are doing inference as part of their pricing and packaging to think of it more as a cost of doing business as opposed to the dominant term in the value equation for your users."
—Scott Woody
Hear Scott’s live answer

Pass-through pricing is easy for users to understand, but it fundamentally caps the amount of margin a company can extract. Ultimately, pricing must reflect the software's unique output, not just the raw cost of the compute underneath it.

📺 Watch the full webinar here