SaaS Says: Your Best Customers Cost the Most

Your Best Customers Cost the Most

Ed Barrow has a warning for any software company with engaged users. "And you often find now that actually, your most loyal customers who actively engage in your product more and more are costing you a lot more."

Barrow runs Cloud Capital, which finances cloud and AI contracts, so rising bills suit him. His company's own survey of about 100 growth-stage CFOs found the average software company now spends 15 to 20% of revenue on compute and AI. Among AI-native companies: "So over half of AI native companies are now spending more than 30% of revenue on AI and compute. That is not the software business of the last decade." He calls the loop of more usage, more cost "the token treadmill".

Todd McElhatton hears it from peers. The Zuora CFO describes a product "that had, you know, a nice 80, 85% gross margin," then AI features get added, "and now all of a sudden, you know, I've tanked my margins." His fix is to hold the line on price until the product earns more. "They've earned the right to have that conversation about having incremental pricing."

Inside Zuora, AI is already a budget line next to people. "And that's one of the things we move to in the upcoming budget cycle is you'll have a people budget, and you'll have a token budget." "But if a leader tells me they can do, you know, they can deliver better with more tokens than people, we'll do more tokens." On the buying side, the seat is going. "a lot of our original AI tools were seat-based and everyone is moving away from those."

At Higgsfield, founder Alex Mashrabov says margin depends on whose model answers. "The margin on own models and open weights models is over 80%. And then it almost doesn't matter. And for close source models, it's probably between 20 and 30%." So the company decides. "So we choose what model to use in over than 40% cases."

Ken Stillwell has the argument from three weeks ago, with a margin reason attached. Pega, where he is CFO, refuses to bill clients by the token, which lets it pick the cheaper model or no model at all. Metering tokens, he says, is like seeing "which household in the neighborhood can use the most electricity in the summer." His forecast, from a company that sells workflow software: "I think clients will expect to pay based on the work that it's being done." The a16z growth team made the same case from the investor side: charge for the finished job, and route to cheaper models underneath it.

LiveRamp CFO Lauren Dillard rebuilt pricing around a "token" that has nothing to do with AI, a unit drawn down as customers act on their data. Same gap. "First, there was a increasing disconnect between the value customers were receiving and the cost to deliver that value and ultimately how we were monetizing our platform, which we knew would only be exacerbated as kind of these AI workflows took hold."

Three weeks ago vendors refused to sell the token. Now their CFOs are counting what it costs to give it away.

Sources: Interviews from GrowCFO Show (Sep 29, Ed Barrow of Cloud Capital), Run the Numbers (Oct 1, Todd McElhatton of Zuora; and Sep 28, Lauren Dillard of LiveRamp), 20VC (Sep 28, Alex Mashrabov of Higgsfield), CFO Thought Leader (Sep 30, Ken Stillwell of Pegasystems), and a16z (Sep 30).