The Trap Disguised as a Gift
Usage-based pricing is a growth story that only works once you've already won.
There is no pricing model in software that sounds more fair, more democratic, more good than usage-based pricing. You use more, you pay more. You use less, you pay less. It aligns incentives! It removes barriers! It’s the pricing equivalent of a Whole Foods reusable bag — virtuous on the surface, slightly more expensive than it needed to be, and ultimately a branding decision.
The companies that made usage-based pricing famous — Twilio, Stripe, AWS — are now held up as proof it works. And it does work. For them. The problem is the lesson most startups draw from this is the wrong one. They see the trophy and copy it, not realizing the trophy came from a completely different sport.
Usage-based pricing is a growth story that only works once you’ve already won. For most startups, it’s not a pricing model. It’s a slow bleed with good PR.
The Survivors Bias is Enormous
Let’s actually look at who made usage-based pricing famous. Twilio charges per text message or API call. Stripe charges per successful transaction. AWS charges per compute second. Notice something? These are infrastructure plays. The unit cost is essentially constant and predictable — sending one million texts costs Twilio roughly the same per text as sending a hundred. Their marginal cost curve is flat. Their pricing axis is bulletproof.
Twilio grew from $277 million in revenue in 2016 to over $3.8 billion in 2022 on this model. But here’s what people forget: the product was mission-critical and nearly impossible to throttle without degrading the core service. If you want your text messages delivered, you use Twilio. There’s no “use fewer messages to save money” option that doesn’t also mean “your product works worse.” That’s not a pricing model. That’s a structural lock-in with flexible billing attached.
Application-layer software — your CRM integration, your analytics tool, your AI-powered workflow product — doesn’t work that way. Users absolutely can throttle their usage without reducing core value. And when a CFO sees an unpredictable line item creeping up, they will.
A 2023 KeyBanc Capital Markets survey found that 68% of SaaS executives cite forecasting difficulties as their primary concern with pure usage-based models. That’s not a minority opinion. That’s a supermajority of the people who have actually tried to run finance inside a company using this model, telling you it’s hard.
Salesforce Just Ran This Experiment So You Don’t Have To
The cleanest case study on why pure usage-based pricing fails isn’t some obscure Series B startup. It’s Salesforce, a company with 150,000 customers, a decade of enterprise sales infrastructure, and Marc Benioff’s entire personality on the line.
When Salesforce launched Agentforce in late 2024, they priced it at $2 per conversation. Clean. Simple. Democratic. Within months, the backlash was severe. Customers couldn’t define what a “conversation” even was when a single query triggered eight backend processes. One support team leader calculated that five agents handling 70 conversations per day would cost roughly $900 daily. Budget teams panicked. Procurement stalled. Of the first 5,000 Agentforce deals, only 3,000 were paid.
Salesforce pivoted to Flex Credits in May 2025 — $0.10 per action. More granular. But as one product officer put it, it still “felt like a black box — hard to predict, hard to explain to stakeholders.” By late 2025, Salesforce introduced per-user licenses at $125 per month, the model they’d spent two years trying to move away from. CFOs got a number they could budget. The chaos subsided.
Salesforce changed its pricing model three times in eighteen months. If the company with the deepest enterprise sales relationships in the world couldn’t make pure usage-based pricing stick for AI, you should have a very good reason why you think your seed-stage company can.
The Revenue EKG Problem
Here is what usage-based pricing does to a startup’s financials: it makes your revenue look like a cardiac monitor. Customers spike in January, go quiet in February, have a big project in March, do nothing in Q2.
This is a problem for three distinct people, and all three of them matter.
The first is your VC. Investors price predictability. Annual Recurring Revenue, the metric that made SaaS companies worth 20x revenue, works because it’s recurring — you can underwrite it. Usage-based revenue is closer to a service business than a software business, and service businesses get service multiples. Hybrid pricing models report the highest median growth rates at 21%, according to Bain, outperforming both pure subscription and pure usage. But the market punishes companies that can’t tell a clean ARR story.
The second is your sales team. Commission structures for usage-based models are genuinely difficult. Do you pay on first contract? On annualized run-rate? On actual consumption over 90 days? Each option creates different incentives, and most of them are wrong. You end up with reps who close deals and immediately disengage, because there’s no comp mechanism that rewards ongoing usage growth.
The third is your customer’s CFO. A 2025 Salesforce study found that 90% of CIOs believe managing AI costs limits their ability to maximize value. The reaction isn’t to use more and spend more. The reaction is to set caps. To throttle. To approve a small pilot and refuse to expand it until the line item becomes predictable. This is not customer success. This is customer sandbagging, and it destroys your expansion revenue.
The Cost Variance Problem Nobody Talks About
There’s a dimension specific to AI-powered products that makes usage-based pricing even harder than it was for Twilio in 2010: your own costs aren’t flat either.
Traditional software had 80%+ gross margins because the marginal cost of serving an additional user was essentially zero. Build it once, distribute forever. In AI-powered products, every action has a compute cost, and that cost varies wildly based on complexity. One customer service platform found that simple queries cost them $0.04 to serve while complex ones ran to $2.80. If you price at average, you either lose money on heavy users or gouge light ones. Neither is a business.
65% of SaaS vendors are now layering usage-based components on top of seat pricing to manage exactly this problem, according to Bain. The market is already voting. Pure usage-based is retreating to hybrid. The companies loudly proclaiming they’ve “moved beyond the seat” are mostly companies that haven’t yet had to file a quarterly earnings call.
When It Actually Works
This isn’t an argument that usage-based pricing is wrong in all situations. It’s an argument that it’s misapplied in most.
It works when your cost per unit is genuinely flat and predictable. Infrastructure, communications, transactions. It works when your buyers are technical and comfortable reasoning about variable costs — developers are fine with it, procurement committees are not. It works when you’re large enough that the variance in any individual customer’s usage averages out across a portfolio. A thousand customers’ EKGs, combined, start to look like a flat line.
It works, most of all, when your product is genuinely impossible to throttle without degrading value. That’s the Twilio test. Apply it before you decide your startup passes.
Usage-based pricing is not a business model innovation. It’s a go-to-market tactic borrowed from infrastructure companies and reapplied, mostly unsuccessfully, to application-layer software that doesn’t share the same cost structure, buyer profile, or lock-in mechanics.
Twilio and Stripe weren’t startups when usage-based pricing made them famous. They were already large, already entrenched, already operating at a scale where variance smooths out. The lesson from their success isn’t “usage-based pricing is good.” It’s “infrastructure with commodity unit economics and irreplaceable switching costs can support usage-based pricing.”
For the vast majority of startups — the AI-native product, the vertical workflow tool, the enterprise automation play — the model sounds like freedom and operates like a trap. You’re handing your customers a dial to turn down your revenue, giving your VCs a forecast they can’t underwrite, and building a sales motion that rewards closing over expanding.
Democratic pricing is a beautiful idea. It just turns out that in software, the market keeps voting for certainty.
Thanks for reading. If you found this useful, the subscription button is right there and it costs a flat, predictable, non-usage-based amount per month.