April 3, 2026
If you've built AI features into your product, you already know the feeling. You look at your cost of goods and something doesn't add up anymore.
Traditional SaaS was a beautiful business model. You built the software once, hosted it on AWS or wherever, and the marginal cost of serving another customer was close to zero. Customer success, some hosting fees, a few odds and ends. Gross margins sat comfortably at 80 to 90% for mature SaaS companies.
AI changed that equation overnight. Every model call costs real money. Every token, every inference, every agent action spins the meter. Bessemer Venture Partners' data shows AI-first companies operating at 50 to 60% gross margins, and some early-stage companies dipping well below that.
Take a product that charges $100 a month. It's not unusual for $30 to $40 of that to go toward token charges and compute costs. That's a 30-point margin hit compared to what the same product would have looked like three years ago.
A lot of startups are masking this problem right now. They burn through AWS credits, then hop to GCP and use Gemini credits, then jump to Azure. It's cloud credit arbitrage, and it works until it doesn't.
The credits will run out. When they do, you have to reckon with your actual unit economics. And if your pricing model was designed for a world where serving customers was essentially free, you're going to be staring at a very uncomfortable spreadsheet.
This isn't a hypothetical. Replit saw gross margins dip to negative territory during usage surges before pricing changes brought them back to the 20 to 30% range. GitHub Copilot was reportedly losing up to $80 per user per month for heavy users in its early days while charging a flat $10. These are cautionary tales, not edge cases.
The first move most companies make is shifting from flat-rate SaaS to usage-based pricing. It makes sense: if your costs scale with usage, your pricing should too.
Usage-based models don't magically fix your margin. You're still paying for inference. You're still running at lower gross margins than pre-AI SaaS. But what usage-based pricing does is eliminate the tail risk. In a flat-rate model, a power user who hammers your AI features can generate a loss on their account. In extreme cases, the more they use your product, the more money you lose.
Usage-based pricing puts a floor under that. Heavy users pay more. Light users pay less. The risk of a single account blowing through your margins disappears.
This is why 92% of AI software companies now use some form of mixed pricing model, combining subscriptions with usage components. Pure flat-rate pricing is becoming an endangered species in AI-native products.
Usage-based pricing is the defensive play. It protects your margins. Outcome-based pricing is the offensive play. It expands your revenue.
For the first time, software products can do a complete job. Not "help a human do the job." Not "make the job 20% faster." Actually do the thing the customer is paying for.
Voice AI is the clearest example. A home services company, say a plumber or an electrician, needs someone to answer phones, qualify leads, and schedule appointments. A full-time employee costs $15 to $25 an hour. That's roughly $40,000 a year.
ZyraTalk, a client of ours, built voice AI that does exactly this for home service companies. The AI answers the call, qualifies the lead, schedules the appointment. The whole job, 24/7. EverCommerce acquired them in September 2025, integrating ZyraTalk's platform into their EverPro ecosystem of 350,000+ home and field service providers. That acquisition tells you where the market is heading.
We see the same pattern across industries. In legal tech, AI listens to intake calls and decides whether they're viable cases. In travel, booking agents plan entire trips. In our own world, agentic project management is moving toward scoping, estimating, and planning software projects end to end.
Here's why outcome-based pricing is so compelling for AI companies, and why the smart ones are sprinting toward it.
Software gets priced like software. A few hundred dollars a month. Maybe a thousand for enterprise. That's the mental anchor buyers have for SaaS tools.
Outcomes get priced against the alternative. If you're replacing a $40,000-a-year employee, even capturing a fraction of that value means thousands of dollars per year in revenue per customer, not hundreds. That's an order-of-magnitude jump in what you can charge.
The math is straightforward. If your AI can deliver an outcome that used to cost the customer $40,000 annually, and you charge $5,000 for it, you're offering a massive ROI while capturing 10x more revenue than a typical SaaS subscription. The trajectory is clear: Gartner projects that by 2030, at least 40% of enterprise SaaS spend will shift toward usage-, agent-, or outcome-based pricing. Companies like Zendesk ($1.50 to $2.00 per automated resolution), Intercom ($0.99 per AI-resolved ticket), and Salesforce (per-action pricing for Agentforce) have already made the move.
This isn't as clean as it sounds. The state of the art in most AI use cases delivers maybe 50 to 60% of the full job. Some simple, well-defined outcomes hit 95 to 100%. But many are still in that messy middle where the AI does most of the work and a human handles the rest.
That's fine. It just means your pricing model needs to reflect the reality. Hybrid approaches, a base subscription plus outcome-based fees, are emerging as the practical middle ground for companies that aren't yet at full automation. Bessemer's playbook recommends hybrid models as the default when you're uncertain, because they give customers predictability while letting vendors capture upside as usage scales.
The companies that wait until their AI can do 100% of the job before moving to outcome-based pricing will miss the window. The ones that start experimenting now, even with partial automation, will have the billing infrastructure, the customer relationships, and the performance data to scale outcome pricing as their AI improves.
If you're building AI products, your pricing model isn't just a business decision. It's a survival decision. Flat-rate SaaS pricing on AI-heavy features is a race to negative margins. Usage-based pricing keeps you alive. Outcome-based pricing is where the real upside lives.
If you're buying AI products, pay attention to how vendors price. Per-seat pricing on an AI tool that replaces headcount is a signal that the vendor hasn't thought through their model yet. Outcome-based pricing, where you pay for results, means the vendor's incentives are aligned with yours.
The pricing spectrum is shifting from left to right: fixed subscriptions, to usage-based, to outcome-based. AI is the force pushing it along. The companies that figure out how to price on outcomes, even imperfectly, will capture dramatically more value than those still clinging to the SaaS playbook.
Describe your software or AI project. Get a full scope with story-point pricing, sprint estimates, and a downloadable plan in minutes. No calls, no waiting.
Free and instant. Try the calculator now.
Why are AI features compressing SaaS margins so much?
Traditional SaaS has near-zero marginal cost per user. Once the software is built, serving another customer barely moves the needle on infrastructure costs. AI flips that: every query, every inference, every agent action consumes real compute resources. Token costs, GPU time, and API fees add up with every user interaction. That's why companies that were used to 80 to 90% gross margins are now seeing 50 to 60%, or worse.
Doesn't outcome-based pricing create unpredictable revenue for the vendor?
It can, which is why most companies are adopting hybrid models rather than going pure outcome-based. A base subscription covers fixed costs and provides revenue predictability, while the outcome-based component captures upside when the AI delivers measurable results. This blended approach gives vendors a floor while still aligning their incentives with customer value.
What qualifies as a measurable "outcome" for pricing purposes?
The best outcome metrics are ones the customer already tracks: support tickets resolved without human intervention, appointments booked, leads qualified, cases evaluated. The key is that the AI can complete the task end to end and the result is objectively verifiable. Fuzzy outcomes like "improved productivity" are harder to price against and typically better suited for usage-based models.
Won't falling inference costs eventually solve the margin problem on their own?
Inference costs are declining, but total AI spend tends to rise because companies deploy AI to more use cases as it gets cheaper. This is sometimes called Jevons Paradox applied to AI: per-token costs drop, but token consumption skyrockets. Pricing strategy still matters even as compute gets cheaper, because the volume of AI work your product does will likely grow faster than the cost savings.
How do I transition from flat-rate pricing to outcome-based pricing without losing customers?
Start with a hybrid model. Keep the base subscription in place and add an outcome-based component on top, priced as a clear win for the customer relative to the alternative cost. Track and publish the outcomes your AI delivers so customers can see the ROI. Over time, as confidence builds on both sides, you can shift more of the pricing toward outcomes and less toward the flat base.
Related: How Much Does It Cost to Build an App? · Story Points Explained
