A founder I know signed their biggest enterprise deal last November. Custom pricing, outcome-based terms, everything modern. Two days later, their billing system couldn't generate the invoice. Not because the product failed. Because the infrastructure they built on was designed for a world where pricing meant counting the seats, multiplying by the number, and sending the PDF.
That's the problem this post is about.
I run Flexprice. We build billing infrastructure for exactly this problem. So read this with that in mind. But it also means I've seen every failure mode up close, at companies that were well-resourced and technically competent. That's why I'm writing this.
On February 3, 2026, the Nasdaq Cloud Index lost nearly $300 billion in 48 hours. No recession, no rate hike, no scandal. Just two product announcements. Anthropic shipped a suite of autonomous agents. OpenAI shipped Project Operator, software that could navigate any software interface without a human at the keyboard.
The market understood the implication before most founders did. If software doesn't need a user, you can't charge for a seat.
Most of the conversation around AI pricing right now is about features and tiers. That's the wrong conversation. The question isn't whether to put AI on your Pro plan. The question is whether the pricing model you've operated on for 20 years still maps to the value you're delivering. For a lot of companies right now, the honest answer is no.
We're not just changing pricing. We're changing what pricing means. For 25 years, pricing was a number you put on a checkout page. Now it's the architecture of how value flows between you and your customer. Get that right, and everything else follows. Get it wrong, and even great products die.
The story of how we got here spans three eras. Understanding all three is the only way to think clearly about where things are going.
The perpetual license
Before SaaS existed, enterprise software was sold like equipment. You paid Oracle or SAP a large upfront fee, received physical media or a server install, and owned that version. Revenue recognition happened at point of sale. Once your check cleared, the vendor's financial incentive to care about your success essentially ended.
There was an honesty to it. The artifact was the product. No pretense of an ongoing relationship. You bought a thing. You had the thing. Clean.
What it got wrong was pricing the artifact instead of the outcome. A company running Oracle every single day and a company that installed it and never opened it again paid the same amount. Oracle's revenue had no relationship whatsoever to whether their software created value. That's not a pricing model. It's a heist with better documentation.
SaaS vendors were quick to point this out when positioning the subscription model as more customer-friendly. And they were right, partially. What they were less quick to point out was that the subscription model also meant vendors got paid every year instead of once. Most of the benefits of SaaS for customers were real. They were also, conveniently, even better for vendors. File that under benefits we packaged as customer benefits.
The subscription lie
Salesforce launched in 1999 with "No Software" on its billboards and a per-user, per-month price. The promise was alignment. You pay as long as it's worth it, we stay motivated to make it worth it, everyone wins.
I want to be precise about what per-seat pricing actually aligned, because this is where the story gets interesting.
It aligned vendor revenue with customer headcount. Not with customer outcomes. Those are very different things.
A 500-seat customer getting enormous value from your product and a 500-seat customer who bought it, set it up, and forgot about it pay you the same amount. That is not alignment. That's billing by coincidence. The coincidence held for so long, because for most of the 2000s and 2010s, more employees really did mean more software usage, that people started calling the coincidence a principle. It wasn't. It was a correlation that held until it stopped holding.
We called it recurring revenue. What we actually built was recurring invoices. Those aren't the same thing, and the difference is catching up with the entire industry right now.
There's a second problem buried in the per-seat model that nobody talked about because it was commercially inconvenient to discuss. Per-seat pricing financially rewards your customers for staying inefficient. Every time they automate a workflow, eliminate a role, or get more productive per person, they have an argument for fewer seats. The vendor's implicit financial interest, the one nobody said out loud, was that their customers wouldn't get too much more efficient.
For most of SaaS history, efficiency gains were slow enough that this tension didn't matter.
Then AI started eliminating roles in months, not years.
The tension became impossible to hold.
AI broke it
One enterprise customer. Two hundred AI agents. Zero new seats. This isn't a hypothetical. This is what's happening right now at companies running agentic workflows on tools priced for human users. The agents are consuming real resources, generating real value, and the vendor's pricing model has no slot for them. Costs go up. Revenue stays flat.
On February 3 and 4, 2026, the Nasdaq Cloud Index lost nearly $300 billion in 48 hours. Anthropic shipped a suite of autonomous agents. OpenAI shipped Project Operator, which could navigate any software interface without a human at the keyboard. The market did the math immediately. If software doesn't need a user, per-seat pricing isn't just suboptimal. It's structurally incoherent.
Jefferies coined the term SaaSpocalypse. Roughly $2 trillion in market cap evaporated from software companies whose revenue models depended on seat counts that AI was making irrelevant. That phrase would be hyperbolic if the numbers weren't real.
The seat made sense when software was a tool humans picked up. Now software is the worker. You don't charge a factory per employee. You charge per unit of output.
What's actually happening now
If you're moving off per-seat right now, there's a good chance you're making the same mistake in a different costume. You're not rethinking where value lives. You're finding a new proxy. Usage-based pricing, badly implemented, is just per-seat pricing with a more complicated invoice. You're still not pricing outcomes. You're pricing tokens consumed, or API calls made, or compute seconds burned. Those are slightly more honest than seats. They're not the same as the value delivered.
High usage does not mean high value. A customer who runs your product twice a month to do something that saves them $500,000 a year is getting extraordinary value from almost no consumption. Usage-based pricing either undercharges them dramatically or creates a perverse incentive for them to manufacture fake usage to justify the spend. Neither is good.
The model looks different. The logic underneath is the same. Find a thing that's easy to count and hope it correlates with value. The bet just changed from headcount to compute.
There are four things that are structurally broken right now, and they're worth naming.
Efficiency now penalizes vendors. Zendesk's own data shows that companies deploying their AI agents resolve 30 to 50 percent of tickets without any human involvement. If Zendesk ran a pure seat-based model, their best product feature would shrink their ARR. They built a feature so good it argues against paying for the product. That's not a pricing edge case. That's a structural inversion, and it's going to show up in every product category where AI does knowledge work.
The incentive was always backwards. We just couldn't see it. When AI can eliminate three roles in a quarter, the backwards incentive becomes impossible to ignore. You're probably already feeling this at renewal. "We used to have 20 people at that function. Now we have 6. Why are we paying for 20 seats?"
CFOs don't hate usage-based pricing because it's complicated. I want to push back on that narrative. They're smart people. They can handle complexity. What they can't handle, emotionally and politically, is honesty. Per-seat billing gave them a number they could plug into a spreadsheet and not think about. Usage-based billing forces a question. How much are we actually using this, and is it worth what we're paying? For a lot of enterprise software, the answer to that question is uncomfortable. That's not a billing problem. That's a value problem that billing has been papering over.
Two trillion dollars in market cap didn't evaporate because of sentiment. It evaporated because investors ran the same scenario your enterprise CFOs are running. What happens to this company's revenue when their customers start automating? The answer, for companies still running per-seat models, is not good.
Right now there are five different answers to the question of what we're actually selling, all fighting for adoption. I'll give you my honest read on each.
Usage-based pricing is a bet that value lives in consumption. Pay for what you use. According to OpenView Partners' 2025 research, 61 percent of SaaS companies now offer it, up from 45 percent in 2021. Those companies grow 38 percent faster than subscription-only peers. The market is moving here and it's working in aggregate. The failure mode is that consumption and value are still just correlated, not equivalent. You've replaced a flawed proxy with a slightly less flawed proxy. Better. Not solved.
Outcome-based pricing is a bet that value lives in results. You pay for what works. Intercom's Fin AI agent charges $0.99 per ticket resolved. Outsourced human support runs $8 to $12 per ticket. In-house runs $25 to $35. Customers can do the ROI math in under a minute, which is exactly why Fin scaled to eight-figure ARR at 393 percent annualized growth. Zendesk followed at $1.50 to $2.00 per automated resolution. Salesforce Agentforce at $2 per conversation.
What gets cited less often is that a lot of companies tried outcome-based pricing in the past two years and quietly abandoned it within six months. Not because the model is wrong. Because they got the execution completely wrong.
Why Intercom made it work while others didn't comes down to three things most companies skip. They obsessed over when not to charge. They spent months defining what counted as a resolution and what didn't, before launch. Most teams don't do that work. They launch, customers dispute the first invoice, and the trust damage takes months to repair. They sequenced it into a pricing model that was already healthy, with customers who expected change and enough relationship capital to absorb a new structure. And they embedded the first batch of resolutions into every plan for free, letting customers experience the value before any money changed hands on outcomes. Most companies reverse this. They lead with the pricing and hope the value follows. It rarely does.
There's a problem that nobody in outcome-based pricing wants to talk about. Aisling O'Reilly at Intercom put it plainly. "A competitor with a lower per resolution price looks cheaper when they might actually be more expensive and deliver less value." That happens because outcomes aren't standardized. For the model to work at an industry level, three conditions have to hold simultaneously. Outcomes need to be explicitly defined upfront with no room for dispute at invoice time. They need to be verifiable, meaning customers can audit what actually happened without taking the vendor's word for it. And they need to be comparable, meaning resolution means the same thing across vendors so procurement can actually evaluate alternatives. Right now, none of those three conditions hold consistently. This is solvable. It's not solved yet.
The honesty cuts both ways. It's honest about value when it works. It's honest about your failure to deliver when it doesn't. Most companies aren't ready for that second kind of honesty.
Credit-based pricing is a bet that pre-commitment changes customer behavior. Figma is the most interesting live implementation right now, and they did several things that most companies get backwards.
Figma surged during the SaaSpocalypse. Their stock jumped 15 percent on Q4 2025 earnings while the rest of the sector was getting obliterated. Revenue hit $303.8 million for the quarter at 40 percent year-over-year growth. The reason wasn't their core product. It was what they did with AI credits.
In December 2025, Figma introduced an AI credit model and crucially, they did not enforce it. Free users get 500 credits per month. Enterprise full seats got 4,200. No one was charged for overages. Figma spent months watching what actually happened. What they found was a power law distribution.
A small subset of users was consuming AI credits at a rate nobody had expected. 75 percent of paid customers with more than $10,000 in ARR were using AI credits on a weekly basis. Some were already exceeding limits with zero enforcement. When Figma announced enforcement starting March 18, they weren't guessing at demand. They had already seen it.
Here's the part every SaaS founder should stop and reread. Credits aren't replacing seat-based pricing. They're selling more seats. Dev and Collab seats come with 500 credits per month. Full seats come with 4,200. If you want a higher allocation, you upgrade your seat. The AI credit model is functioning as a seat upgrade driver. Figma built AI monetization on top of their existing model in a way that reinforces it rather than cannibalizing it. That's not an accident. It's what thoughtful sequencing looks like.
The broader failure mode for credit-based pricing is the credit cliff. The moment a customer's balance runs low and they face a decision. Done well, that moment is a natural expansion conversation. Done badly, it's a surprise that feels punitive. Figma's answer is the shared pool subscription launching in Q2, where admins can buy credits the whole team draws from with a spending cap. That removes the individual user anxiety while keeping the consumption model intact.
Hybrid pricing is a bet that customers need a floor. A predictable base subscription plus variable usage charges above it. 46 percent of SaaS companies already use some version of this.
Bain called it the dominant interim strategy. I agree with dominant, I agree with interim, and I want to be clear about what interim means here. Hybrid pricing is what happens when you know the old model is wrong but you're not ready to commit to the new one. It's not a principled answer. It's a diplomatic one.
That's not criticism. It's the right call for most companies right now. The base protects your revenue floor. The usage component signals good intent on alignment. Just don't mistake diplomacy for strategy.
AI-as-labor pricing is a bet that the right comparison isn't other software. Price the agent like a contractor at a discount that makes the math impossible to argue with. A senior analyst costs $120,000 a year. This AI does 80 percent of that job for $15,000 a year. Here's exactly which 80 percent, enumerated.
This is the most honest frame for agentic AI products. Software isn't competing with other software anymore. It's competing with headcount. "Replacing an analyst" is marketing. "Processes 400 financial disclosures per month, flags anomalies within a 98 percent accuracy threshold, generates structured summaries, at $12,000 per year" is pricing. The specificity is what makes the deal credible and repeatable.
When software does the work of a human, pricing it like software is wrong. You should price it like a human but with a deal that's too good to refuse. That's the whole pitch. That's the whole company.
