Key Takeaways
- Per-seat AI pricing often masks significant waste—organizations typically pay for 500 seats but get value from only 200 active users
- The true cost structure of AI (compute, tokens, retrieval) varies dramatically by user, making per-seat models misaligned with actual value delivery
- Usage-based models better align spending with business value, ensuring you only pay for the AI your team actively consumes
Let's do some math that nobody in your organization has probably done yet.
Your company has 500 employees. You're evaluating AI tools. The one your team likes costs $30 per user per month.
That's $180,000 per year.
But here's what the math doesn't show you: within six months, maybe 100 of those people will use the tool regularly. Another 150 will log in occasionally. The remaining 250 will forget it exists.
The effective cost per active user when only 40% of seats are actually being used—more than double the sticker price.
You're paying $180,000. You're getting value from maybe $60,000 worth of seats.
The rest is waste—but it's invisible waste, buried in a line item that looks like it's working as intended. This is the hidden economics of per-seat AI pricing, and it's costing companies far more than they realize.
How We Got Here
Per-seat pricing made sense for traditional software. When you bought a CRM license or an office productivity suite, you were buying access to a static tool. Every user consumed roughly the same resources. The marginal cost to serve user #500 was nearly identical to that of user #5.
So vendors charged per seat, customers budgeted per seat, and everyone understood the model.
Then AI came along, and everyone just... kept using the same pricing model. It made the transition feel familiar. Enterprises knew how to budget for it. Vendors knew how to sell it. But many organizations learn this lesson the hard way when their AI pilots fail to scale.
But AI economics are fundamentally different, and forcing them into a per-seat model creates problems that compound over time.
The Vendor Incentive Mismatch: When vendors lock you into per-seat contracts, their revenue is secured regardless of your deployment success. They are incentivized to sell licenses, not to drive the deep, daily engagement that actually transforms your business.
The Actual Cost Structure of AI
Here's what's really happening when someone uses an AI tool:
Compute resources: Every query consumes compute resources. Large language models run on expensive GPU infrastructure, and that infrastructure costs money every time it processes a request. Simple questions cost less. Complex analyses cost more.
Token processing: AI models charge by tokens—roughly, words—both input and output. A quick question might use a few hundred tokens. A document analysis might use tens of thousands.
Retrieval costs: When AI searches through your knowledge base to find relevant information, that's not free either. More sophisticated retrieval means better answers, but also higher costs.
A power user who runs 50 complex queries a day might consume 100x the resources of someone who asks one simple question a week. But under per-seat pricing, they pay the same amount.
This disconnect between price and cost creates strange incentives for everyone involved.
What Per-Seat Pricing Actually Costs You
Beyond the sticker price, per-seat AI pricing has several hidden costs that don't show up in your initial analysis.
The rollout tax. You can't afford to give everyone access at $30/head, so you start with a pilot team. Then you expand to another department. Then another. Each expansion requires budget approval, procurement cycles, and political capital. By the time AI reaches your whole organization, you've spent months in bureaucratic friction that a different pricing model would have avoided entirely.
The rationing problem. Someone has to decide who gets seats and who doesn't. That means someone is deciding who gets access to productivity tools and who doesn't. Your highest-paid knowledge workers probably get seats. Your frontline employees—the ones answering customer calls, processing orders, and onboarding new hires—probably don't. The people who might benefit most from AI assistance are often the last to get it.
The shelfware effect: You're paying for 500 seats, but you have no mechanism to reallocate unused licenses to people who might actually use them. That marketing coordinator who requested access, used it twice, and forgot about it? You're still paying for her seat. And you probably don't even know.
The budget ceiling. At some point, you hit the maximum number of seats that finance will approve. You're stuck at that number even if demand exceeds it, even if the ROI would justify more. The pricing model creates an artificial cap on value creation.
The shadow AI problem. Employees who don't get seats find workarounds. They use personal ChatGPT accounts. They paste company data into free tools. They find ways to get AI assistance that IT can't see, govern, or secure. Your per-seat pricing just pushed sensitive data outside your control.
The Math Nobody Does
Here's an exercise worth doing before your next AI procurement.
Look at your current software tools that charge per seat. For each one, calculate the percentage of licensed users who actively use the tool each month. For most enterprise software, this number is somewhere between 40% and 70%.
Now apply that same percentage to your AI pricing. If you're paying for 500 seats and expect 50% active usage, you're really paying $60 per active user, not $30.
But it gets worse. Among active users, engagement varies wildly. Some people use AI constantly. Some use it once a week. When you account for this, the effective cost per meaningful use might be $100 or more per user.
Would you have approved this purchase at $100 per user? Probably not. But that's what you're actually paying.
The Alternative: Usage-Based Pricing
There's another model that's gaining traction, especially among AI-native companies: charge for what people actually use.
The logic is simple. AI has real marginal costs. Usage-based pricing aligns the price you pay with the value you receive and the costs the vendor incurs. Heavy users pay more because they're getting more. Light users pay less. Non-users pay nothing.
This changes the economics in several important ways.
Removing Barriers to Entry: This model creates a "try before you buy" dynamic within your own organization. Because there is no upfront penalty for adding a user, you can grant access to the entire company immediately. This allows organic discovery of use cases you never anticipated.
Everyone can have access. When adding a user costs nothing until they actually use the tool, you can give your whole organization access from day one. No rationing. No deciding who deserves innovation. Everyone can discover what AI does for them.
Cost tracks value. If a team is using AI heavily, it's presumably because they're getting heavy value. You're paying more, but you're getting more. If a team isn't using it, you're not paying for it. The budget naturally flows to where the value is.
Adoption isn't punished. Under per-seat pricing, successful adoption means budget pressure. "Everyone wants access" becomes a problem to solve. Under usage-based pricing, adoption is just... adoption. It scales naturally.
Budget becomes predictable in a different way. You can set spending caps that guarantee your maximum monthly cost. You know your ceiling. What varies is how much value you extract within that ceiling.
The CFO's Checklist
If you're evaluating AI tools, here are the questions to ask:
What's the real cost per active user? Don't accept the sticker price. Estimate realistic adoption rates and do the math on effective cost.
What happens when we want to expand? Is there a natural path to company-wide rollout, or will each expansion require a new budget battle?
How do we handle variable usage? If one team uses AI ten times more than another, does the pricing model accommodate that? Or are we subsidizing light users with heavy users' budget?
What's the cost of non-adoption? If 40% of seats go unused, what does that mean for our effective ROI? How would that change our decision?
Are there spending controls? For usage-based models, can we set caps? Get alerts? Prevent runaway costs?
What are employees doing without access? If we limit seats, are people finding workarounds? What's the security cost of shadow AI?
A Different Way to Think About It
The real question isn't "what does AI cost?" It's "what does AI cost relative to the value it creates?"
Per-seat pricing makes this calculation harder than it needs to be. You're paying a fixed amount regardless of how much value you extract. The price is predictable, but the ROI is opaque.
Usage-based pricing ties cost directly to activity. If you're paying more, it's because people are using it more. If they're using it more, presumably they're finding value. The price varies, but the relationship between cost and value is clear.
You can pay a perfectly predictable $180,000 per year for a tool that delivers $50,000 in value. The budget line looks clean. The ROI is terrible.
For CFOs used to the predictability of per-seat models, usage-based can feel uncertain. But predictable costs aren't the same as predictable value.
The Transition
The industry is moving toward usage-based pricing, slowly. The major cloud platforms already work this way. API-based AI services work this way. The per-seat model is a holdover from an era when software economics were different. Understanding how to build a CFO-ready AI business case helps navigate these transitions.
If you're evaluating AI tools today, you likely have options. Some vendors still charge per seat. Others have moved to usage-based models. Some offer hybrid approaches.
The right answer depends on your organization. If you're confident that every seat will be heavily used, per-seat might work fine. If you expect variable adoption—which is the reality for most organizations—usage-based models are worth serious consideration.
At a minimum, do the math. Real math, not vendor math. Account for realistic adoption rates. Consider the cost of rationing and the cost of shelfware. Factor in the shadow AI problem.
When you run those numbers, the "predictable" per-seat model often looks a lot less attractive than it did on the first slide of the sales deck.
JoySuite uses usage-based pricing with unlimited users. Your whole organization gets access from day one—no rationing, no deciding who deserves innovation. You pay for the AI you actually use, and Budget Shield lets you set spending caps so costs never surprise you. See all the capabilities your team gets from day one.
