Per-seat pricing is the wrong shape for AI

Most AI vendors are charging the way SaaS vendors have charged for fifteen years: pay per seat, per month, whether the seat is used or not. The pricing is borrowed reflexively from the old playbook. It’s also wrong.

The mismatch is structural. Per-seat pricing exists because access to a tool was the unit of value. You give a salesperson a CRM seat, the salesperson logs in every day, the seat earns its cost in deals tracked. The math works because the seat is the thing being used.

AI doesn’t work that way. The thing being used is the model’s compute — and that’s billed by token, not by seat. A user with a Copilot license who never opens Copilot costs the same as a user who runs ten thousand completions a week. The vendor is okay with that because they’re betting on average behavior across a population. But the buyer is paying for headcount they may or may not be activating.

That gap — between what’s being charged for and what’s being consumed — only gets worse as AI use intensifies. A firm with a thousand employees can’t predict whether twenty of them or two hundred will actually pull value from the tool. Per-seat pricing forces them to pay the bigger number to keep their options open. So they overbuy, then under-deploy, then write the whole thing off as “AI didn’t work for us.”

What I price for instead

Everything I build for clients runs on utilization. You pay for the work the system produces. Use it once, pay for once. Use it a thousand times, pay for a thousand. Don’t use it, don’t pay.

This isn’t a clever pricing trick. It’s the only model that mirrors how the underlying AI is actually billed. OpenAI charges me by the token. Azure charges me by the page processed. Google charges me by the embedding. When the cost the vendor pays scales with use, the price the buyer pays should scale the same way. Anything else is the vendor pocketing the gap.

It also aligns three things that per-seat pricing actively misaligns.

The provider’s incentive. If I get paid only when the system produces value, I’m motivated to make the system produce more value. I’m not motivated to inflate license counts or upsell seats. The product gets better because that’s where the revenue is.

The buyer’s risk. A firm trying a new AI workflow doesn’t have to commit to a year of licenses to find out whether it pays back. They commit to one process, run it through the system, see what comes out the other side, and pay for what got produced. If the system doesn’t move the needle, the bill stops on its own.

The buyer’s growth. Per-seat punishes adoption. Every new employee onboarded is an additional cost — so the firm rations access. With utilization pricing, more usage means more value generated, which means more cost — but proportionally, and with offsetting value already in hand. The buyer scales without flinching at the next renewal.

What I tell prospects

When I quote an engagement, the conversation goes like this:

I’ll build the system for a fixed scope. After that, you pay per run — title commitment processed, legal description parsed, plan reviewed, whatever the unit of output is for your workflow. Here’s the per-run rate. Here’s an estimated monthly cost based on your current volume. If volume goes up, your bill goes up — but you’re shipping more work. If volume goes down, your bill goes down. You’re not paying me to be on retainer. You’re paying me for things I produced for you.

The conversations end faster. Prospects either say “that aligns with how I think about cost and value” and we keep going, or they say “I’d prefer a flat monthly retainer for budget predictability” and we both find out earlier we’re not a fit. Either outcome saves time.

A note on the tools themselves

I’m aware this puts me at odds with how most AI vendors price themselves. That’s fine. The big tool vendors price for population averages — they have to, because they’re selling to a million firms at once. I price for one client at a time and one process at a time, where I can actually measure what the system is producing. Different scale of business, different shape of pricing.

If you’re a vendor selling to a million firms, per-seat probably still makes sense for you. If you’re a buyer, or a builder embedded inside a firm, it almost never does.