For thirty years, enterprise software has priced itself the same way: count the humans who touch the tool, then charge for each one. A CRM, a helpdesk, a support platform, it didn't matter which category. If a person logged in, that was a seat, and a seat was a line item. It worked because it matched the thing being sold. Software was a tool a human used to do their job, so the price followed the human.
That logic breaks the moment software starts doing the job itself.
The unit of work changed. The pricing didn't.
When AI can carry a customer conversation from start to finish with no person involved, the seat stops being the unit of value. Nobody is sitting in that seat. Nothing is being licensed to a human working through a queue. The work moved from person to software, but most vendors kept billing as if it hadn't.
The result is a strange kind of doubling. Take a typical support operation today, as an example: somewhere around $115 per seat, per month, for the core helpdesk software. Another $50 to $60 per seat for the “value-added feature” add-ons layered onto that same platform. Then, separately, a bill from a standalone AI vendor, priced on usage or outcomes, for the conversations it resolves on its own. Three bills. Three pricing logics. None of them built with the other two in mind.
Pricing AI by its work fixed the metric, not the fragmentation
To be fair, part of the industry has already made progress here. A number of AI vendors have moved away from seat pricing entirely, charging instead for what the AI actually does: the conversations it handles, the volume it processes, the outcomes it delivers. You pay for the work, not for a seat that may or may not get used. That's a real improvement over the old logic, and it deserves credit.
But look at what these vendors are pricing, and why. They charge for the AI because the AI is what they built, and that's a fair way to price it: it reflects what they're actually selling. What they don't sell is the other half of the job.
AI still can't resolve every case on its own, and when it hits that limit, the problem doesn't go away, it needs a person. That person needs a system to actually work from: the full context of what the AI already tried, the history, the tools to act on it. Most AI vendors don't build that, because it isn't what they do. It ends up as a separate product, from a separate company, with its own login, its own contract, and its own definition of what counts as a win.
So a business trying to solve one customer's problem ends up buying two things that were never designed to work as one: an AI layer priced by conversation or resolution, and a system for the cases that AI can't close on its own. One counts conversations. The other counts tickets and seats. Neither is measuring whether the customer's problem got solved end to end, because neither one owns the whole path, just its own half of it. The business is the one left holding that gap: stitching two logins, two bills, and two definitions of success into something that has to feel like a single experience to the customer.
Pricing AI by what it does instead of by seat is the right instinct. It just stops at the edge of what AI can do today, and leaves the rest, the part still run by people, priced and built as a completely separate product.
One customer, one problem, split across systems that don't share a definition of success
A customer's problem doesn't arrive pre-sorted into "what the AI can handle" and "what needs a person." It's one continuous thing: a person with a question, at a moment in time. The systems handling it, though, were bought separately, built by separate companies, and each one only sees its own half. The AI layer counts its own conversations. The helpdesk counts its own tickets and the live agent queue. Nobody is measuring what the customer actually experienced end to end, because no single system was built to see the whole interaction.
That's the deeper problem with seat pricing, and it's also the problem usage-based pricing hasn't fully solved yet. It's not just about which unit you bill against. It's about whether the tools were ever designed to work as one system in the first place. A better price on a fragmented stack is still a fragmented stack. You may be paying for reduced seats (over time), but you're still paying for the seams.
What pricing should actually track
If software is doing the work end to end, the price should track the work end to end: not one team's seats, not another vendor's conversation count taken in isolation, but the full path a customer's problem takes from the moment it shows up to the moment it's solved, whether that's fully automated or finished by a person picking up where AI ran into exceptions that it does not yet own. That only works if the systems along that path share one definition of what "solved" means, instead of being stitched together after the fact by whoever's left holding the integration.
We built our own product on that premise instead of adding a pricing tweak to any legacy models. The AI and the human escalation step live in one system here, sharing one record of what happened and why, and we don't charge a seat fee for the human side of it. If we made money every time a case needed a person, we'd have no reason to make that happen less often.
Pricing AI by what it does, not the seat someone sits in, was the right correction to make. Building the thing you're pricing as one system, instead of stitching that correction onto a stack that was never built to be one, is the next one.