Why commission, in the AI era.
The cheap, flat software price you are reaching for is a leftover from an era that AI just ended. Everything in AI is usage-based underneath, so the honest question is what you pay for. Our answer is outcome-based pricing: you pay for the thing the AI was hired to achieve, and only when it happens.
By Naveh Mevorach, founder of Jyper · Published August 23, 2026
For thirty years software had one magic property: once it was built, one more copy cost nothing. That single fact is where the cheap, flat price comes from. AI does not have that property, and that changes what you should pay for it.
A CRM licence is small and flat because the vendor writes the code once and copies it a million times for free. You are paying for access to something that costs almost nothing to run. Use it once a month or ten thousand times, the vendor’s cost barely moves. That is why a flat monthly licence felt honest, and why it became the shape every travel tool takes.
AI is a different kind of thing entirely. Every brief it reads, every supplier it emails, every price it works out spends real money on the machines doing the thinking, every single time it runs. The cost does not vanish once the product is built. It comes back on every task, for the life of the product. AI does not behave like software. It behaves like labor.
AI is priced like labor, because it works like labor.
Labor has a cost every time it happens. An employee’s hour costs you whether or not that hour wins a booking. AI is the same shape: the cost lands on every task, tied to how much work you pull out of it. So the price of AI, if it is honest, has to track the work. It cannot sit flat and disconnected the way a software licence does, because the cost underneath it does not sit flat either.
This is why “unlimited frontier AI, one flat monthly fee” is a promise that cannot hold. Either the AI is throttled hard enough to be cheap to run, which means it is too weak to actually do your quoting, or it is genuinely capable and the vendor is quietly losing money on every heavy user until the day they cap it, meter it, or raise the price. A cheap, flat AI price is not a bargain. It is the tell that the AI underneath is not really doing the work.
You have already met this rule, probably on your own phone. The $20 chat subscriptions from OpenAI and Anthropic come with usage limits, and the $200 tiers come with bigger usage limits, and heavy users hit both. Push past them and the answer from every lab is the same: switch to paying for usage. Even the companies that make the models cannot sell you a true flat fee, because underneath, every single thing in AI is usage-based. The subscription tiers are starting points, not prices.
So the flat number you are reaching for is, more often than not, the price of the demo. Real AI, the kind that reads the contract and builds the costing and chases the supplier who went quiet, has a real cost behind every run. A real price has to reflect that. The only open question is which version of “pay for the work” you sign up to.
Once the price tracks the work, one question is left.
Accept that AI has to be priced against the work it does, and the whole argument narrows to a single fork. You can pay for the work attempted, or for the work that paid off.
Pay for the work attempted and you are on a usage meter: tokens, runs, credits. Every draft, every dead-end supplier search, every quote that never closes still adds to the bill. The meter does not care whether any of it made you money. It is honest about cost, but it quietly puts the vendor back on the wrong side: they earn more the more the machine churns, win or lose.
Pay for the work that paid off, and the bill only moves when money actually lands in your account. This has a name: outcome-based pricing. You do not pay for tokens, seats or attempts. You pay for the thing the AI was hired to achieve, and nothing else. It is the same “price tracks the work” logic, turned the one direction that is unambiguously on your side, and it is the only pricing where the vendor has real skin in the game: if the AI does not deliver the outcome, the vendor eats the cost of every token it burned trying.
We did not invent this, and that is the point. The clearest example outside travel is Fin, Intercom’s AI support agent and the biggest success story in AI customer service. Fin does not charge per seat or per message. It charges $0.99 per resolution: a customer conversation actually resolved, or you pay nothing. The outcome Fin exists to produce is the only thing on the invoice. The serious AI companies are converging on this shape for the same reason we did: it is the only price that proves the product works.
You pay Jyper only when you get paid.
Jyper is paid the way the travel industry itself is paid: on confirmed business. We invoice a commission only on business you actually confirm through Jyper. No confirmed booking, no fee. Our invoice is not a cost sitting on top of a maybe. It is proof that money already reached you. This is outcome-based pricing applied to travel: the outcome Jyper exists to produce is confirmed business, so confirmed business is the only thing on the invoice. Fin’s outcome is a resolved conversation. Ours is a won booking.
Follow it through and every incentive lands where you want it. We earn nothing from your headcount, so we have no reason to keep people parked at screens. We earn nothing from quotes that die, so we have every reason to make your proposals land faster, read sharper, and close more often. We get paid at the very end of the pipeline, so we have to care about all of it: the reading, the supplier replies, the costing, the proposal your client actually opens. The only way our number goes up is if yours goes up first.
That is the best deal AI can offer you. Not the lowest sticker. The one where the company selling you the machine has staked its own revenue on the machine earning you money.
You decide how much of our own pay rides on the promise.
Because it is a salary, you get to set it like one. There are three ways to pay your AI employee, and the real difference between them is how much of Jyper’s own money we put behind the promise.
Fixed is a flat monthly salary and no commission at all: budget it once and forget it, for when you just want the work done and the number still. Aligned is a smaller base plus a commission on confirmed business: we split the outcome with you. Starter is a minimal base and a higher commission: your AI employee works nearly for free until you win, and we carry almost all of the risk. That last one is the tier we are happiest to put on the table, because it only pays us if we are right about ourselves.
You choose which of those we live on, you move between them whenever you want, and the whole thing switches off at the start of any month. Three things are fixed in writing: the commission only ever touches business confirmed through Jyper, never your overall turnover; the numbers are set with you on a call so they fit your margins before you commit; and the first 30 days are fully money-back. You are not signing up to an open meter. You are deciding how hard we have to bet on our own product.
When flat pricing is exactly right.
None of this makes per-seat pricing a trick. For software you operate, it is the honest price. A CRM, an itinerary builder, a quoting platform where your team still does every step: you are renting a nicer workspace, so paying per person renting it is exactly fair. We said the same thing from the other direction in Is Jyper worth it?: never pay a commission for a better broom.
Match the price to what the thing actually is. Per-seat for a tool your team drives. Outcome for an employee that does the work itself. The mistakes are the crossed pairs: a commission on a better broom is a bad deal for you, and a flat per-seat fee for real AI is a bad deal too, just a quieter one, because it only pencils out for the vendor if the AI never quite does its job.
Ours is the second kind, priced accordingly. Whether Jyper truly does the work in your operation is not something to take from this page: run the pilot, measure it, and hold us to the result.
Four questions for every AI vendor. Including us.
Whatever you end up buying, from us or anyone else, put these on the table and keep the answers in writing:
If this product works perfectly, what happens to my bill?
With per-seat pricing the bill stays flat while the value story is supposedly enormous. With commission pricing the bill grows only if your confirmed business grows. One of those two invoices is evidence the product worked.
What happens to your revenue if I need fewer seats next year?
This is the reverse-incentive test. A per-seat vendor loses revenue when your team does more with fewer people, so your successful AI transformation is their bad quarter. Ask them how they feel about that sentence.
What exactly triggers a fee, and what never does?
For Jyper: a fee is triggered only by business you actually confirm through Jyper. Quotes that do not close cost you nothing beyond your plan base. Get any vendor’s trigger in writing, including ours.
How do I leave?
You should be able to switch off month to month and export everything. A pricing model only proves alignment if walking away is easy; otherwise it is just a different way to lock you in.