Why the Map Is Free and the Interview Is Not

Most software pricing is a guess dressed as a tier. Three columns, a middle one highlighted, and a set of feature ticks arranged so the thing you actually need sits one row below the plan you were going to buy.

There is a more honest way to decide what should cost money, and it is not a marketing question. It is an arithmetic one: what costs us something every time you use it?

The test

Take any feature and ask what happens to the supplier's costs when one customer uses it a thousand times instead of once.

For most software the answer is nothing measurable. Storing your records, rendering a view, letting you edit and export, serving a page. These are close enough to free that charging by usage would cost more to meter than to provide.

For anything that calls a large language model, reads a document, transcribes a call, or runs a research query, the answer is different. Every single run costs the supplier real money, immediately, and that cost scales exactly with how much you use it.

If those two categories are priced the same way, somebody is being lied to. Either the light users are subsidising the heavy ones, or the pricing has a cliff in it that nobody mentions until you hit it.

What this means in practice

The parts of abi. Clone that cost nothing per use are free and stay free. Building your map, editing it, exporting it, presenting it, sharing it as a view-only link. No card, no trial timer, no expiry, because a timer on something that costs nothing to run would be a lever rather than a price.

The parts that cost money every time they run are the paid ones. Letting the system interview a team, reading a folder of documents, reviewing a process and proposing changes. Those consume something per use, so they are metered per use, in credits included with a plan.

The pleasant consequence is that the free version is not a demo. It is the actual product with the expensive operations turned off, which is a different thing from a trial designed to become annoying on day fourteen.

Where usage pricing goes wrong, and what to look for

Credits are not automatically honest. Three things separate a usage model that respects you from one that does not.

You can see the price before you spend it. If an operation costs twenty-five credits, that should be visible at the moment you press the button, not discoverable in a billing report. abi. Agent Studio prices a process review at twenty-five credits and charges nothing for designing and editing on the canvas, because editing costs nothing to provide.

Unused capacity is not a trap. Watch for models where the meter creates anxiety, because anxious users use less, and a supplier whose revenue rises when you use the product less has an incentive pointing the wrong way.

The expensive thing is the valuable thing. If the metered operation is something you would happily do more of, the incentives line up. If the meter sits on something you have to do, like inviting a colleague or exporting your own data, that is a toll booth on a corridor, and it will get worse.

The counter-argument

Usage pricing is harder to budget than a flat fee. A finance director who has to approve a variable number every month will reasonably prefer a slightly worse deal that is predictable, and anybody telling you predictability has no value has not sat in that meeting.

The workable compromise is a plan with a generous monthly inclusion, so ordinary use never touches the meter and only genuinely heavy months cost more. That is what a credit allowance is for, and it is worth checking, before you sign anything, how many of the operations you actually intend to run the allowance covers. If the honest answer is "about a third of them", the meter is the price and the plan is the marketing.

Work out your own number first. Then compare it to the allowance rather than to the headline.