Why an unlimited free trial beats a metered one for AI products

What Everyone Says

Meter your AI usage in the free trial. Otherwise the LLM costs will eat you alive.

It's the default advice in every SaaS pricing thread right now. LLM costs are usage-based. Trial users are freeloaders. So you give them a few tokens, a handful of resolutions, a small number of prompts. Just enough to see the shape of the product. Then you make them pay to actually use it.

Every SaaS finance blog says the same thing. Never subsidize trial usage at scale. You'll either go bankrupt or attract the wrong customers. It sounds like the only sane move.

Why That's Wrong

The logic falls apart at the moment that matters most. The moment a prospect decides whether your AI is actually any good.

Karel Papik at Product Fruits tried the metered version first. Customers got "a few tokens or we call that resolutions" to try the AI copilot. The result was simple: it was a barrier. Prospects hit the cap before they got to the value. They dropped out of the trial confused about what the product even did.

The hidden assumption in the conventional wisdom is that trial users are all the same. They're not. A capped trial filters out the people who need to see the product work under real load. Which is exactly the segment that would have converted.

What you save on LLM costs, you lose in conversion. And you multiply that by however many qualified buyers never made it past the meter.

What Karel Did Instead

He removed the cap entirely.

Here's how he puts it. "We will give them zillions, zillions. Do anything you want to. We will cover the expenses. Because you want to try. Because we know that if they will try, they will become customers because the value is there."

That's the whole rule. If the value is real, friction during the trial costs you more than the infrastructure does.

And Product Fruits had some external validation on this. Karel had a call with Gartner about AI pricing strategy. Their advice: "this is the way you should do. That's the only way." Subsidize the trial. Let prospects hit the limits of what the product can do, not the limits of your budget.

Then they solved the pricing side separately. Customers wanted flat pricing, so Product Fruits charges a flat monthly fee starting at around $100/month. They eat the variable LLM cost and manage it in aggregate. It's risky. Karel is open about that. But the trial is unlimited, and that's what drives conversion.

The Principle Underneath

The old pricing playbook was built for deterministic software. You measured seats, API calls, or storage because usage was predictable. You could model it.

AI breaks that assumption. Usage is stochastic. Value is experiential.

When value can only be felt through use, anything that limits use limits your conversion rate. The job of a free trial isn't to control costs. It's to create belief.

That's the reframe most founders miss. They treat the trial like a budget problem when it's actually a belief problem.

Should You Do This?

Do this if your product's value becomes obvious within the first real use, and if a paying customer's LTV covers the trial subsidy with room to spare.

Skip this if your product needs weeks of use before value shows up. In that case, an unlimited trial just bleeds money.

The question to ask: what do trial users need to feel before they believe? Then remove every cap that gets in the way of feeling it.