Tolla shows what each free user costs you in AI usage, which ones go on to pay, and which accounts are the same person — then tells you which changes pay for themselves.
A free user of a classic SaaS app cost almost nothing. A free user of an AI product burns tokens on every click. It’s hard to say what a free tier costs — or what it buys you.
Some free users become your best customers. Others open five accounts and never pay. Today both look the same on the bill from your AI provider.
Average gross margin on AI products in 2025, against about 80% for software overall.
ICONIQ · Jul 2026 ↗ · Aleph × Benchmarkit · Jun 2026 ↗Median free-to-paid conversion across 200 B2B software products, trials and freemium, within six months.
Growth Unhinged × ChartMogul · Feb 2026 ↗More than one in six sign-ups at AI companies is linked to multi-account abuse: one person behind several accounts.
Stripe · May 2026 ↗More abusive free trials detected across Stripe's network from November 2025 to February 2026, with AI companies driving much of the rise.
Stripe · Mar 2026 ↗Every AI call is priced from its model and tokens, or from the cost you send, and added up per person. You see what each free user costs, which features drive it, and how few people account for most of it.
What you seeYour 20 costliest free users, what each one cost this month, and the model behind it.
Stripe tells Tolla who pays and since when. Tolla lines that up with free usage: how much customers used before buying, how long they took, and which free users look like them today.
What you seeHow much free AI your customers used before paying, next to the free users who never did.
Accounts are linked when they share a card, an email with dots or a +tag removed, a device or a network, each weighted by how rare it is. Every link shows its evidence and a confidence: a lead, not a verdict.
What you seeThree accounts on one card, a week apart, each with the signals that link them.
Instead of one free tier for everyone, each group gets its own share of it. Scripts get none until a human check, repeat accounts add a card first, heavy users who never look at pricing get less, and people about to buy get a little extra. Prices never change.
What you seeSomeone who hits the limit right after viewing pricing gets 10% more credits and 3 more trial days.
A change first runs in shadow on your live data and touches nobody. Then it goes live while some people keep today's free tier, and Tolla measures savings and conversion against them. If conversion drops clearly, it rolls itself back.
What you seeThe saving as a range, conversion in both groups side by side, and a verdict: keep, roll back or too early.
A read-only Stripe key tells Tolla who pays and since when. An export of your AI usage tells it what each free user costs; your provider's bill alone gives an estimate. Tolla reads the last 90 days.
No code to start. Tolla can't charge, refund or change anything in Stripe.
Tolla reads.
It never touches billing.
Asked before signing up.
Tolla uses a restricted key you create with read permissions only. It cannot charge, refund or change anything, and you can revoke it in Stripe with one click.
No code. A read-only Stripe key, which Tolla walks you through creating, and a CSV of your AI usage. If you don't have one, Tolla gives you a prompt for your coding agent (Claude Code, Cursor or Copilot) that writes the query for you. Using a tool like Langfuse or Helicone? Export from it, or let your coding agent pull the usage through its API. Only have your AI provider's bill? Upload its cost export and say how many people used your product: Tolla estimates the free users' share as a range, and a week of its SDK makes it exact.
Not unless you turn a change on. The audit only reads. A change runs in shadow first, where it only watches; when you take it live, through the SDK or your own code, a control group keeps today's free tier and Tolla rolls it back if it costs sales.
Every link shows its evidence and a confidence score. Families sharing a card can look alike, so links are leads, not verdicts.
Radar stops fraud at payment. Helicone shows AI cost per request. Tolla joins both with sign-ups, so you see cost, conversion and duplicates per free user, and then tests what each person should get.
A query gives you one number, once. Tolla links accounts that belong to the same person, tries a change in shadow, measures it against a control group and rolls it back if it costs sales. That is weeks of work, redone at every pricing change.
You get a report of what your free tier costs and the changes that would pay for themselves. Your first 3 months of Tolla are free, with no card. Keep it running if it is useful; delete your workspace if not.
A free-tier audit takes a read-only Stripe key, an export of your AI usage, and about ten minutes. You see what each free user costs before you change anything.