Research
Almost everything written about AI prediction tools is written by people who have never run one at scale. These studies are the opposite: measured figures from MiroFish.us, a live product with 17,974 registered users and 126 paying customers, published with the method attached so you can judge them.
Where a number is an estimate, it says so. Where the sample is too small to be meaningful, it says that too. Everything is aggregate — no individual user, scenario, or address appears in any dataset we publish, ever.
What an AI prediction actually costs to produce
One prediction makes 74 model calls and costs $0.0123 — about 1.2 cents. Card processing on a $2.99 plan still costs about 3 times more than a whole month of AI.
Where AI prediction users come from — and where they pay
A 38-country breakdown of signups against paying customers. The largest source of signups converts at 0.24%; the best-converting real market has no translated site at all.
How we work
- Aggregate only. No scenario text, email address, or full IP address is ever published. Country resolution uses IP addresses truncated to their /24 network before lookup, so precise addresses never leave our systems.
- Method stated up front. Sample sizes, observation windows, and the difference between counted and projected figures are given on every study.
- Small numbers are labelled. A country with three signups and one customer is noise, not a 33% conversion rate, and we mark it as such rather than quoting the flattering figure.
- Corrections stay visible. If a figure turns out to be wrong, the page is updated and the change noted rather than quietly edited.
Research by Zinedine, Founder & Developer. Questions about methodology, or a request for a figure not published here? Get in touch. Findings may be quoted with attribution and a link to the source page.