On this page
What the factory is
A small machine for testing product ideas with real money. An AI agent does most of the work: it screens ideas, builds the pages, runs the ads and keeps the books. A person sets the spending limit in the ad account, approves every payment link and reads every entry before it is published here.
The money
- €500 of our own money, counted without VAT, to last until 13 April 2027.
- No more than €100 on any one idea.
- Money a product earns goes back into the budget fourteen days after it is paid, once the refund window has closed.
The measure
One number: money received minus money spent. Received means what the payment processor reports, after fees, refunds and VAT. Spent means ads and services. Both sides are counted without VAT. The model's opinion of an idea is not part of the measure.
How an idea gets in
Before any money is spent, an idea is checked at the desk against measured numbers:
- Enough people search for the problem: at least 10,000 searches a month.
- A click costs no more than €1.
- The price is at least 33 times the cost of a click. Put another way: no more than 3 in 100 visitors need to pay for the ads to pay for themselves.
- Someone already sells a solution at that price or more.
- The product can be delivered within seven days.
Numbers come from measurements, with a source and a date. A number the model guessed is marked as a guess and cannot let an idea in. Ideas that fail the check are published too, with the reason.
Stage one: signal
Up to 50 clicks, costing no more than €50, within ten days. The page says plainly that the product is in early access. What counts is a refundable deposit, or a reservation at the stated price. An email address alone does not count.
- Fewer than 50 clicks: no reading. The idea goes back in the queue once. A second time, it is dropped.
- With 50 clicks or more, it passes only if at least three people act and they are a large enough share of the visitors: three times the share that would need to pay.
Stage two: money
What is left of the €100. A real product, real payments, refunds on request.
- Kept: within 30 days the idea earned at least everything spent on it.
- Dropped: it did not. Or earlier: nobody has paid by the time the ad spend equals the price of two sales.
Predictions
Before spending on an idea, the factory writes down what it expects: cost per click, clicks per day, the share of visitors who act and the share who pay. Afterwards it publishes how far off it was. After each finished attempt one assumption is changed, the one that was most wrong, and only if the change would have made the earlier predictions better too.
When it stops
- The money or the time runs out.
- Three ideas in a row pass stage one and fail stage two. Then the rules are the problem, not the ideas.
- After three finished attempts the predictions are not getting better.
- An ad account or a payment account is restricted, or a legal question comes up.
What will not happen here
No invented reviews. No countdown timers and no fake scarcity. No promises of income. Pre-orders only with a delivery date; if the date is missed, the money goes back automatically.
What came before
We ran paid experiments before this one. They are told here separately, without names, as background. They are not counted in the ledger.
The numbers
The same numbers the ledger uses. They are read from this record, not from the code.
- Budget
- €500, without VAT
- Most on one idea
- €100
- The experiment ends
- Version of the rules for choosing ideas
- 1
Amendments
None yet.