Method
How Taquet picks ideas
Every idea in the catalog gets six scores from 0 to 5. Code computes each score from a fact we measured, and the page shows the score next to that fact and its source. When we haven’t measured something, the page says so and gives no score.
The six criteria and their thresholds
| Criterion | What we measure | How the measure becomes a score | Default weight |
|---|---|---|---|
| Proven demand | How long a seller’s ad has been running, and the price listed on the seller’s sales page. | Days the seller’s ad had been running when we checked it. Below the first threshold: 0. From 1: 1. From 30: 2. From 60: 3. From 120: 4. From 240: 5. With no price listed on a sales page, the score is capped at 3. A price alone, with no ad, scores 1. | 3 / 5 Starting weight, the same for every criterion until enough tests have closed |
| Ads can reach the audience | Monthly Google searches for the service, and the audience size Meta estimates for the idea’s targeting. | Monthly Google searches for the service, added up. Below the first threshold: 0. From 1: 1. From 500: 2. From 2,000: 3. From 10,000: 4. From 50,000: 5. Meta’s estimated audience, lower bound. Below the first threshold: 0. From 1: 1. From 50,000: 2. From 250,000: 3. From 1,000,000: 4. From 5,000,000: 5. The better of the two gives the score. | 3 / 5 Starting weight, the same for every criterion until enough tests have closed |
| People know the problem | Monthly Google searches about the problem, and how they changed over twelve months. | Monthly Google searches about the problem, added up. Below the first threshold: 0. From 1: 1. From 500: 2. From 2,000: 3. From 10,000: 4. From 50,000: 5. One point more when searches rose 20% or more in twelve months, one point less when they fell 20% or more, never below 1 while there are searches. | 3 / 5 Starting weight, the same for every criterion until enough tests have closed |
| Margin per order | Whether one of our recipes makes the product and sends it automatically after payment, and the margin at the listed price, based on what that recipe spends on one order. | Margin at the listed price: the price minus what the recipe spends on AI and tools for one order, divided by the price. That cost is the average across real orders or, before the first orders, the most the recipe is allowed to spend. Below the first threshold: 0. From 1%: 1. From 25%: 2. From 50%: 3. From 75%: 4. From 90%: 5. No score without a recipe. | 3 / 5 Starting weight, the same for every criterion until enough tests have closed |
| Easy to buy on the spot | The listed price: a price people pay without a call or a quote. | The listed price, as written, whatever the currency. Up to 30: 5. Up to 60: 4. Up to 150: 3. Up to 500: 2. Above that: 1. No score without a price. | 3 / 5 Starting weight, the same for every criterion until enough tests have closed |
| Few constraints | Whether the field is regulated, and whether you have to buy tools before starting. | Every idea here is outside regulated fields. 3 while the tools haven’t been assessed, 4 when a tool has to be bought first (its price is shown), 5 when no tool is needed. | 3 / 5 Starting weight, the same for every criterion until enough tests have closed |
How the ranking works
- You choose how much each criterion matters, from 0 to 5. Three presets do it in one click.
- Ideas are ranked by the average of their scores, weighted by your choices, over the criteria we’ve measured. A criterion we haven’t measured neither lifts nor lowers an idea.
- If you name a trade or an audience you know, the ideas for it come first.
- Each idea says in one line why it ranks where it does, with the measurement behind it.
Three checks before an idea is published
- A recipe that delivers it. A recipe is the program that makes the product and sends it to the buyer right after payment, with no one stepping in. An idea is published only if one of our recipes makes exactly that product. A list of tools is not enough: until the recipe is written, nothing reaches the buyer.
- A tool chain and its cost. For each idea we list the tools that would deliver one order, step by step, and add up what they cost. The prices come from our tool inventory, where each one has its source and the date we read it. An idea with no tool chain, or with a tool whose price we don’t know, stays out of the catalog.
- A jury. A second AI model, not the one that wrote the idea page, answers three yes-or-no questions. Does the seller’s page we saved prove that people pay for this? Can the tool chain deliver it at the quality the seller shows? Can the buyers be reached with a targeted ad? A single no keeps the idea out, and we record the reason.
- An idea whose tools would call or text people, or use property sale records, waits until we have read the law that applies.
What we leave out, and why
- An idea whose proof page also backs another idea: one page proves one idea.
- Evidence that is twelve months old or more, or dated in the future.
- A quote we can’t find word for word in the copy of the page we saved, and a figure or a price that isn’t in its source.
- An idea proved by an ad that was no longer running when we checked it.
- Anything in a regulated field: health, credit and finance, individual legal advice, employment, housing.
- A score that would come from an opinion instead of a measurement.
What we don’t know yet
- Every criterion starts with the same weight. Nobody has finished a real test of these ideas on Taquet yet, so we can’t say which criteria best predict a sale. Each closed test goes, anonymously, into Taquet’s test base with its net result (sales minus ad spend and tool costs). Once 20 tests have closed, the weights adjust to those results on their own, and the table above says where each weight comes from.
- Google gives no figure for some searches, those aimed at children for example. The idea’s page says so.
- The cost of one order and the tools it takes are being worked out idea by idea. Until then, “Margin per order” has no score.
- Meta’s audience estimate hasn’t been collected yet for the ideas online today.
