From a complaint somebody wrote to an idea you can build
Five stages, each one resumable, each one traceable. The whole design goal is that you can get from any claim on this site back to the original post in one click.
Ingest
Twelve connectors pull raw text from where people describe problems.
Forums, software review sites, app stores, job boards and revenue datasets. Each document is stored with a content hash computed on the prose, so the same complaint reposted elsewhere is caught as a duplicate rather than counted twice.
Every run is logged with its counts and any error. Failed runs stay visible rather than being quietly swallowed — a broken connector you can see beats a silent gap in the data.
Extract
A model reads each document and pulls out the specific friction.
Not a summary — a structured record: the normalised problem statement, a verbatim quote, the job to be done, who has the problem, how severe it is, and what workaround they use today.
Most documents produce nothing, and that is the correct outcome. The prompt is explicit that returning nothing beats inventing a pain point to fill a slot, because a fabricated signal poisons everything downstream.
Cluster
Near-identical complaints group into one problem.
One person complaining is an anecdote. The same friction described independently by fifteen people is a market. Clustering is what turns the first into the second.
Two paraphrases of one complaint often share only a third of their words, so matching has to be smarter than keyword overlap. A merge pass afterwards fixes the order-dependence that greedy grouping introduces.
Synthesise
Each cluster becomes one concrete product idea.
The model receives every pain point in the cluster plus revenue data for comparable companies, and returns a single product: the problem, the solution, the target customer, pricing, named competitors and the gap they leave, plus an ordered build plan.
One idea, not a menu. If the cluster is thin, the prompt requires it to say so and scope the idea narrowly rather than inflating weak evidence into a confident pitch.
Score
Seven weighted dimensions produce a 0–100 score.
Pain severity, competition gap, market size, monetisation clarity, build feasibility, evidence strength and distribution ease. All seven and their weights are shown on every opportunity page.
Evidence strength is the one dimension a model never touches. It is computed from how many independent complaints back the cluster and how many distinct platforms they came from, with cross-source agreement weighted more heavily than raw volume.
Right now
What is in the database today
Live counts from the database this site runs on.
Worked example
What one opportunity actually looks like
Fifteen complaints, four platforms, one product.
What we found
“I spend the first three days of every month copying numbers off PDFs.”
“No way to import supplier invoices without re-keying every line.”
“Looking for a VA to enter invoice data into QuickBooks weekly.”
+ 12 more across four platforms
What it becomes
Invoice line-item extraction for bookkeepers
Pull line items from supplier PDFs straight into QuickBooks, with a review step for the rows the parser is unsure about.
Questions about the method
Does a language model invent any of this?+
It summarises and judges, but it never invents the underlying evidence. Every opportunity traces back to documents you can open, and every pain point carries a verbatim quote copied from the source. If our summary and the original disagree, the original is right.
How fresh is the data?+
Connectors pull continuously and the pipeline re-runs on new material. Every piece of evidence carries the date of the post it came from, so you can see for yourself whether a complaint is current or three years stale.
What stops one loud community from skewing everything?+
Evidence strength weights breadth over volume. Forty complaints from one forum score lower than ten spread across four platforms, precisely because single-community volume is the easiest signal to mistake for a market.
Why do some opportunities have no competitors listed?+
Because we would rather show an empty list than a plausible-sounting name that does not exist. An empty competitor list is usually a warning rather than an opportunity — it often means nobody has found a way to make money in that space.