About Demands
What this is
Demands reads public complaints — forums, software reviews, app store ratings, job posts — and turns the repeated ones into scored, buildable opportunities. Every conclusion links back to the posts that produced it.
The premise is simple: the best evidence that somebody wants something is that they said so in public, unprompted, before anyone tried to sell them anything.
How it is built
Twelve connectors handle ingestion. A model extracts structured pain points from each document. Similar pain points cluster together. Each cluster is synthesised into one product idea and scored across seven weighted dimensions.
Six of those dimensions are judged by a model reading the evidence. The seventh — evidence strength — is computed arithmetically from how many independent sources back the cluster, specifically so it cannot be inflated.
The decisions that shaped it
- Show the evidence on every page, because a claim you cannot check is a claim you have to take on faith
- Publish the scoring weights, so disagreement can be about the weighting rather than about opacity
- Prefer an empty competitor list to a plausible-sounding invented one
- Log failed ingest runs publicly rather than hiding gaps in the data
- Keep the calculators free and ungated, because gating arithmetic is a poor way to build trust
Where it falls short
People who complain online are not a random sample of people with problems. The data skews technical, English-speaking, and toward industries with active online communities. A quiet industry is under-sampled, not problem-free.
We compensate by weighting cross-source agreement heavily, but the bias is real and worth knowing about. The limitations page in the documentation sets it out in full.