Complaints in. Scored opportunities out.

The evidence comes from collected posts and reviews. AI summaries and scoring rules help organize it; they are hypotheses to investigate, not proof that a business will succeed. Counts on this page come from the database.

The pipeline

  1. Collect

    Nine collectors read public complaints on a schedule — app store reviews, Hacker News, Product Hunt, GitHub issues, Stack Exchange and configurable forum feeds. Only text matching a pain phrase is kept, so “great app, five stars” never enters the corpus.

    68,506 collected
  2. Extract

    A language model reads each one and answers a fixed set of questions: is this a real unmet need, who has it, what tools were named, what is broken about them, is there any sign of willingness to pay. Generic venting is discarded here.

    35,658 pain points kept
  3. Cluster

    Each pain becomes a 384-dimension embedding, and pains within 0.73 cosine of each other are grouped. That threshold was chosen by reading actual pairs: below it, different products get merged; above it, the same complaint written two ways stays apart.

    2,303 clusters with repeat evidence
  4. Score

    Demand, pain intensity, supply gap and feasibility, weighted 30/25/25/20. The result is then scaled down when a cluster rests on few mentions — three of those four inputs come from the model's reading of the members, so at one member they are one opinion of one post rather than a measurement.

    0–100

Live, by source

Where the evidence comes from

34,710App Store reviews51%18,056Play Store reviews26%7,431Product Hunt11%5,093Hacker News7.4%1,399firefox2.0%858GitHub issues1.3%815wordpress1.2%125discourse0.2%19Stack Exchange<0.1%

What the badges mean

Rising
Mentions grew more than 40% over the last 90 days.
Broken incumbent
One product is named by most of the cluster, and complained about. A market with an owner people dislike.
Fragmented
Three or more products named, none dominant. A market with no leader to displace.
Void
Demand is present and almost nobody names an existing tool.

What this method cannot do

It finds complaints that repeat. A brilliant idea nobody has complained about yet is invisible to it, and always will be — that is a property of the method, not a gap we intend to close.

Most evidence is mobile app reviews, because that is where dissatisfied users repeat themselves at scale. Expect markets with existing paid software, not greenfield.

Scores are estimates from public text. A language model summarises each complaint and can misread one. Every opportunity links to the original posts precisely so you can check rather than trust.

See the opportunities Read the FAQ