An agentic app that reads the 714 public reviews, and ranks customer requests to build product roadmaps.

01 · Findings

What it produces

Each run below is a real corpus, collected and verified the same way. A company is added or removed by pointing the collectors at its public sources; nothing on this page is hand-written.

02 · Limits

What this cannot see

1. 01Only people with something to say write reviews. Satisfied customers and quiet cancellations look identical from the outside.
2. 02Review volume tracks marketing effort as much as product quality. Some of these reviews arrived in vendor campaigns, weeks apart.
3. 03Reddit skews to power users and detractors. Marketplace reviews skew to the administrators whom support had already helped.
4. 04G2 is excluded because its terms prohibit automated collection. Capterra blocks automated access, and some sources are English only.
5. 05One reader made every clustering call. There is no second rater yet, and the agreement check on the page exists so someone else can challenge it.

03 · Method

A ranked list is a guess until someone tests it

Ranking feedback is the part that demos well, and it is the easy half. The harder question is what evidence would change the ranking. So each company’s top opportunity carries a survey written to test it, a critique of that survey by a second pass, and the decision thresholds fixed before any answers exist.

1. 1 →
   Listen The feedback they do not own

   Public reviews and posts, collected within each site’s rules, stored word for word with the date, the rating, and the reviewer’s stated role. Names are not kept.

2. 2 →
   Ask A survey that can change the answer

   Every question is tagged to the part of the hypothesis it tests. A second pass removes the leading questions, the ones that ask two things at once, and any question no decision depends on.

3. 3 BuildIgnore
   Decide The rule before the data

   Thresholds are set in advance. If the answers come back one way the work moves to Build, another way and it drops to Ignore. The sliders show what each distribution would do. No responses are invented.