← AI Tools

LIVE · ~25–30s

Review Mining Agent

Clusters real App Store reviews into JTBD (Jobs-To-Be-Done) themes (churn, feature gaps, pricing, UX friction, bugs, praise) — no sentiment scores, just the actual reason behind the rating.

About this case study

Problem: manual competitive review analysis doesn't scale — reading hundreds of reviews and grouping them by hand takes hours, and sentiment analysis alone can't tell you whether a low rating means churn risk, a missing feature, a pricing objection, or a plain technical bug. Each of those needs a different fix. Approach: a pipeline resolves a brand name to its App Store listing and collects public reviews — this demo runs it against three preloaded apps — then classifies them (multi-label, not sentiment) against a fixed 6-tag JTBD taxonomy — churn, feature gap, pricing objection, UX friction, technical bug, praise — so the results are comparable across competitors instead of just "more negative" or "less negative." Result: a pilot across YAZIO, MyFitnessPal, and Noom surfaced three genuinely different competitive stories from what looked like the same negative-review noise — MyFitnessPal's pain was reliability, Noom's was billing/support, YAZIO's was missing features and UX friction. This demo runs the same live classification against real, pre-collected reviews — pick an app above and watch it cluster.

Methodology Note

Reviews are real, pre-collected from the public App Store RSS feed — a safe, stable dataset to validate the classification methodology on. The classification and paraphrasing run live against real AI.

Example: MyFitnessPal

Live access is available on request — I personally review each one.