Corpus at a glance
Dashboard
How much was fetched, how much survived into analysis, where it came from, and which kinds of buyer it resolves into.
Where the reviews came from
User segments
The answer, ranked · every figure computed in SQL
Findings
The five pain points and five themes that carry the most weight, each with a couple of verbatims.
Inferred from buying behaviour narrated in reviews.
The commissioned questions
Behaviours and needs
Why baskets stay narrow, what blocks category exploration, what people say they need before trying something new, and the jobs they are hiring the app to do. Where the corpus cannot answer a question, the tab says so instead of padding to five.
Buying behaviour · inferred, not disclosed
User segments
Kinds of buyer, derived from how people describe shopping — frequency, basket breadth, what triggers an order, whether they ever leave familiar categories. Shares are of reviews that carried enough signal to place, never of users.
Every number above is a link into this
Corpus
The full evidence base: all themes ranked, the granular open codes, every validated finding, and free-text search over every review.
| # | Theme | Reviews mentioning | Share | Severity | MoM | Source mix |
|---|---|---|---|---|---|---|
| loading themes… | ||||||
theme
Select a theme
Share over time
Each period against its own denominator, so corpus growth cannot masquerade as a trend.
Splits
Every split carries its own denominator.
Sub-themes
Root-cause hypotheses
Reviews give symptoms. Causation is inference — each hypothesis is tagged with what would actually validate it.