"Analyze customer reviews with AI" can mean almost anything — a one-paragraph summary, a star-rating average, a word cloud. Review X-Ray means something specific: every review is individually classified across six dimensions, and the results are tallied in code before anything gets written up.
What gets extracted from every review
- Category — which topic the review is actually about (pricing, support, quality, shipping, and whatever else the data shows).
- Sentiment — positive, neutral or negative, judged on a real three-point scale.
- Explicit request — does the review directly ask for a feature or change?
- Implied need — does the complaint point at a gap without asking for anything by name?
- Actionability — is the review specific enough to act on, or too vague ("terrible," "love it") to be useful evidence?
- Severity — for negative reviews, how serious: a minor annoyance, a real frustration, or a dealbreaker.
Why six dimensions instead of one sentiment score
A single sentiment score collapses a lot of useful information into one number. "60% negative" doesn't tell you whether that's six hundred people mildly annoyed about shipping speed or sixty people furious enough to demand refunds over broken products. Splitting sentiment from severity, and separating what customers explicitly ask for from what they only imply, is what turns a sentiment score into something you can actually act on.
From classification to insight
None of this is useful sitting as raw per-review data. Review X-Ray aggregates it into category breakdowns, a severity-weighted pain map, and ranked opportunities — each one backed by the real counts and real quotes behind it, never an estimate.
The model that writes the final report is only ever handed numbers that were already computed in code. It's never asked to count, and it's told explicitly not to invent a percentage or a quote that isn't given to it.