Reading a report

A Vantage report answers one question: what changed, and what should you do about it? Every section is built to be acted on, not admired.

A report in full: the headline, the counted figures, the rating distribution, and the complaint themes ranked by how many customers raised them.
A report in full: the headline, the counted figures, the rating distribution, and the complaint themes ranked by how many customers raised them.
Reports are written once a week per store, and can also be generated on demand.
Reports are written once a week per store, and can also be generated on demand.

The headline

One sentence at the top summarizing the period. If sentiment fell sharply, that is the headline. If a third of your reviews were negative, that is the headline. If nothing much happened, it will say so — a quiet week is useful information.

The numbers

FigureWhat it means
ReviewsHow many reviews were posted in this period. A small number makes everything else less reliable.
Avg ratingMean star rating for the period.
SentimentMean sentiment on a −1 to +1 scale. Roughly: above +0.2 is positive, below −0.2 is negative, in between is neutral.
vs previousChange against the previous period. This is usually the most important number on the page.
Positive / neutral / negativeHow the reviews split.

Why sentiment and star rating can disagree

Star ratings are blunt. A reviewer who gives four stars and then writes two paragraphs about how the packaging arrived crushed is not a happy customer, and sentiment scoring catches that where the rating does not. When the two diverge, the text is usually telling you more.

The rating histogram

Under the numbers is a breakdown of how many reviews gave each star rating. Its shape tells you more than the average does: a product averaging 3.0 because everyone gives it three stars has a very different problem from one averaging 3.0 because half give five and half give one. The second is a consistency problem — usually manufacturing or fulfillment.

What customers complained about

The core of the report. Each theme shows:

Themes come from a fixed vocabulary. Shipping, packaging, build quality, durability, value, sizing, battery life, ease of use, instructions, support, missing items, listing accuracy, performance, noise and appearance. Using a fixed list rather than free text is deliberate: it means "shipping delay" this week and "slow delivery" next week are counted as the same thing, so trends actually hold up over time.

What customers praised

Same structure, opposite direction. Useful for two things: knowing what not to change, and knowing what to put at the top of your listing copy. If reviewers keep mentioning battery life unprompted, that belongs in your product photos.

SWOT

Strengths, weaknesses, opportunities and threats — derived from what your reviewers actually said, not from a template.

QuadrantWhere it comes from
StrengthsWhat customers consistently praise
WeaknessesWhat they consistently complain about
OpportunitiesFixable problems, and things customers ask for
ThreatsTrends that will damage the listing if they continue — a falling star average, rising returns risk

What to do next

A ranked list of concrete actions, each with a reason and an estimated impact. These are specific — "add a measured size chart to the listing images" rather than "improve the customer experience". The first item is the one to do first.

How this report was produced

At the bottom of every report, collapsed by default, is exactly which engine produced it, how many tokens it used and what it cost. Worth checking occasionally so there are no surprises on your provider bill.

The numbers are always real. Counts, averages, shares and changes are calculated from your review data — never generated. When an AI provider is connected, it only writes the prose. A Vantage report cannot invent a statistic.

Regenerating a report

Use Regenerate on any report to rebuild it — useful after importing more reviews for that period or changing AI model. It replaces the existing report and, if AI is configured, uses tokens.

When a report says "not enough reviews"

Below five reviews in a period, Vantage declines to draw conclusions rather than producing something confident and wrong. Nothing is broken; there simply was not enough to say. It will try again next period.