When a reader finishes a book they loved, what actually brings them back to the shop? Not the overall star average, according to a growing body of marketplace research. The more interesting question for booksellers is whether a pair of specific, narrowly defined skill badges—say, "Reliable Recommender" and "Series Completer"—predict repeat purchases better than the familiar five-star aggregate. The evidence points toward yes, and the mechanism is behavioral, not cosmetic.
The Signal Problem With a Single Average
A five-star rating is a compressed summary of many different judgments: prose quality, physical condition, shipping speed, whether the ending landed. When a buyer scans that number, they are not learning anything they can act on. Worse, the average invites a specific cognitive shortcut. Kahneman and Tversky's work on loss aversion suggests buyers weight the possibility of a bad outcome more heavily than an equivalent good one, so a 4.6 average reads as "some people were disappointed" rather than "most people were pleased."
Badges work differently because they are earned through repeated, verifiable behavior rather than a single impression. A seller who has completed 40 multi-book series orders without a cancellation is not expressing an opinion. They are demonstrating a pattern. That distinction—opinion versus demonstrated pattern—is where the predictive power lives.
Why Two Specific Badges Beat One Broad Number
Narrow criteria force a real trade-off
Marketplace data from platforms like eBay and Etsy has long shown that structured seller metrics outperform free-text reputation for repeat purchase. The reason is partly mechanical: a badge with a clear threshold ("50 repeat buyers" or "zero unresolved disputes in 12 months") is hard to fake and easy to compare. A star average is neither.
Two badges create a profile, not a score
One badge can be gamed or misread. Two badges that measure different things—one about accuracy, one about follow-through—begin to describe a type of seller. Behavioral research on variable-ratio reinforcement explains why this matters for return visits: unpredictable but reliable reward keeps people checking back. A bookseller who reliably surfaces the right obscure title is delivering exactly that kind of intermittent payoff, and a badge that certifies the pattern makes the expectation explicit.
The Concrete Case: Powell's and the Staff Pick
Powell's Books in Portland has run staff-pick shelf talkers for decades, and the format is instructive. Each card names a specific employee and a specific reason—"if you liked Lincoln in the Bardo, start here." The card is a badge in miniature: attributable, narrow, and earned through having actually read the book. Customers return to the shelf talkers, not to the store's overall review score, because the card tells them something the average cannot.
Independent bookshops that have digitized this—tagging staff reviewers by genre and tracking which reviewers' picks get bought again by the same customer—report that reviewer-level attribution drives more repeat visits than store-level ratings. The badge is doing the work the star average was never able to do.
Designing Badges That Actually Predict
If you run a shop or a marketplace, the test for any badge is simple: can a customer use it to make a decision they couldn't make from the star rating? "Reliable Recommender" passes. "Top Rated" fails, because it is the average wearing a costume.
Start by identifying the two behaviors most correlated with your own repeat purchases—often something like completion rate (did the buyer come back for the next book in the series?) and precision (did the recommendation match the stated taste?). Then publish thresholds, keep them visible, and retire badges that stop discriminating. The forward-looking move is not more badges. It is fewer, sharper ones that describe a pattern a reader can trust enough to act on.