Correction — 12 September 2026: We have corrected the retained-sale count from four to three and the comparison denominator from 132 to 103. The earlier “6× lift” claim is withdrawn. These groups were neither randomised nor fully matched, so the comparison does not establish a causal effect. The original URL and publication date are retained.
We repriced eight listings in our own store and wanted to know whether it had helped.
Counting the sales afterwards was an obvious start. But some listings might have sold without a price change. To understand the result, we also needed to examine what happened to other stock over the same period.
Our first version called that a control group and reported a six-times conversion rate. On review, the wording and the calculation went further than the evidence could support.
Here is the corrected comparison, and why we changed it.
Start with retained sales
For listings repriced on 3 August, we examined the 30-day window ending on 2 September 2026, with 2 September excluded. The saved 11 September calculation reported:
| Group | Listings | Listings with a qualifying retained sale | Share |
|---|---|---|---|
| Repriced on 3 August | 8 | 3 | 37.5% |
| Observational comparison | 103 | 11 | 10.7% |
The earlier version counted four sales in the repriced group. One was subsequently cancelled. An order is evidence that someone placed an order; it is not retained revenue. For this outcome measure, that cancelled transaction does not count, whatever caused the cancellation.
The comparison denominator also changed from 132 to 103 under the corrected calculation. These are revised counts, not additional sales since the first article.
Each conversion means a listing had at least one qualifying transaction with positive retained merchandise revenue. It is not a measure of units sold or seller profit, and later reversals can require an update.
What the other group represents
The comparison was reconstructed from inventory records. It selected listings with no recorded markdown date, a recorded price below $200 and above zero, and an age of at least 90 days at the start. Listings recorded as sold before the window were excluded.
We did not randomly assign comparable listings to be repriced or left alone. One of the eight repriced items was above the comparison group’s price cap. Historical availability, price changes and other interventions were not completely recorded.
Those differences matter. The groups could have had different prospects before any price changed. A shared observation window does not make them equivalent.
That is why we now call this an observational comparison, and why we are not replacing “6× lift” with another lift claim calculated from the corrected percentages.
What we can take from it
The repriced group had a higher recorded retained-sale share during this window. That is worth investigating. It does not establish how many sales the repricing caused.
Eight listings are also a fragile basis for a broad claim: one outcome changes the group’s rate by 12.5 percentage points.
The other group’s 10.7% is not a reliable monthly forecast for all aged stock. We cannot turn it into a statement about an individual listing’s chance of selling next month. And choosing a target of thirty observations would not, on its own, settle the question. Selection and measurement quality matter as well as sample size.
A better next test
Before another comparison, we want to preserve the eligible listings and their starting state: prices, availability, quantities, traffic and the reason for each proposed change.
We also need to record concurrent changes to titles, photos, shipping or promotion. Otherwise we may credit a price change for a result that had several possible causes.
Where practical, randomly assigning eligible listings within relevant price and category groups would make the comparison stronger. If that is not practical, the selection limits should stay visible. Either way, the report must include non-sales, cancellations, refunds and missing evidence.
A consenting second seller’s isolated audit is the next useful test of whether the workflow transfers beyond our own stock. We would ask whether the recommendations were useful, whether the audit saved time, and what retained outcomes followed. A second seller alone would not prove repricing caused the sales.
For now, we have a small recorded comparison and a clearer measurement method. We do not have a demonstrated causal lift. That is the result TurnGoods can stand behind.
Read the companion analysis . These figures describe our own Still In The Box AU store; no external seller outcome has yet been measured.