Gross merchandise value is a useful operating total. It is a poor loyalty instrument. Marketplace apps routinely post a “record GMV month” while the share of buyers who return inside a defined window quietly falls. Leadership celebrates. Supply feels busier. The base is thinning.
The mismatch is mechanical. GMV adds every paid baht in the period, including first-time promo baskets, one-shot category spikes, and sellers dumping inventory. Churn — or more precisely, failed second orders — lives in a cohort denominator that GMV never sees. If you only plot the total, a campaign that buys new buyers at a discount looks identical to a healthier repeat engine.
The split we ask for in week one
In Listing to Loyalty Studio we ask for three lines on the same chart: GMV, first-order GMV, and GMV from buyers whose previous paid order was inside a window you actually believe (often 28 or 45 days, category-dependent). When the first line rises and the third sags, you do not have a growth story. You have a substitution story.
Grocery apps in Bangkok see this during mid-year sales: free-delivery codes inflate first baskets; the 28-day repeat among those new buyers underperforms the organic baseline. Fashion resale sees it when a celebrity listing spike never converts into a second browse session. Neither pattern is visible if finance only exports a single GMV column.
Counter-metrics worth defending
Pick one counter-metric before you approve the next campaign: second-order rate among new buyers, time-to-second-order, or GMV from buyers with two or more paid orders in the last quarter. The point is not the perfect formula. The point is that a “win” must survive the counter-metric, or it is not a win in Marketplace App Analytics terms.
Seller-side GMV can lie the same way. A take-rate holiday that floods the catalogue with low-quality listings will raise GMV and raise cancellation and chat-latency complaints. If ops owns cancellations and product owns GMV, nobody owns the leak. Module 04 of the flagship exists because of that split.
What not to do
Do not “fix” the chart by lengthening the repeat window until the line looks kind. Do not average Thai payday spikes into a 12-month smooth and call the residue churn. And do not add twelve more vanity totals. Three honest lines beat a dashboard that cannot lose an argument.
If you want the workbook version of this split, the studio seat is described on the flagship page. If you only need a written leak ranking on one extract, look at Snapshot Audit.