"Are we gaining or losing share?" is the question every category review opens with, and most answer it badly — from a panel that covers a sample, or from internal sell-in numbers that describe what you shipped rather than what sold.
Marketplace data answers it differently, because the marketplace already counted. Every SKU, every month, whether or not you sell it.
What "share" should mean here
Share of what the marketplace actually sold in your category, computed across full-category coverage rather than a sampled basket. That is a stricter definition than most brands work with, and it produces uncomfortable numbers — usually because the denominator turns out to be bigger than the one internal reporting had been using.
The uncomfortable number is the useful one. A brand at 12% of a category it thought it held 20% of is not a reporting error to be argued away; it is 8 points of category somebody else is serving.
The four things worth computing
Top ten brands by GMV in the most recent month, with share. The baseline. Everything else is movement against it.
How each share has moved since the first month in your window. Levels tell you where you stand; deltas tell you where you are going. A brand at 6% and climbing is a more urgent problem than one at 15% and flat.
Realised price per unit, indexed against the category average with the average at 100. This is what separates two brands with identical share. One is holding it at a 40% price premium; the other is buying it with discount. Those are different businesses and different threats.
Any brand whose share moved by more than a third either way. Big movers are where the story is, and they are easy to miss in a table sorted by size.
The check that stops you reporting an artefact
Before you present any of it, look at the row count per month. If one month is sharply out of line with its neighbours — half the rows, or double — that is a collection artefact, not a market event.
Categories do not lose half their SKUs in a month. Collection pipelines occasionally do. Reporting the second as the first is the most common way marketplace data produces a confidently wrong answer, and the fix is a single sanity check you run every time.
Say so in the output rather than smoothing it. "March looks anomalous and is excluded" is a defensible line in a deck. Silently including it is not.
Running it
Call the catalogue first to confirm your category and market exist and for which months — use its category strings verbatim, because an invented name errors and omitting the field exports everything. Then estimate, check the number, and only then submit for real.
For share work you want a top-N slice rather than the full category: the brands that matter are all near the top, and the long tail costs a great deal for very little. Six months of one Indonesian category at "top": 100 came to 600 rows and USD 37.50, measured. The same scope in full was 374,169 rows and about USD 7,015.
How far back you can look
Indonesia runs from November 2020 and carries 121 categories. Thailand, Singapore, the Philippines, Vietnam and Malaysia start November 2022 with twelve to fifteen each. Every market is currently complete to June 2026.
Five years of monthly history in Indonesia is enough to separate a trend from a season, which is the distinction most share conversations actually turn on.
Recipe 1.1 in the agent skill is this whole analysis as a prompt, including the anomaly check.