An ecommerce platform can lead by number of detected stores and trail by transaction value. Another can have a small installed base but a large share of high-traffic merchants. Both statements may be true because “market share” is not one statistic until the market, unit, and weight are defined.
This guide explains how operators should read platform statistics built from store counts, web traffic, or gross merchandise value. It does not publish a current vendor ranking. The numerical examples are hypothetical and designed to expose how weighting changes interpretation.
Table of Contents
- Define the population before the percentage
- Separate three valid questions
- Calculate the weighting effect
- Control detection and coverage bias
- Compare periods on a stable basis
- Use platform statistics in selection
- Publish a reproducible methods box
- The EcomToolkit view
Define the population before the percentage
Start by writing the denominator in plain English. It might be all detected ecommerce domains, active stores in a country, the top 100,000 retail sites by traffic, merchants above a revenue threshold, or marketplaces included in a named competitive set. Each creates a different market.
Then define the observation unit. One merchant can operate several domains, storefronts, countries, brands, or headless frontends. A domain share is not necessarily a merchant share. A technology can support checkout while a separate CMS renders content; detection rules must decide which layer qualifies.
Finally define the period. A live-site snapshot, monthly engaged visits, and annual transaction value are not directly comparable. Include the capture date, inclusion criteria, and whether the result represents a point in time or accumulated activity.
Separate three valid questions
Store-count share asks how common a platform is among the observed sites. Traffic-weighted share asks how much observed audience activity runs through those sites. GMV-weighted share asks how much observed transaction value is associated with them. The measures should not be substituted for one another.
| Weighting basis | Numerator | Useful interpretation | Main limitation |
|---|---|---|---|
| Store count | Detected eligible stores on platform | Installed-base prevalence | Treats tiny and huge stores equally |
| Traffic | Visits or engaged visits to those stores | Audience exposure in defined set | Traffic source and bot controls matter |
| GMV or turnover | Transaction value attributed to stores | Economic activity in defined set | Private, estimated, and definition-sensitive |
The OECD’s marketplace methodology provides a concrete example of denominator discipline: its published share calculation uses a platform’s engaged visits divided by total engaged visits across platforms in the defined scenario and month. See the OECD Similarweb data analysis methodology.
That traffic share does not become seller share or GMV share merely because the chart says “market share.” Readers need the full label.
Calculate the weighting effect
Assume a hypothetical census of 1,000 eligible stores. Platform A powers 500, B powers 300, and C powers 200. Store-count shares are therefore 50%, 30%, and 20%.
Now suppose the observed annual GMV is 200 million for A, 450 million for B, and 350 million for C. Total GMV is one billion, so GMV-weighted shares become 20%, 45%, and 35%. Platform A leads in installed-store count while Platform B leads in observed transaction value.
| Hypothetical platform | Stores | Store-count share | Observed GMV | GMV-weighted share |
|---|---|---|---|---|
| A | 500 | 50% | 200m | 20% |
| B | 300 | 30% | 450m | 45% |
| C | 200 | 20% | 350m | 35% |
| Total | 1,000 | 100% | 1,000m | 100% |
Neither result invalidates the other. Platform A may have broad small-business adoption; B may be concentrated among larger merchants. The correct measure depends on the question: ecosystem reach, enterprise relevance, developer opportunity, payment volume, or competitive exposure.
Weighting needs governance. The Office for National Statistics explains that weighting business responses by count gives each business presence in the population, while turnover-based ratio estimation gives larger-turnover businesses greater emphasis. Its Business Insights and Conditions Survey methodology illustrates why weighted and unweighted business statistics answer different questions.
Control detection and coverage bias
Public technology detection is imperfect. Custom domains can hide vendor signatures. Headless implementations may expose the CMS but not the commerce engine. Password-protected, regional, app-only, and wholesale channels may be absent. A vendor’s own customer total may include development, inactive, trial, or multiple stores under one merchant.
Measure precision and recall on a labelled validation set where possible. Precision asks how many detected stores truly use the platform. Recall asks how many true platform stores the detector found. Report unknown and multi-platform cases rather than forcing every site into one vendor bucket.
Coverage bias is separate. A detector can classify its visible universe accurately while that universe underrepresents China, mobile apps, private B2B portals, or small regional merchants. The platform detection precision and recall guide covers this distinction in more detail.
Traffic and GMV weights add their own missingness. Traffic providers model panels and crawls; GMV is often private or estimated. State whether missing weights are excluded, imputed, or assigned to an unknown category. Excluding every unweighted store can skew the result toward large public merchants.
Compare periods on a stable basis
Apparent platform growth can come from a larger crawl, a new detector, a changed traffic panel, or newly available GMV estimates. Preserve both the current best estimate and a comparable-series view.
For a fixed-panel comparison, retain stores observed in both periods and show entry and exit separately. For a current-market snapshot, use the latest eligible population but explain that share movement includes composition change. Do not splice the two approaches into one unlabeled trend.
When weights evolve materially, consider a linked index rather than applying today’s weights to every historic period without explanation. Official statistics use chain-linking to update weights while preserving a continuous series. The ONS chain-linking methodology provides useful background, though an ecommerce platform index still needs its own explicit design.
Show absolute counts beside percentages. A platform can gain stores while losing share if the measured market grows faster. It can lose detected sites while gaining traffic-weighted share if its remaining stores are larger.
Use platform statistics in selection
Market share is context, not a platform requirements document. High store-count prevalence can imply a large app and freelancer ecosystem. High GMV-weighted presence can signal experience at scale. Neither proves that the platform fits a specific catalog, checkout, B2B pricing, international, or operational model.
Use several lenses: comparable merchant adoption, agency and integration availability, release velocity, operational workload, data ownership, total cost of change, and required workflow depth. The mid-market platform selection guide provides a broader decision framework.
Avoid turning observed correlation into vendor capability. Large merchants may choose a platform because they already have engineering resources; their traffic does not prove the platform created that traffic. Conversely, a platform with many small stores is not inherently less capable.
For ecosystem strategy, segment by merchant size and geography rather than publishing one global rank. An integration vendor may care about addressable store count. A payments provider may care about processed value. An agency may care about merchant complexity and service spend.
Publish a reproducible methods box
Every platform-statistics article should include the population, unit, period, data source, detection rule, weighting basis, missing-data treatment, multi-platform rule, and update cadence. Include raw denominators and a confidence or uncertainty statement where supported.
Version the methodology. If classification changes, rerun history where feasible or mark the break in series. Store the input snapshot so a published figure can be reproduced after the live web changes.
Use precise titles. “Share of detected active stores in the sampled UK domain set, August 2026” is longer but safer than “UK ecommerce market share.” The title tells readers what the percentage can and cannot mean.
Separate measured facts from analyst inference. A detected-store count may be observed under a rule; the claim that it indicates ecosystem strength is an interpretation. Both can be useful when they are labelled correctly.
The EcomToolkit view
Platform share changes when the denominator or weight changes. That is not a statistical nuisance; it is the central meaning of the number. Publish store count, traffic, and economic activity as separate lenses, then connect the lens to the decision being made.
If vendor comparisons mix installed sites, traffic, revenue, and anecdotes into one ranking, request an EcomToolkit audit to build a reproducible platform scorecard.
