A platform market-share chart can make an ecosystem look either diverse or concentrated depending on how it groups smaller vendors. Ten platforms displayed separately tell a different visual story from the same ten platforms combined into one large other category. Squaring that combined category in a concentration calculation changes the result as well as the presentation.
EcomToolkit’s position is that platform statistics need an auditable population and grouping policy before they can inform a buying decision. This guide explains the Herfindahl–Hirschman Index, or HHI, through hypothetical platform distributions. It does not publish a current market ranking, infer legal market power or claim that concentration determines the best platform for a merchant.
Table of Contents
- Define the market-like population
- Calculate concentration without changing units
- Why the other category overstates concentration
- Treat unknown detections separately
- Choose vendor and ownership boundaries
- Compare periods on a stable basis
- Use concentration in a platform decision
- The EcomToolkit view
Define the market-like population
Begin with what is counted: detected live stores, merchant organisations, visits, transaction value or platform revenue. Each represents a different distribution. A platform serving many small stores can lead a store-count dataset without leading transaction value. The resulting concentration indices answer different questions.
Then specify geography, observation date, activity rule and coverage. A crawl of accessible domains is not necessarily a census of trading merchants. Several domains can belong to one business, and one domain can expose several commerce technologies. Decide how these situations enter the denominator before calculating any shares.
The US Department of Justice’s discussion of market shares illustrates why the selected market and measurement basis matter to interpretation. Here, HHI is used only as a descriptive summary of a defined technology dataset. A technology census should not automatically be described as a legally defined economic market.
Keep the distribution mutually exclusive for the calculation. If a store can belong to several platform categories, shares may sum above 100%. A conventional HHI applied to those overlapping figures loses its usual interpretation. Establish a primary-platform rule or choose a different descriptive measure.
Calculate concentration without changing units
HHI is the sum of squared shares. Using percentage shares gives an index on a scale up to 10,000. Using proportions gives a scale up to 1. These are equivalent only after the appropriate scaling; mixing them makes comparisons meaningless. The DOJ definition documents the percentage-share calculation.
Consider four hypothetical platforms with shares of 40%, 30%, 20% and 10%. The index is 40² + 30² + 20² + 10² = 3,000. For four equally sized platforms, it is 25² × 4 = 2,500. Both examples have four vendors, but the uneven distribution has greater concentration.
| Hypothetical distribution | Percentage-share calculation | HHI |
|---|---|---|
| Four equal platforms | 25² + 25² + 25² + 25² | 2,500 |
| Uneven four-platform split | 40² + 30² + 20² + 10² | 3,000 |
| One platform has every observation | 100² | 10,000 |
| Ten equal platforms | 10 × 10² | 1,000 |
These calculations describe distributions, not platform quality. They say nothing directly about checkout reliability, merchant satisfaction, development cost or switching difficulty. Avoid borrowing regulatory thresholds to label a technology dataset as safe, competitive or monopolistic without the separate analysis such claims require.
Why the other category overstates concentration
Suppose a chart shows platform A at 40%, platform B at 30% and other at 30%. Treating other as one firm produces 1,600 + 900 + 900 = 3,400. But if other consists of three independent platforms at 10% each, the correct platform-level index is 1,600 + 900 + 300 = 2,800.
The 600-point difference comes entirely from grouping. It does not reflect a change in adoption. Squaring a combined share introduces cross-products between vendors that should have been counted separately. This is why a presentation category is not necessarily a valid analytical entity.
If the remaining 30% is spread across ten equal vendors, its contribution is 90 and the total becomes 2,590. If those ten vendors have unequal shares, their contribution is higher. For a fixed total share spread across a known number of vendors, equal shares give the smallest squared-share sum.
| Treatment of the same 30% remainder | Remainder contribution | Total HHI |
|---|---|---|
| One platform | 30² = 900 | 3,400 |
| Three equal platforms | 3 × 10² = 300 | 2,800 |
| Ten equal platforms | 10 × 3² = 90 | 2,590 |
| Composition unavailable | Not exactly identified | Publish assumptions or bounds |
When the vendor count is unknown, do not manufacture a precise index. Show the known contribution, the unresolved remainder and a clearly described scenario range. A table with transparent limits is more informative than an exact-looking value calculated from a chart’s convenience labels.
Treat unknown detections separately
Other means identified vendors grouped for display. Unknown means the technology could not be assigned under the detection method. Combining them hides two different uncertainties. Unknown sites might belong to a major platform, an in-house system or several smaller products.
One option is to calculate concentration among detected sites only. Label that population explicitly and report detection coverage alongside it. The result is conditional on successful detection. It should not be presented as concentration across all stores unless that broader claim is justified.
Another option is sensitivity analysis. Allocate unknown observations under several transparent scenarios and recalculate the distribution. For example, test whether they follow the detected mix or disproportionately belong to one vendor. Those scenarios are not estimates unless there is evidence supporting their probabilities.
The platform detection guide explains why labels can be wrong as well as absent. A precise HHI computed from systematically misclassified stores remains a precise description of the labels, not necessarily of the underlying ecosystem.
Choose vendor and ownership boundaries
Technology brands and corporate owners are different entities. Two separately branded products may share an owner, while one product name may cover several operating models. Whether to combine them depends on the question being asked.
For developer skills and integrations, separate product brands may be useful. For dependence on a corporate supplier, ownership grouping may be more relevant. Publish both if they answer distinct procurement questions, and name them product-level and owner-level concentration rather than treating one as the uniquely correct value.
Maintain a dated mapping table. An acquisition can increase owner-level concentration without moving any store to another product. A product rename can change labels without changing either adoption or ownership. A methodology note should explain such breaks before readers interpret them as migration.
Keep self-built stores and technology combinations explicit. An undifferentiated custom category is often a collection of unrelated implementations, so squaring it as a single supplier can repeat the other-bucket problem. Depending on the research question, it may need separate treatment rather than inclusion as one vendor.
Compare periods on a stable basis
A change in HHI can result from adoption, coverage, classification or grouping. To distinguish them, preserve raw counts and the methodology version for each observation date. Recalculate historical periods under a common mapping when feasible, while retaining the originally published series for auditability.
Compare like populations. Adding newly discovered small stores can reduce measured concentration even if no existing merchant changes platform. Restricting a report to high-traffic stores can increase it because the sample now represents a different part of commerce.
The store-count versus weighted-share guide explains this denominator distinction in more detail. For concentration reporting, keep separate series for each basis. Do not splice a store-count index into a transaction-weighted series simply because both use the HHI formula.
Use concentration in a platform decision
A concentration measure can prompt useful questions about ecosystem dependence. Are your required integrations available across several providers? Can your team export critical data? How much of the operating process depends on one supplier’s roadmap? These questions require direct evidence beyond market shares.
A fragmented market does not guarantee easy switching. A merchant can be deeply dependent on one implementation even when many alternatives exist. Likewise, a concentrated ecosystem may have a substantial skills pool and mature integration options. Neither conclusion follows automatically from the index.
For a buying committee, include the population definition, known shares, treatment of the remainder and ownership policy on the same page as the result. Then evaluate operational fit separately using the platform total-cost model. This keeps a descriptive statistic from becoming an unsupported procurement score.
The EcomToolkit view
The calculation is easy; deciding what belongs in each share is the substantive work. An honest concentration report makes the other bucket, unknown detections and ownership boundaries visible. Those details often matter more than another decimal place.
If platform research is driving a migration or procurement decision, request an EcomToolkit audit to connect the statistics with your actual operating requirements.
