Abstract
Public companies must disclose what the figures they report actually mean. This makes annual filings a rare place where the definitions of marketplace metrics are available verbatim and can be checked.
Fifteen definitions were extracted from the fiscal 2025 annual reports of seven companies. Metrics with identical or near-identical names describe different quantities. Three things differ: the observation window, the unit being counted, and the event that qualifies for the numerator.
No company breaks any rule here. Each disclosed its definition, and each filing is internally consistent. The problem belongs to whoever places the numbers side by side.
Question
Do identically named marketplace metrics mean the same thing across companies?
Method
Selection: companies filing Form 10-K or 20-F with the SEC and disclosing marketplace operating metrics. The resulting set is Airbnb, DoorDash, eBay, Etsy, Fiverr, MercadoLibre, and Upwork.
Source: each company’s most recent annual report, filed in February and March 2026 for fiscal year 2025.
Extraction: passages containing an explicit definition of an operating metric were taken from the filing text. Quotations are verbatim, with no elision inside a sentence. Each record links to the filing.
Classification: every definition was coded by observation window, unit of count, and qualifying event. The coding is the author’s interpretation; the verbatim text remains the source of truth.
The fifteen-record dataset is available as CSV and JSON.
Finding 1. Demand-side activity
Five companies publish a demand-side activity metric. The names nearly coincide.
| Company | Name | Window | Unit counted | Qualifying event |
|---|---|---|---|---|
| Etsy | Active buyer | 12 months | buyer | completed purchase |
| eBay | Active buyer | 12 months | buyer account | paid transaction |
| Fiverr | Annual active buyers | 12 months | buyer | order placed, cancellations included |
| MercadoLibre | Unique active buyers | reported period | user | completed purchase |
| Upwork | Active client | 12 months | client | spend activity |
Three differences matter.
Window. Four companies use a trailing twelve months; MercadoLibre uses the reported period. Under quarterly reporting these are different quantities by construction.
Unit of count. eBay counts accounts rather than people: “all buyer accounts that paid for a transaction”. Etsy and Fiverr count buyers, MercadoLibre counts users. One person holding two accounts contributes differently under each.
Qualifying event. The sharpest divergence. Fiverr counts an order “irrespective of cancellations”, so a cancelled order still makes the buyer active. eBay requires a paid transaction. Etsy requires a completed purchase. Upwork requires spend activity.
Applied to identical user behaviour, these five definitions return five different numbers.
Finding 2. Volume
Six companies publish a volume metric under four different names: GMS, GMV, GOV, GBV. The names are not where the difference lies.
DoorDash (Marketplace GOV) includes taxes, tips, consumer fees, and subscription membership fees.
Etsy (GMS) excludes shipping fees and is net of refunds.
Fiverr (GMV) excludes value added tax and equivalents, service chargebacks, and refunds.
Airbnb (GBV) includes host earnings, service fees, cleaning fees, and taxes, net of cancellations and alterations.
MercadoLibre (GMV) excludes Classifieds transactions.
One measure includes taxes and membership fees, another removes shipping, a third removes taxes. Relating these figures to revenue, or comparing them across companies, requires a common base that the filings do not provide.
Finding 3. Take rate
Two definitions, one name.
eBay: “net revenues divided by GMV”. Fiverr: “marketplace revenue for such period divided by GMV for such period”.
The numerators differ — total net revenue against marketplace segment revenue. The denominators, as shown above, are built on different rules as well.
Finding 4. Comparability of a company with itself
One finding concerns not cross-platform comparison but a company against its own past.
In the notes to its metrics table, MercadoLibre discloses that as of 1 January 2024 the indicator “Unique Active Users” was replaced by “Fintech monthly active users” and “Unique active buyers”, and that as of 31 December 2024 the company no longer discloses “Number of items shipped”.
This is diligent disclosure: the change is recorded in the document itself. The consequence for a reader is unchanged, though. The historical series for those indicators is not continuous.
A single case does not establish that the practice is widespread. It establishes that it occurs, and that series continuity needs checking even within one company.
Alternative explanations
The differences may reflect the design of the platforms rather than arbitrariness. A cancelled Gig on Fiverr and a cancelled booking on Airbnb are operationally different events with different costs to the platform, and it is reasonable that they enter the metrics differently. Tips exist in food delivery and not in a handmade-goods marketplace.
This does not overturn the finding; it sharpens it. Differences can be substantively justified and still make the numbers incomparable. Whether a definition is well-founded and whether it is comparable with someone else’s are separate properties.
A second explanation is selection. Companies filing with the SEC may not be a neutral sample, and the differences could be an artefact of this particular seven.
What the study does not claim
It does not claim that any company discloses metrics improperly. Every definition was published by the company and is consistent within its own filing.
It does not establish which definition is correct. The question has no single answer: a definition serves a business model, not a comparison with a competitor.
It does not measure how far the numbers diverge. That would require raw data held only by the platforms themselves.
And it does not extend beyond the sample. Seven companies are seven companies.
Practical implications
Comparing marketplace metrics across companies without reconciling definitions is invalid — not merely imprecise, but invalid, because different quantities are being compared.
The definition of a metric should be stored under version control alongside its value. A change of definition invalidates the historical series, and without a recorded change this is discovered late and by accident.
Any benchmark figure should carry its window, unit of count, and qualifying event. Without them the number does not travel into someone else’s context.
Data
A fifteen-record dataset: company, ticker, filing form and date, metric name as published, metric family, verbatim definition, window, unit of count, qualifying event, and source link.
CSV and JSON, licensed CC BY 4.0. Version 0.1.
Next steps
Extending the sample beyond SEC filers. A retrospective across several years of one company’s filings to test how common definition changes are. Independent methodological review of the classification.