由 admin 
Marketplaces increasingly combine multiple datasets into one decision product. When several suppliers contribute, a naïve “pay by rows uploaded” formula rewards quantity, not commercial usefulness. A mathematically sophisticated formula can be equally misleading if its inputs are unverifiable. This Finndy Research model offers a practical allocation ledger for a hypothetical enterprise data product. Every amount is illustrative; it is neither a benchmark for typical marketplace take rates nor a statement that any contributor has a particular legal entitlement.
Start with the distributable revenue pool
A royalty is not meaningful until the contract defines the money being divided. Begin with invoice-level revenue actually collected during a reporting period, then subtract taxes collected on behalf of authorities, validated refunds and any explicitly agreed pass-through charges. Contractual platform fees and serving costs should be visible separately, rather than deducted through an undefined “net revenue” label. The royalty base should specify whether credits, bundled subscriptions and minimum commitments are recognized immediately or allocated over delivery periods.
Take an illustrative monthly $50,000 of collected receipts. Assume $3,000 in refunds and adjustments, $2,000 of tax or mandated pass-through amounts, and a separately agreed $10,000 platform operations allowance. The remaining $35,000 is an example contributor pool. This calculation does not imply that 70% is a fair standard share; it simply exposes which amounts need bilateral negotiation and reconciliation.
Score economic contribution without equating it to row count
A feed contributing a million duplicated rows can be less valuable than a small dataset supplying unique identifiers that make the entire product usable. Score contributors along separately evidenced dimensions: incremental coverage after deduplication, field validity, update timeliness, license breadth, continuity during outages, and measured downstream model or workflow performance. Publish definitions before suppliers see their own results; changing weights after a dispute erodes legitimacy.
For example, three contributors A, B and C receive assumed, audited contribution points of 50, 30 and 20. A proportional allocation of the hypothetical $35,000 pool produces $17,500, $10,500 and $7,000. Such a score is a contractual proxy, not a measurement of each supplier’s true marginal economic value. It should never be described as mathematically proven ownership of revenue.
Use marginal-value experiments carefully
Shapley-value thinking is useful because it asks how much the whole product gains when a contributor is added to different coalitions. But full calculations can be expensive and unstable when data sources overlap or users change behaviour. A pragmatic approximation runs leave-one-source-out tests on a fixed, rights-cleared evaluation sample, then reports the change in task completion, coverage and error rates. Document the sample, time window and uncertainty bands.
Avoid a single inflated “AI uplift” number. If removal of B reduces valid task completions by 3 percentage points on a narrow customer workflow, report that limited observation, not a claim that B generated 3% of marketplace turnover. Keep statistical measurement as an input into a negotiated score. As the OECD explained in its November 2019 work on data sharing, benefits of access must be balanced against governance costs and risks.
Put rights and legal restrictions before calculation
Not every supplier can authorize every downstream use. A contributor may permit internal analytics while prohibiting model training, onward resale or geographic transfers. A high contribution score must never override that restriction. Maintain a use-rights matrix by dataset, field and buyer purpose; block incompatible bundles before sales are booked. W3C DCAT 3 provides a standards-based vocabulary for describing distributions, versions and dataset series, which makes catalog identifiers and change tracking easier. It does not decide intellectual-property ownership or royalty entitlements.
The EU Data Act, generally applicable since September 12, 2025, affects defined access-and-use situations. It is not a blanket authorization to commercialize all accessible data. Legal review must distinguish statutory access from voluntary marketplace rights. Royalty agreements should name the jurisdiction, provenance evidence, sublicensing powers, dispute channel and retention obligations.
Maintain a correction and dispute ledger
A trusted payout system should be append-only at the accounting layer. Record original invoice, buyer entitlement, supplier mapping, calculation version, settlement date, chargeback and correction entry. Never silently overwrite last month’s score. A dispute can then be resolved by producing the assumptions and input evidence used at the time, not by rebuilding a spreadsheet from memory.
Consider holding back an illustrative 5% of each payout for 45 days where enterprise refunds or data-quality credits typically settle after consumption. The reserve rate and duration are negotiation variables, not externally verified norms. When the hold expires, release the balance with an immutable statement of any adjustments. A supplier should be able to reproduce its allocation using redacted aggregate figures without receiving other suppliers’ confidential rows or customer data.
Run a controlled shadow settlement before launch
For the first two months, calculate royalties in parallel with the pre-existing commercial agreement. Compare the formulas at invoice level, stress-test missing feeds, forced removals and disputed rights, and check whether a supplier can manipulate the score by uploading repetitive or low-quality records. Include customer value tests rather than ranking suppliers only by technical throughput. Publish a versioned methodology and provide an appeal route staffed by someone empowered to correct mistakes.
The success measure is not the sophistication of the allocation formula. It is whether the model rewards lawful marginal usefulness, explains the money, and survives a credible counterexample. If stakeholders cannot independently understand why A received $17,500 and C received $7,000 in the example, the allocation is insufficiently auditable for a durable data market.
Conclusion: Data marketplaces should sell verifiable, licensed outcomes and distribute economic rewards against evidence. A transparent proxy with revision and dispute controls is preferable to an opaque perfect-looking model. Publish the pool definition, rights matrix, scoring version and the correction log together; otherwise a royalty percentage merely hides a new argument.
Sources
- W3C Data Catalog Vocabulary Version 3, 22 August 2024
- OECD Enhancing Access to and Sharing of Data, 26 November 2019
- EUR-Lex Data Act summary, Regulation EU 2023/2854