Skip to content

Deal economics

Direct data licensor business model vs data broker: a four-question test

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

A direct data licensor holds rights that run straight from the records' creator and licenses an agreed scope onward; a data broker collects or buys data, often about people it never dealt with, and resells copies on its own terms. Neither label settles the economics. Compare who owns the rights, who performs the labor, who supplies the tooling and who captures recurring value, with every metric normalized to the licensed scope.

License terms

Direct data licensor

This page covers direct data licensor business model vs data broker: a four-question test.Illustrative

Key takeaways

  • The line between a direct licensor and a broker is contractual role and relationship, not the word license: Vermont's definition and a federal rule both count licensing as brokering in some cases.
  • Four questions separate the models: who owns the rights, who performs the labor, who supplies the tooling and who captures recurring value.
  • Gross and net revenue reporting under ASU 2016-08 can make two intermediaries with identical economics look very different on the headline.
  • Normalize every metric to the licensed scope: revenue retained after creator payments, delivery hours, reuse and concentration.
  • Each model has real risks: brokers face growing regulation, while direct licensors carry bounded reuse, buyer concentration and creator opt-outs.

What is the difference between a direct licensor and a data broker?

The difference is contractual role. A direct licensor holds rights that run straight from the creator of the records, either because it is the creator or because it signed a license with the creator for a defined scope, and it grants rights onward only within that scope. A data broker collects or buys data from many sources, often about people it never dealt with, and resells copies on terms it sets.

The law draws the line on relationship and personal information, not on the word license. California's Delete Act defines a data broker as a business that knowingly collects and sells to third parties the personal information of a consumer with whom it does not have a direct relationship. Vermont's definition covers a business that sells or licenses brokered personal information, so calling a sale a license does not move a company outside it. The FTC's 2014 study of nine brokers found that consumers are often unaware the firms exist, because brokers seldom deal with them directly.

Whether a company licensing its own records counts as a broker is a supplier question, covered in the state-law guide linked below. This page asks the investor question: what each model owns, does and keeps, and how to compare them on equal terms without assuming one is always better.

The four-question test: rights, labor, tooling and recurring value

Four questions separate the models better than any label: who owns the rights, who performs the labor, who supplies the tooling and who captures recurring value. A fifth, who carries the liability, decides what happens when the first answer turns out to be wrong.

Read the table as definitions, not as a ranking. A broker that owns a compiled copy can resell it many times with little added labor, which can be a strong business where the law allows it. A direct licensor works within each creator's license, which limits reuse but leaves a documented chain of rights a buyer can audit. A marketplace pushes most of the work and the rights question to its sellers.

The four-question test: rights, labor, tooling and recurring value
QuestionDirect licensorData brokerMarketplace
Who owns the rights?Shared: the creator keeps ownership and grants a defined licenseThe broker, over its compiled copy, subject to the law on its sourcesEach seller; the platform holds none
Who performs the labor?The licensor: rights review, preparation and delivery for each scopeThe broker, mostly at collection and compilationMostly the sellers
Who supplies the tooling?The licensor, for review, preparation and deliveryThe broker, for collection, matching and distributionThe platform, for listing and transacting
Who captures recurring value?Shared with the creator under each licenseThe broker, across repeat sales of the same copySellers, with the platform taking a cut of each sale
Who carries the liability?The licensor and creator, through rights and privacy warrantiesThe broker, as far as its contracts and the law assign itMostly sellers, under platform terms
A matrix of four responsibilities (owning the rights, doing the labor, supplying tooling, keeping recurring value) against three models (direct licensor, data broker, marketplace), marked carries it, shares it or not its role.
The four-question test drawn from the comparison of roles. Definitions, not a ranking or measured data.

Who does the work in a transaction, step by step

In a direct license, the licensor does most of the delivery work itself; in a broker sale, most of the work happened before the buyer arrived. Following one transaction through its steps shows where the cost and the risk sit, and that is what a business model really is.

Steps

Who does the work in a transaction, step by step

A direct licensing transaction typically runs through the steps below. Each one is labor or tooling the licensor must pay for on every new scope, which is why its margins depend on how much of that work can be reused.

A broker's sequence is different. It collects or buys data first, compiles it, and then sells copies to many buyers, so its labor and tooling sit at collection and matching. A marketplace leaves most steps to the seller and supplies the listing, payment and sometimes delivery tooling. None of the three is free of labor; they differ in when it happens and whether it repeats with each sale.

Checklist · 6 items

  1. Source the rights: find a creator willing to license and agree the scope, meaning which records, systems, years and permitted uses.
  2. Review the rights: confirm the creator may license the records under its own contracts and notices.
  3. Prepare the records: de-identify, link and document them to the agreed scope.
  4. Deliver and support: ship the dataset, answer buyer questions and handle acceptance.
  5. Pay and report: pay the creator its share and keep the records that let a buyer audit the chain of rights.
  6. Honor ongoing duties: deletion, renewal, opt-out or termination terms that outlast the sale.

0 of 6 ticked. Tick each item as you go.

What public filings and rules show about each role

Public documents show each role clearly enough to compare, and one of them shows that labels give no protection. We read ten primary documents on 2026-10-26: a state privacy regulator's page, a state statute, a federal regulation, the FTC's broker study, an accounting standard, a public company's annual report, two company announcements, a model developer's data partnerships page and a model developer's training documentation.

Sampling frame: primary documents from regulators, legislatures, standard setters, a public company filing, company announcements and model developers that describe a data licensing or brokerage role in their own words. Excluded: vendor marketing pages, press coverage and any document we could not verify on its publisher's domain.

The example that challenges the headline is the law itself. Vermont's definition covers businesses that sell or license brokered personal information, and the Justice Department's rule on access to bulk U.S. sensitive data by countries of concern defines data brokerage to include the licensing of access to data where the recipient did not collect it directly from the individuals. A company can call itself a direct licensor and still be brokering under these texts if its datasets carry personal data about people it never dealt with. The relationship and the content decide, which is why the four-question test asks where the rights came from, not what the company calls itself.

What public filings and rules show about each role
DocumentRole it showsWhat it says about rights, labor or valueWhat it does not show
Reddit annual report for 2024 (SEC filing)Platform licensing data generated on its own serviceSubstantially all licensing contract value to date comes from two partners, and the deals may not be renewed or may renew on worse termsPer-record pricing or delivery labor
Shutterstock investor announcementsLicensing intermediary paying contributorsA Contributor Fund compensates contributors whose content is used, contributors can opt out of data deals, and OpenAI holds a six-year license to its libraries and metadataHow much of each license reaches contributors
OpenAI data partnerships pageBuyer contracting directly with organizationsTwo tracks, an open-source archive and private datasets, and no interest in personal or third-party informationPrices or volumes
FTC data broker report (May 2014)Data brokerNine brokers studied; consumers often unaware of them because brokers seldom deal with them directlyCurrent practices after 2014
California and Vermont definitions; 28 CFR 202.214The legal lineBrokering turns on the absence of a direct relationship; Vermont and the federal rule both count licensingHow any one company is classified

The comparison table: metrics normalized to the licensed scope

Compare the models only after restating every metric per licensed scope, meaning one dataset defined by its records, systems, years and permitted uses under one license. Headline revenue, customer counts and transaction counts mean different things in each model, and the table shows how to put them on one footing.

Methods note for the table: the metrics and normalizations are an editorial framework built on 2026-10-26 from the four-question test, FASB's principal-versus-agent guidance and the disclosures in the evidence set. They describe what to look for under each model's definition, not measured values for any company, and they have not been tested against audited financials.

The comparison table: metrics normalized to the licensed scope
MetricHow to normalize itWhat to look for: direct licensorWhat to look for: data brokerLabel trap
Revenue basisRestate to revenue retained after payments to creatorsGross license value if it acts as principal, with the creator's share as a costGross resale revenue with little owed to sourcesGross and net reporting make headline revenue incomparable
Creator obligationsPayments, opt-outs and duties owed per licensed scopeRecurring shares, opt-outs and deletion dutiesUsually nothing owed to the people describedA marketplace label can hide either royalties or their absence
Services laborHours of rights review, preparation and delivery per scopeHigh on a new scope; falls only as reuse risesLow per sale once data is compiledA platform label can hide bespoke delivery work
ReuseLicenses sold per acquired scope within its permitted usesBounded by each creator's licenseOften high: one compiled copy, many buyersHigh reuse can signal loose rights limits
ConcentrationShare of contract value from the top two buyers and the top two creatorsWatch buyer concentration, as Reddit's two-partner disclosure showsWatch dependence on a few upstream sourcesA few large deals look like growth and behave like renewals
Retained liabilityWarranties and indemnities per contract, and their capsRights and privacy warranties on delivered recordsWhatever its contracts and the law assignIndemnity terms decide who pays when rights fail

Do labels like platform or marketplace hide a services business?

Often they do, and two checks expose it. The first is revenue presentation. Under FASB's principal-versus-agent guidance in ASU 2016-08, a company is a principal if it controls the specified good or service before it passes to the customer, and then reports the gross amount it is entitled to; an agent reports only the fee or commission it earns for arranging the sale. Two intermediaries arranging identical licenses can therefore report very different revenue with identical economics.

Illustrative example: intermediary A controls each license before the buyer receives it and books the full license value as revenue, recording the creator's share as a cost. Intermediary B arranges the same licenses as an agent and books only its commission. A looks much larger on revenue and identical on what it keeps. Compare them on revenue retained after creator payments, never on the headline.

The second check is labor. If every new dataset needs a bespoke rights review and hand preparation, the business behaves like a professional services firm whatever it calls itself, and its margin depends on utilization. Ask for delivery hours per licensed scope over time. A licensor whose hours fall as scopes repeat is building leverage; one whose hours stay flat is selling services with a data label.

What could break the direct licensing model

Five developments could weaken the direct licensing model, and each has a public marker an investor can follow.

  • Rules can reach licensors. Vermont's definition counts licensing, and the federal rule on bulk sensitive data treats licensing of access as brokerage where the recipient did not collect the data directly. A licensor whose datasets drift toward personal data about people it never dealt with inherits broker duties, such as the California requirement that registered brokers process deletion requests through the state's mechanism from August 1, 2026.
  • Buyers can concentrate. Reddit's annual report for 2024 ties substantially all of its licensing contract value to two partners. A licensor selling to a handful of model developers carries renewal and pricing risk however clean its rights are.
  • Creators can pull back. Shutterstock lets contributors opt out of data deals; a licensor whose creators renegotiate or withdraw loses supply it cannot replace from a compiled copy.
  • Synthetic data can substitute. NVIDIA's Nemotron documentation says its models use synthetic data in post-training and that the pre-training corpus includes synthetic and rephrased sources. Every licensed category that synthetic data can match loses pricing power.
  • Liability can outgrow fees. A licensor that accepts broad indemnities for buyers' models and products takes on risks it cannot control; the indemnities guide linked below lists the ones to resist.

Where SourceX sits, and the questions to ask any licensor

SourceX sits in the direct licensing intermediary position, and the same four questions apply to it. As that intermediary, it helps companies turn records they pick into licensed AI datasets, and the labor sits with it: it sources and qualifies suppliers, reviews their rights, negotiates contracts, prepares the data, then delivers it and settles payment. On rights the order is fixed: the company signs the agreement and approves its scope and terms before any data reaches SourceX, and SourceX's later use, sublicensing and distribution of the de-identified dataset stay inside what that agreement permits. Rather than keeping stock, it sources each buyer's dataset from U.S. businesses on request, which is the opposite of a broker's compiled inventory.

That position carries the costs the table describes: labor on every new scope, reuse bounded by each company's agreement, and obligations that outlast the sale. SourceX has not published revenue basis, delivery hours, reuse or concentration figures; it shares company and commercial metrics with investors directly. Its raise, under Rule 506(c), is open only to accredited investors. Investors can request the investor overview and bring a specific diligence question, such as how it would report revenue retained after payments to companies.

Questions to ask any licensor, broker or marketplace, SourceX included:

Checklist · 6 items

  1. For each dataset sold last year, whose license did the rights come from, and can you show the chain?
  2. Do you report revenue gross or net, and what do you retain after payments to creators?
  3. How many delivery hours did a new licensed scope take, and how has that changed as scopes repeat?
  4. How many licenses have you sold per acquired scope, and what limits reuse?
  5. What share of contract value comes from your top two buyers and your top two creators?
  6. Which warranties and indemnities do you give, and how are they capped?

0 of 6 ticked. Tick each item as you go.

Methods note: the evidence set and what it leaves out

The evidence set is ten primary documents read on 2026-10-26 and verified on their publishers' own domains or repositories: the California Privacy Protection Agency's data broker page, Vermont's statute, the FTC's May 2014 report, 28 CFR 202.214 in the eCFR, FASB's ASU 2016-08, Reddit's annual report for 2024, two Shutterstock announcements, OpenAI's data partnerships page and NVIDIA's Nemotron documentation. We selected documents that describe a data licensing or brokerage role in the publisher's own words and excluded vendor marketing, press coverage and anything we could not verify at the source.

Not measured: revenue, margins, delivery hours, reuse or concentration for any company, including SourceX; how any company is classified under broker laws; and whether either model earns better returns. Each requires company records or legal analysis of specific facts. The statutes and rules are summarized as general information, and counsel should review any specific classification question.

Frequently asked questions

What is a direct data licensor?

A direct data licensor holds rights that come straight from the creator of the records, either because it created them or because it signed a license with the creator for a defined scope. It grants rights onward only within that scope and can show the chain of rights to a buyer.

Is a data licensing company the same as a data broker?

Not necessarily. State broker laws turn on whether a business sells or licenses personal information about consumers it has no direct relationship with, so a company licensing its own business records is in a different position from one compiling profiles of people. The label does not decide it: Vermont's definition expressly includes licensing, and specific facts need legal review.

Why do gross and net revenue matter when comparing data businesses?

Under ASU 2016-08 a company that controls a license before it reaches the buyer reports the gross amount, while one acting as an agent reports only its fee or commission. Two intermediaries with the same economics can therefore report very different revenue, so compare revenue retained after payments to creators.

Which business model is better, direct licensing or brokering?

Neither is better in general. Brokers can resell one compiled copy many times with little added labor where the law allows it, while direct licensors accept bounded reuse in exchange for documented rights. The comparison table and the risks section show what to measure for each.

Does SourceX publish its business metrics?

No. SourceX has not published revenue basis, delivery hours, reuse or concentration figures, and it shares company and commercial metrics with investors directly. An investor can request materials at sourcex.si/investors and ask how revenue retained after payments to companies is reported; any offering is limited to verified accredited investors.

Sources

  • California's Delete Act defines a data broker as a business that knowingly collects and sells to third parties the personal information of a consumer with whom it does not have a direct relationship; from August 1, 2026, registered brokers must process deletion requests through the state's deletion mechanism. Source
  • Vermont defines a data broker as a business that knowingly collects and sells or licenses to third parties the brokered personal information of a consumer with whom it does not have a direct relationship. Source
  • The FTC's May 2014 report studied nine data brokers and found consumers are often unaware of them because brokers seldom interact with consumers directly. Source
  • 28 CFR 202.214 defines data brokerage as the sale of data, licensing of access to data or similar commercial transactions where the recipient did not collect or process the data directly from the individuals. Source
  • FASB ASU 2016-08: an entity is a principal if it controls the specified good or service before transfer and recognizes the gross amount; an agent recognizes the fee or commission it expects to earn. Source
  • Reddit's annual report for 2024 says that, to date, substantially all of the contract value associated with its licensing revenue is derived from two of its partners and that the arrangements may not be renewed or may be renewed on less favorable terms. Source
  • Shutterstock says its Contributor Fund compensates contributors whose content is used in training and that contributors can opt out of data deals. Source
  • Shutterstock announced a six-year agreement giving OpenAI a license to its image, video and music libraries and associated metadata. Source
  • OpenAI's data partnerships page describes an open-source archive and private datasets and says it is not seeking sensitive or personal information or information that belongs to a third party. Source
  • NVIDIA's Nemotron FAQ says Nemotron uses synthetic data in post-training and that its pre-training corpus includes synthetic and rephrased sources. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies