A data broker is a business that collects information about individuals, households, or organizations and supplies that information, or services derived from it, to other parties. Brokers commonly combine records from multiple sources to produce contact lists, identity-verification services, advertising audiences, risk indicators, and searchable background information. Their defining role is intermediation: the people described in a broker’s products often have no direct relationship with the broker and may not know that it holds information about them.
Data brokerage connects the economics of information with data privacy, commercial decision-making, and data governance. It can reduce the cost of finding customers, verifying identities, and detecting fraud, while also making personal information available beyond the setting in which it was originally collected. Its consequences depend on the information involved, the reliability of the links between records, the customer’s purpose, and the legal protections surrounding collection and use.
The industry has no single worldwide boundary. Some legal definitions cover only businesses that sell information about consumers with whom they lack a direct relationship; other policy discussions include a wider range of information intermediaries. A broker may sell records, license access to a database, answer a narrowly framed query, or deliver an audience to an advertising platform without disclosing the underlying records.
Historical development and industry boundaries
Modern data brokerage developed from several older information businesses rather than from a single technological invention. Nineteenth-century mercantile reporting agencies assembled information about businesses and their owners to help suppliers assess whether to extend credit. The Mercantile Agency, established by Lewis Tappan in 1841, was an important predecessor of later commercial credit-reporting organizations. Consumer credit bureaus subsequently developed local files about borrowers, while directory publishers, mailing-list businesses, and public-record researchers maintained other forms of commercially useful information.
Computers changed the scale and organization of these activities. During the second half of the twentieth century, magnetic storage and database systems made it practical to maintain millions of records, standardize names and addresses, and select customers using combinations of attributes. Direct marketers could move beyond purchasing a magazine’s subscriber list to combining purchase histories, geographic characteristics, and modeled interests. Records collected for one commercial purpose increasingly became inputs into products for another. The development of the Internet and electronic commerce added browsing activity, online purchases, account registrations, and other digital sources to established offline files.
Mobile devices and automated advertising further expanded the industry. Advertising identifiers, application-generated events, and location signals allowed some intermediaries to describe activity across devices and physical places. At the same time, corporate data services developed products for identifying businesses, mapping ownership relationships, and maintaining sales contacts. Established firms such as Acxiom, Experian, and LexisNexis Risk Solutions illustrate different combinations of marketing information, credit-related operations, identity services, and public-record analysis. These activities should not be treated as legally or operationally interchangeable merely because they occur within a related corporate group.
An influential 2014 report by the U.S. Federal Trade Commission examined nine data brokers and grouped their products broadly into marketing, risk mitigation, and people-search services. That classification remains useful because it distinguishes purposes rather than assuming every broker sells the same kind of file. An advertising intermediary, a background-information website, and a fraud-prevention provider can all mediate access to information, but their customers, accuracy requirements, and legal obligations differ.
The boundary also depends on who determines the purposes of processing. A company that hosts a customer database solely under that customer’s instructions is not necessarily functioning as a broker. Conversely, an intermediary need not transfer a downloadable database to perform brokerage: allowing a customer to query information or select an audience can provide commercially valuable access. “First-party” and “third-party” data describe relationships between collectors, users, and individuals; they are not universal legal classifications or reliable indicators of data quality.
Sources, acquisition, and the data supply chain
A broker’s source material can include public records, commercial transactions, surveys, website activity, application events, and information obtained from other intermediaries. Public sources may contain property ownership, business registrations, professional licenses, court filings, and other records made accessible under particular legal rules. Commercial sources can include retailer purchase histories, loyalty-program information, subscription records, and customer contact details. The availability and permitted reuse of these sources vary by jurisdiction, record type, and contractual terms; public accessibility does not by itself establish an unrestricted right to collect and resell information.
Data commonly move through a multi-stage supply chain. An organization that collects information directly may provide it to an analytics company, which standardizes the records and passes selected fields to another intermediary. A further business may combine these fields with its own database and license a resulting product. Brokers can acquire information through direct purchases, reciprocal exchanges, partnership arrangements, acquisitions of other companies, or services offered in return for access to contributing records. A record’s commercial value may therefore be realized at several different stages.
Digital collection introduces additional mechanisms. Websites can use tracking technologies to record page visits and associate them with browser identifiers. Mobile applications may contain software development kits that support advertising, analytics, or other functions and transmit information to outside providers. Depending on permissions and implementation, transmitted information can include device identifiers, interaction events, and location signals. Data exposed during automated advertising transactions can also provide information about devices or browsing contexts. These mechanisms are distinct: the presence of an analytics tool does not establish that every resulting event is sold onward.
After acquisition, brokers normalize formats, remove some duplicates, check field validity, and attach metadata. A telephone number may be converted into an international format; an address may be standardized against postal conventions; an outdated record may be marked with its last observation date. These operations make records easier to combine but cannot guarantee that the original information was correct. A validly formatted address can still belong to the wrong person, and a correctly recorded purchase may no longer describe that person’s current interests.
Provenance—the record of where data came from and how they were transformed—is consequently central to both quality and accountability. Meaningful provenance includes the source, collection date, permissions or other legal basis, permitted purposes, and downstream recipients. A supply chain can lose these distinctions when information is repeatedly aggregated. Contracts may prohibit onward sale or specified uses, but the practical effectiveness of those restrictions depends on monitoring, auditing, and the ability to identify where particular information has traveled.
Identity resolution, enrichment, and inference
Much of a broker’s value comes from linking records that appear to describe the same person, household, business, or device. This process, usually called identity resolution, is necessary because information sources rarely share a single dependable identifier. One file might contain a name and postal address, another an email address, and a third a device identifier. Linking them creates a more extensive profile than any source provides separately. The resulting structure is sometimes called an identity graph, although its nodes may represent different kinds of entities rather than verified individual identities.
Deterministic matching uses a shared identifier or a predefined matching rule. For example, records containing the same verified account identifier can be linked with relatively high confidence within that system. Probabilistic matching instead evaluates combinations of evidence, such as similar names, overlapping addresses, and patterns of device use. Machine learning can estimate whether two records refer to the same entity. Neither approach eliminates ambiguity: names are shared, telephone numbers are reassigned, households change, and multiple people may use one device or network connection.
Matching systems face a trade-off between false matches and missed matches. A false match combines information about different people; a missed match leaves records about the same person separate. The appropriate threshold depends on the application. An incorrect advertising association may waste a marketing impression, whereas an incorrect link in a tenant-screening or identity-verification product can affect access to housing or services. A high overall accuracy rate can also conceal weaker performance for people with common names, sparse records, or frequently changing addresses.
Enrichment adds attributes to an existing record. Some attributes are observed, such as a recorded transaction; others are inferred, such as an estimated income range or likely product interest. Statistical models may use neighborhood characteristics, purchases, or other behaviors to assign a person to an audience category. A geographic average is not an individual measurement, and a probability is not an established fact. Responsible interpretation therefore requires a distinction between observed information, model-derived estimates, and categories created for a particular commercial purpose.
Removing direct identifiers does not necessarily prevent linkage. Pseudonymization replaces identifiers with codes while preserving the possibility of associating records with an entity. Hashing an email address, for example, can facilitate matching between parties that apply the same procedure; it does not make the address intrinsically anonymous. Anonymization requires stronger protection against identifying individuals from the released information and reasonably available auxiliary data. Detailed movement histories or unusual combinations of attributes can be identifying even when names have been removed.
Products, delivery methods, and customer uses
Marketing products are among the most familiar forms of data brokerage. A broker may supply a contact list, add demographic estimates to an advertiser’s existing customer file, identify households likely to purchase a product, or create a digital audience segment. Segments can be based on recorded behavior, such as buying a category of goods, or inferred characteristics, such as an estimated likelihood of moving home. Geographic and contextual information may also be used without necessarily identifying a named individual. The product is often access to a selection capability rather than a complete dossier.
In digital advertising, a broker can match an advertiser’s customer records to identifiers recognized by an advertising platform. This process, often called onboarding, connects offline information with online delivery. Real-time bidding systems automate the sale of individual advertising opportunities, although not all brokered data are used in such auctions and not all auction participants are data brokers. Measurement products then attempt to associate an advertisement with an outcome, such as a purchase or visit. Such attribution is a measurement claim, not automatically evidence that the advertisement caused the outcome.
Risk and identity products serve different purposes. Banks, payment providers, insurers, and online services may use external information to check whether a name, address, telephone number, and device history are mutually consistent. An unusual combination can trigger additional review rather than an automatic rejection. Other products help locate debtors, identify duplicate accounts, or investigate suspicious transactions. Credit-reporting and screening products may overlap with brokerage, but their use in eligibility decisions can bring additional statutory duties. A product permitted for marketing is not necessarily lawful or suitable for assessing credit risk, employment, or insurance eligibility.
People-search services make information accessible through searches for a name, address, telephone number, or related identifier. Results may combine current and historical addresses, possible relatives, property records, and other public or commercially sourced information. Business-to-business services instead maintain professional contacts, firm characteristics, corporate hierarchies, and sales leads. Other brokers provide aggregated purchasing trends, mobility measures, or transaction-derived indicators to researchers, investors, and commercial planners. Aggregation changes the privacy and analytical issues but does not by itself establish that a dataset is anonymous or representative.
Delivery methods include bulk files, periodic database updates, web interfaces, and an application programming interface that returns a response to a query. Customers may pay for each record, each search, each successful match, each advertising impression, or a subscription. Some services return only a score or confirmation, limiting exposure of underlying fields. Others use controlled matching environments, commonly called data clean rooms, to permit specified analyses without unrestricted copying. Such arrangements can reduce some disclosure risks, but their protections depend on access rules, permitted queries, and controls on exported results.
Economic characteristics and market organization
Data brokerage exists partly because information is costly to collect, reconcile, and maintain. A firm seeking customers or checking identities could gather some information itself, but specialized intermediaries can spread those costs across many buyers. Brokerage can therefore reduce transaction costs and some forms of information asymmetry. A small merchant, for example, may gain access to identity checks that would be uneconomical to build independently. The broker’s contribution is not simply possession of raw records but their organization, coverage, timeliness, and integration into a usable service.
Information has unusual cost characteristics. Once a dataset and delivery system exist, supplying another authorized copy or query may have a low marginal cost, even though acquisition, security, compliance, and updating remain expensive. Broader coverage can create economies of scale, while reusing a common identity infrastructure across marketing, verification, and analytics can create economies of scope. Data are also generally non-rival: one customer’s use does not physically consume the information. Nevertheless, unrestricted resale can reduce exclusivity, and widespread use can change the behavior or environment the data describe.
These characteristics influence competition. A large historical database may be difficult for a new entrant to reproduce, particularly when it contains records no longer available from the original source. Exclusive contracts, established distribution relationships, and integration with customer systems can become barriers to entry. Network effects may arise where additional contributors improve matching or coverage and thereby attract more participants, but size alone does not demonstrate such an effect. Acquisitions and vertical integration can connect collection, identity services, advertising delivery, and measurement within one group.
Prices reflect more than the number of records supplied. Coverage of a difficult-to-reach population, freshness, verified provenance, permitted uses, update frequency, and expected commercial value can all matter. A list of names may be inexpensive compared with a continuously maintained service that returns a fraud signal during a transaction. Licensing terms commonly specify permitted purposes, retention periods, user numbers, and restrictions on onward transfer. These contractual rights differ from a simple transfer of ownership, and property rights in a database do not displace the legal rights of people described within it.
The market also generates externalities. A transaction between a broker and a buyer can impose privacy, security, or exclusion costs on people who are not parties to the contract. Conversely, shared fraud intelligence can benefit parties beyond the immediate purchaser. Economic assessment therefore cannot rely on transaction revenue alone. It must distinguish gains from better information from transfers between market participants and from harms borne by third parties. Estimating these effects is difficult because data flows are often opaque and the same information can support both beneficial and harmful uses.
Accuracy, privacy, security, and unequal effects
Data quality problems arise at every stage of brokerage. Source records may be incomplete or wrong; matching can attach information to the wrong person; old observations can remain in use after circumstances change; and models can convert uncertain evidence into overly definite labels. An address history may describe a previous resident, a purchase may have been a gift, and a device may be shared. Quality is therefore task-specific. A dataset adequate for estimating broad retail trends may be inappropriate for making an individual eligibility decision.
Correction is particularly difficult when records circulate through multiple organizations. Updating one broker’s file does not necessarily update copies already distributed to customers or other brokers. A deleted record may also reappear from a later source unless the system maintains an appropriate suppression mechanism. Effective governance distinguishes correction of an underlying fact from correction of an identity link, withdrawal from a marketing segment, and deletion of information. These are different operations, and each can require different changes across the supply chain.
Privacy risks are not limited to disclosure of names or financial account numbers. Combining otherwise ordinary observations can reveal routines, relationships, or visits to sensitive places. Location records can indicate where a device spends nights, works, or attends services; such patterns may allow a person to be identified. Inferences about health, financial vulnerability, or other intimate circumstances may expose information never explicitly provided to the broker. The central issue is often secondary use: information collected in one context is repurposed in another with different consequences and expectations.
Cybersecurity adds a separate dimension. Concentrated datasets can be attractive targets for unauthorized access, credential theft, or misuse by customers and insiders. The 2017 Equifax breach illustrates the consequences of compromising a large repository of personal information, although credit bureaus operate under obligations distinct from those of many marketing brokers. Security controls include access restrictions, logging, encryption, customer verification, and monitoring for unusual queries. Encryption protects against some forms of unauthorized access but does not prevent an authorized customer from using information for an improper purpose.
Unequal effects can result even without explicitly using a protected characteristic. Geographic patterns, purchase behavior, and other variables may act as proxies for group membership or socioeconomic circumstances. Errors can also fall unevenly across populations when underlying records differ in completeness. The use of algorithmic discrimination as an analytical concept concerns these differential outcomes and their mechanisms; whether a particular practice constitutes unlawful discrimination depends on applicable law and evidence. Establishing harm requires attention to the complete decision process, including which populations are represented, how scores are interpreted, and whether affected individuals can challenge the result.
Regulation in the United States
The United States has historically regulated data brokerage through a combination of sector-specific federal laws, general consumer-protection authority, and state legislation rather than through a single comprehensive federal broker statute. Legal duties often turn on the purpose for which information is assembled or used, not simply on whether a company calls itself a broker. The same organization can therefore operate products subject to different rules. Marketing information, credit reports, health records, and motor-vehicle information cannot be treated as one undifferentiated legal category.
The Fair Credit Reporting Act, enacted in 1970, is especially important where information is used for credit, employment, insurance, or other statutorily specified eligibility purposes. It regulates consumer reporting agencies, users of consumer reports, and furnishers of information. Its framework includes permissible-purpose restrictions, accuracy-related duties, consumer access and dispute rights, and requirements associated with adverse decisions. A company can fall within the statute because of what it does; a contractual disclaimer alone does not necessarily determine whether a product is a consumer report. Conversely, not every database containing personal information is covered by the Act.
Other federal statutes protect particular sources or sectors. The Gramm–Leach–Bliley Act of 1999 addresses financial institutions’ handling of nonpublic personal information. The Driver’s Privacy Protection Act of 1994 restricts disclosure and use of personal information from state motor-vehicle records, subject to enumerated exceptions. Federal health privacy rules apply to covered entities and their business associates, not automatically to every company holding health-related information. This distinction matters for information inferred from purchases, browsing, or app activity outside traditional healthcare settings.
The Federal Trade Commission can address unfair or deceptive practices under Section 5 of the FTC Act. In its 2014 data-broker report, the agency described extensive information collection and limited consumer visibility, and recommended legislation to improve transparency and control. Enforcement has also addressed particular data practices. In 2024, the FTC announced settlements involving X-Mode Social and its successor Outlogic, and separately InMarket, concerning location-data practices. The resulting restrictions included limits involving sensitive locations and requirements addressing consent and data handling. These actions concerned specified companies and conduct rather than establishing that every transfer of location information is unlawful.
States have added broader privacy rights and broker-specific requirements. Vermont enacted a data-broker registration law in 2018. California enacted a registration law in 2019 and its Delete Act in 2023, providing for a centralized deletion-request mechanism and additional broker obligations on a statutory timetable. California’s broader privacy framework also gives covered consumers rights concerning access, deletion, correction, and opting out of sale or sharing, subject to applicable definitions and exceptions. Registration is a transparency obligation, not a government endorsement of a broker’s accuracy or business practices.
European and other international legal approaches
In the European Union, data brokerage involving personal information is principally governed by the General Data Protection Regulation, which became applicable on 25 May 2018. The GDPR does not depend on “data broker” being a separate industry category. Instead, obligations follow from processing personal data and from the organization’s role as controller or processor. A broker determining why and how personal data are combined and marketed will generally have responsibilities as a controller for those operations, although legal roles must be assessed for each activity.
Processing requires a lawful basis and compliance with principles including fairness, transparency, purpose limitation, data minimization, accuracy, storage limitation, and security. Consent is one possible basis, not the only one. Reliance on legitimate interests requires an assessment of necessity and the interests or fundamental rights of affected individuals; it is not a blanket authorization for resale. Where data are obtained indirectly, Article 14 establishes information duties, subject to specified exceptions. These duties are especially relevant to brokerage because individuals commonly do not interact directly with the intermediary.
The GDPR provides rights of access, rectification, erasure in specified circumstances, and objection, among others. Objection to processing for direct marketing requires that processing for that purpose cease, including related profiling. Special-category data, such as information revealing health or religious beliefs, are subject to additional restrictions. The regulation also addresses certain solely automated decisions producing legal or similarly significant effects. These provisions should not be conflated: an advertising segment, an inferred sensitive attribute, and an automated eligibility decision raise different legal questions.
Other European rules operate alongside the GDPR. The ePrivacy framework addresses matters including access to information on user devices, while international transfers of personal data are subject to separate safeguards. The EU Data Governance Act, generally applicable from September 2023, establishes a framework that includes certain data intermediation services. Its provisions do not mean that all data brokers are regulated as the same type of intermediary, nor do they replace data-protection obligations. In the United Kingdom, the UK GDPR and Data Protection Act 2018 provide a related but legally distinct framework.
Outside Europe, approaches differ in institutional design, scope, and enforcement. Brazil’s General Data Protection Law establishes general rules for processing personal data, while China’s Personal Information Protection Law took effect in 2021 and imposes requirements concerning processing, sensitive personal information, and cross-border provision. These systems cannot be reduced to a uniform worldwide consent requirement. A brokerage transaction can involve the laws of the place where individuals are located, where collection occurs, where the broker operates, and where customers receive the information. The international character of the supply chain consequently creates overlapping obligations rather than a single global brokerage regime.
Accountability, measurement, and unresolved research questions
Effective oversight requires distinguishing the stages at which information is collected, transformed, and used. A source organization may be responsible for a misleading collection notice; a broker for an incorrect identity match; and a customer for an impermissible decision. These responsibilities can coexist. Contractual warranties help allocate duties between businesses, but they do not necessarily determine the rights of affected individuals or remove statutory obligations. Auditing therefore examines both legal documentation and actual technical behavior, including what fields are transferred, which queries are available, and whether restrictions are enforced.
Consumer-facing access and deletion systems address only part of this problem. Identity verification must be strong enough to prevent someone from obtaining another person’s file, yet excessive verification can collect additional sensitive information or exclude legitimate requesters. Deletion also differs from preventing future reacquisition. A suppression record may be needed to recognize that later-arriving data should not be processed for a particular purpose. Where lawful retention exceptions apply, deletion of some information may coexist with restricted retention of other records. The operational question is which data remain, why they remain, and what systems can still use them.
Assessing commercial effectiveness presents another challenge. A broker’s audience model may predict who buys a product without demonstrating that targeting those people increases sales. Customers already likely to purchase can make a campaign appear successful even when the advertising has little effect. A randomized controlled trial can help distinguish prediction from causal impact by comparing appropriately assigned groups. Identity-matching errors, missing transactions, and cross-device measurement complicate such evaluation. Similar problems arise in fraud systems, where a reduction in detected fraud may reflect changed detection rather than a genuine reduction in underlying attempts.
Industry-wide measurement is difficult because firms bundle brokerage with software, analytics, credit services, advertising, and other activities. Revenue figures based on different definitions are not directly comparable. Counts of profiles can be misleading when a database contains duplicates, household records, inactive identifiers, or modeled rather than directly observed individuals. More informative measures include coverage of the relevant population, update frequency, error rates for a specified task, and the ability to document provenance and permitted uses. Even these measures can vary substantially across customer applications.
Emerging technical methods alter the available controls without eliminating governance questions. Differential privacy can place formal limits on what certain statistical outputs reveal about individual participation, while secure computation can support selected analyses without exposing all source records. Neither method establishes whether the original collection was lawful or the analytical purpose acceptable. Advances in artificial intelligence also make it easier to extract attributes from unstructured material and generate new inferences. Evaluation consequently has to consider not only which records a broker holds, but which conclusions customers can derive from them, how reliable those conclusions are, and what consequences follow from their use.