What Is a Data Broker? What Marketers Should Check

The short answer: a data broker is a company that collects personal information about consumers from sources other than the consumers themselves, compiles it into profiles, and sells or licenses access to other businesses. The defining feature is the missing relationship. A broker never sold those people anything or signed them up for anything; it assembled what it knows from public records, licensed third-party files and other commercial sources. For growth and performance marketers, that matters because every audience, append or enrichment file bought from a broker inherits the broker's sourcing, matching and refresh habits. The formal definition and the state registration rules live on the data broker glossary page. This guide covers the buying side.

The part most marketers miss

"Data broker" describes a business model, not a quality level.

Two brokers can both be registered, both operate legally, and still sell files that perform completely differently in a campaign. Registration tells you a company has disclosed what it does. It doesn't tell you whether its records are current, how it decided that two rows are the same person, or what happens to a record after someone opts out. Those three things decide whether a paid audience reaches real people or burns budget on bounces and mismatches, and they're the questions most vendor calls never get to.

No wonder so many teams write off broker data after one bad file. Usually the file was the problem, not the channel.

Where broker data comes from

Most consumer data brokers draw on a familiar mix of inputs:

• Public records, such as property records and other government files that are open by law.
• Licensed third-party files from other data companies, aggregators and identity graph providers.
• Commercial and transaction sources, where a business licenses data it collected in the normal course of operating.
• Online signals, for brokers that run website tags or pixels and link visits back to known profiles.

Exact Match's consumer database, for example, is assembled from licensed third-party and public-record sources. A broker that can't name its source categories in plain language is a broker to skip, even if the sample file looks good.

How raw records become something sellable

Raw inputs are messy. The same person shows up with three spellings of a name, an old address and two phone numbers, one of them disconnected. The real product a broker sells is the compiling work that happens next:

1. Identity resolution: deciding which records belong to the same person or household.
2. Attribute building: attaching demographic, behavioral, interest and intent fields to each resolved profile.
3. Refresh: re-running the process as source files change, people move and records go stale.
4. Suppression: pulling records that shouldn't be marketed to, such as deceased individuals and consumers who've opted out.

Step one is where file quality is won or lost. Deterministic matching links records only when several identifiers agree (name, email, phone, address). Probabilistic matching estimates a likely identity from partial signals and returns its best guess. The second approach can produce a bigger file that's quietly attached to the wrong people, and a wrong match looks just as complete as a right one. The post on how to append demographic data to a CSV shows how that difference plays out on a real upload.

How brokers sell it now

Delivery has changed more than sourcing has. The older model was a static list: choose counts and selects, receive a file, use it for a set period. Plenty of brokers now sell access instead of files.

API access is the fastest-growing of the three. The AI agent audience data API post walks through what an agent needs from that kind of connection, and the AI-native data enrichment tool post covers how enrichment changes when it runs inside an agent workflow.

Exact Match sells the platform-and-API version. One identity graph holds 250M+ verified U.S. consumer profiles plus 60M+ business records, organized into 80,000+ targeting clusters across 9 data domains. Matching is deterministic, and the data refreshes daily. Four products share that graph: Audience ID builds and exports segments, Clean ID cleans and enriches uploaded lists, Predict ID surfaces consumers who are actively in-market right now, and Site ID identifies 25–40% of verified human website visitors (bots excluded). The same data is available through an MCP server, so it can be queried from Claude or Claude Code.

Seven questions to ask before buying broker data

1. What are your source categories? Expect public records, licensed third-party data and commercial sources, named plainly rather than waved at.
2. Are you registered where the law requires it? State registries, including California's and Texas's, are public, so this takes a few minutes to check yourself.
3. How do you match records? Ask whether a match needs several agreeing identifiers or accepts a statistical best guess.
4. How often does the data refresh? Cadences across this market range from daily to quarterly or slower. Get the actual schedule, not "regularly."
5. How do opt-outs and deletions reach exported files? Brokers have to honor opt-outs within the time applicable law requires. In California, the Delete Request and Opt-out Platform (DROP) lets a resident send one deletion request to every registered data broker, and since August 1, 2026 brokers have had to process those requests, checking the platform at least every 45 days.
6. Which uses are prohibited? Read the acceptable use policy before writing the campaign brief, not after. Exact Match's Acceptable Use Policy, for instance, prohibits housing, employment, credit and insurance targeting unless Exact Match has authorized it in writing (Section 8.4), and it bars building audiences around sensitive categories such as health.
7. What match rate will a given file get? Be wary of any blanket number quoted before the vendor has seen your data. Match rate depends on how many usable identifiers each row carries, so an honest answer starts with "send a sample."

Question seven is the one that separates serious vendors fastest. A broker that promises a fixed match rate sight unseen is selling the pitch, not the file.

What a good broker can't fix

Even a well-run broker can't rescue a thin input. If an upload carries only first names and partial addresses, the match rate will be low no matter whose identity graph it runs against, because there's nothing reliable to verify. Registration doesn't change that either: a registered broker with stale records is still a stale broker.

So the work splits in two. The broker owns sourcing, matching, refresh and suppression. The buyer owns the quality of what it uploads and the questions it asks before signing an order form.

For teams weighing Exact Match against other providers, pricing for the Unlimited plan is shared in a short consultation rather than published. Schedule a consultation to see the data run against a sample of your own file.