Data enrichment is the practice of adding attributes from an outside source to records a company already holds. An enrichment provider might take a retailer's file of customer names and postal addresses, match it against its own database and return age bands, estimated household income or lifestyle segments; for business contacts, it returns job titles and company size. The same operation happens inside programmatic auctions, where a publisher, supply-side platform (SSP) or curator attaches audience and contextual segments to a bid request before buyers see it.

The practice exists because first-party data is narrow. A company knows what its customers did on its own properties, not who they are, and a bid request describes an impression far better than the person looking at it. Enrichment sells the missing columns. Its central tension is that those columns describe people who never dealt with the company supplying them.

How a record is enriched

Offline, the client sends match keys: email addresses, phone numbers or name-and-address combinations, usually normalised and hashed with SHA-256. The provider matches them against a reference database, appends the requested fields with confidence flags and returns the file, which is loaded into a customer relationship management (CRM) system, a customer data platform or an ad platform.

A study presented at the Federal Trade Commission's (FTC) PrivacyCon 2022 shows the mechanics. Levi Kaplan, Alan Mislove and Piotr Sapiezynski of Northeastern University used a self-service "data append" interface that Experian offered until April 30, 2021. Attributes ran from age and sex to net worth and segments such as new parents, priced from $0.007 to more than $0.20 per record, with a $500 minimum. Each record came back graded from Non-Match to Person Match, and each attribute as Exact, Estimated or None. Appending birth years to 35,717 North Carolina voter records cost $517.94.

Delivery takes three forms. Batch files remain standard for direct mail. Real-time application programming interfaces (APIs) enrich a record as it is created, the model Clearbit built for business-to-business (B2B) sales teams; when HubSpot agreed to buy it in November 2023, Clearbit claimed more than 100 data points on 20 million companies and 500 million decision-makers. Clean rooms, the third form, join two parties' data without moving it, as when Nielsen Marketing Cloud data became available on InfoSum in August 2025.

Inside the bid request

OpenRTB, the real-time bidding protocol now stewarded by the IAB Tech Lab, has carried a Data object and a Segment object inside the user object since version 2.0, released for comment on June 30, 2011. Each Data object names a provider and holds segments, each with an id, a name and an optional value. The 2.0 specification's sample request sends "auto intenders" and an age value of "30-40" from "Data Provider 1". A community extension called segtax, introduced with seller-defined audiences, identifies the taxonomy behind segment IDs. According to Google's Authorized Buyers documentation, 4 denotes IAB Tech Lab Audience Taxonomy 1.1 and 7 Content Taxonomy 3.0.

On the web, most sell-side enrichment passes through Prebid, the open-source header bidding library. Its First Party Data Enrichment module fills fields such as site.ref, site.keywords and device.sua from the browser at the start of each auction, according to Prebid's documentation. Outside vendors plug in through Real-Time Data (RTD) submodules, a framework added in Prebid.js 2.39.0 in November 2019 alongside a Browsi provider. A submodule's getBidRequestData function writes segments into ortb2Fragments, while the publisher's auctionDelay setting caps the wait. Prebid expects tens of milliseconds. Vendors ask for more: Permutive's documented example uses 50, 152media recommends 500 and Stack Up's example sets 1,000.

Downstream, SSPs and curators add segments of their own, and demand-side platforms (DSPs) join requests to audience files advertisers have onboarded or bought. Third-party data is billed on top of media; advertisers using it have historically spent close to 20% of media costs on it.

From mailing lists to bid streams

The practice started with postal marketing. Acxiom, founded in Conway, Arkansas in 1969, built its business on mailing lists and in the late 1990s created AbiliTec to reconcile name and address variants into single records, the precondition for any append. The FTC's study of nine brokers, published on May 27, 2014, found that seven supplied data to one another, so an attribute could pass through several companies before reaching an advertiser.

Europe changed the terms on May 25, 2018. Article 14 of the General Data Protection Regulation (GDPR) requires anyone processing personal data not obtained from the individual to tell that person, within one month at the latest. Poland's data protection authority tested the rule in March 2019, fining Bisnode, a business data aggregator, 943,470 zloty (about 220,000 euros). Bisnode had emailed about 682,000 people but relied on a website notice for more than 6 million others drawn from public registers, arguing that letters would be disproportionate. The regulator disagreed.

In the bid stream, IAB Tech Lab finalised seller-defined audiences on February 24, 2022, letting publishers send taxonomy segment IDs without user identifiers, and renamed the specification Curated Audiences on December 16, 2024. B2B enrichment consolidated in parallel: HubSpot folded Clearbit into Breeze Intelligence, an enrichment product launched on September 18, 2024.

Why marketers care

Enrichment has grown more valuable as other signals have faded. Google's data strength framework names signal enrichment as its second step, asking for hashed emails and conversion values so automated bidding can price auctions. Experian's July 30, 2026 partnership with AUDIENCES leaned on a 72% figure for marketers struggling to move first-party data into activation.

Curation turned enrichment into a sell-side business. The German digital association BVDW's March 2026 whitepaper lists data enrichment among six curation use cases, adding contextual data, audience segments or performance metrics to inventory. Moving that step upstream is also a claim about addressability: Equativ put match rate losses between platforms at 40% to 70% in October 2024, a vendor figure that has circulated widely.

The input matters as much as the append. In a survey published by Firmable, a sales data vendor, on September 9, 2026, 222 B2B sales professionals estimated that 32% of their CRM records were inaccurate, incomplete or outdated. Those records become the customer lists sent to ad platforms.

Accuracy, legality and open disputes

Independent tests have been unkind. Nico Neumann, Catherine Tucker and Timothy Whitfield examined more than 90 third-party audiences from 19 data brokers in a study published in Marketing Science in 2019. Gender accuracy averaged 42.3%, below the roughly 50% a random guess would achieve. Against random selection, the profiles raised the chance of reaching a user with one desired attribute by between 0% and 77%.

Errors are not evenly spread. In the Northeastern study, 27% of Hispanic voters returned no match against 18% of white voters, and correct birth years came back for 51% against 64%. Experian questioned voter rolls as ground truth, according to the paper's appendix. Truthset, which sells accuracy scores, reported in its State of Data Accuracy 2026 study age segments accurate 13% of the time and gender segments 61%.

The legal position is contested. The UK Information Commissioner's Office (ICO) issued an enforcement notice to Experian in October 2020 over its offline marketing business. On February 20, 2023 the First-tier Tribunal largely sided with Experian, rejecting the ICO's view that Experian's privacy notice lacked transparency, while agreeing that data on more than 5 million people from public sources had been processed without notice. The Upper Tribunal dismissed the ICO's appeal on April 22, 2024. Italy's data protection authority took a harder line with a B2B provider, fining Lusha 2 million euros in a decision published on July 27, 2026 and finding that legitimate interest could not support selling enriched contact profiles.

Customers push back too. HubSpot's plan to pool customer CRM data into a shared enrichment dataset, set out on July 1, 2026, was withdrawn within days. In the bid stream, RTB House sampled about 135 billion requests enriched with seller-defined audiences and in March 2023 found 75% of user signals non-compliant. Every RTD module also draws on the pre-auction time budget used by consent and identity modules.

Not the same as

Data onboarding. Onboarding converts offline records into online identifiers so a list can be targeted. It makes a file addressable without adding attributes, which places Nexxen's July 2026 tool promising activation within 24 hours in this category.

Identity resolution. Identity resolution decides which identifiers belong to one person or household. Enrichment depends on it, but a resolved identity carries no new facts.

Data cleansing. Hygiene corrects fields already held, such as moved addresses. Vendors often sell both under one contract.

Curation. Curation packages inventory and data into deals; enrichment is one step within it. Data engineers also use the phrase for any lookup that adds fields to records in a pipeline.

Recent developments

July 2026 brought a cluster of enrichment deals. Nexxen brought Acxiom household and purchase data into Nexxen Discovery on July 21 without publishing match rates. Two days later Acxiom launched an Identity Boost Accelerator with Salesforce, again without a published match-rate benchmark.

The payload is changing too. IAB Tech Lab's AAMP 2.3, released on July 30, 2026, lets audience vector embeddings travel inside OpenRTB, and a Prebid module writes them into user.data. A string of numbers is a different kind of attribute from a taxonomy node.

Litigation continues. Florida's Attorney General filed a 66-page complaint against Netflix on September 9, 2026, alleging among other things that behavioural data flowed to Experian and Acxiom for data enrichment. Netflix has not been found liable. The case turns on the question enrichment has always raised: who agreed to the extra columns.

Timeline

  • 1969: Acxiom is founded in Conway, Arkansas as a mailing list business
  • Late 1990s: Acxiom creates AbiliTec to link name and address variants to single records
  • June 30, 2011: OpenRTB 2.0, including the Data and Segment objects, is released for comment
  • May 27, 2014: The FTC publishes Data Brokers: A Call for Transparency and Accountability
  • May 25, 2018: The GDPR applies, including Article 14 notice duties for data not collected from individuals
  • March 2019: Poland's data protection authority fines Bisnode 943,470 zloty under Article 14
  • November 2019: Prebid.js 2.39.0 adds the Real-Time Data framework with a Browsi provider
  • November 2019: Neumann, Tucker and Whitfield's field study of third-party audiences appears in Marketing Science
  • October 2020: The ICO issues an enforcement notice to Experian
  • April 30, 2021: Experian stops offering its self-service data append interface
  • February 24, 2022: IAB Tech Lab finalises seller-defined audiences, with the segtax extension identifying taxonomies
  • February 20, 2023: The First-tier Tribunal largely allows Experian's appeal
  • March 8, 2023: RTB House reports 75% of sampled seller-defined audience signals non-compliant
  • November 2023: HubSpot agrees to acquire Clearbit
  • April 22, 2024: The Upper Tribunal dismisses the ICO's appeal in the Experian case
  • September 18, 2024: HubSpot launches Breeze Intelligence for data enrichment
  • December 16, 2024: IAB Tech Lab renames seller-defined audiences as Curated Audiences
  • August 20, 2025: Nielsen Marketing Cloud data becomes available on InfoSum
  • March 3, 2026: BVDW publishes its programmatic curation whitepaper
  • July 1, 2026: HubSpot publishes Contact Discovery terms, withdrawn within days
  • July 21, 2026: Nexxen adds Acxiom household and purchase data to Nexxen Discovery
  • July 23, 2026: Acxiom launches its Identity Boost Accelerator; Truthset audiences reach Yahoo DSP
  • July 27, 2026: Italy's Garante publishes a 2 million euro fine against Lusha
  • July 30, 2026: Experian announces its AUDIENCES partnership; IAB Tech Lab releases AAMP 2.3
  • September 9, 2026: Florida files its complaint against Netflix; Firmable publishes its CRM data survey

Summary

Who. Data brokers and credit reference agencies such as Experian and Acxiom, B2B providers such as HubSpot's former Clearbit unit and Lusha, publishers, SSPs, curators and DSPs in the bid stream, and the advertisers that buy the output. IAB Tech Lab and Prebid.org set the technical formats; regulators including the FTC, the ICO, Italy's Garante and Poland's data protection authority set the limits.

What. The appending of attributes from an outside source to records already held, whether a CRM file, a customer list or an OpenRTB bid request. Offline it runs through match keys, reference databases and confidence-graded returns; online it runs through the Data and Segment objects, segtax values and Prebid's Real-Time Data submodules.

When. It grew out of postal list businesses such as Acxiom, founded in 1969, entered real-time bidding with OpenRTB 2.0 in 2011, came under GDPR notice duties in 2018 and has been reshaped since 2022 by taxonomy-based publisher segments, curation and, in 2026, audience embeddings.

Where. In batch files, real-time APIs and clean rooms for customer data, and in header bidding wrappers, SSPs, curation platforms and DSPs for programmatic media, across the United States, the UK and the European Union, where the legal treatment differs.

Why. First-party records and bid requests say little about the people behind them, and enrichment fills that gap for targeting, bidding and measurement. Independent studies show the added attributes are often wrong and unevenly accurate across groups, while regulators disagree over whether people who never dealt with the supplier can lawfully be profiled this way.