I took a data journey—here are the disturbing things I found

I don’t own an SUV, a Mercedes or a million-dollar home. I’m not 86. I don’t speak Spanish fluently and I’m not planning on having a colonoscopy anytime soon—despite what some data broker records say. Ad Age set out to find if information data brokers including Acxiom, Epsilon and Experian collect and share with marketers is actually worth the money. It turns out a lot of the data marketers receive from such data brokers is completely inaccurate, and when it’s not, it raises questions about whether brokers should be storing it. Check out the latest Ad Age events and award programs Over the course of six months, using what are called data subject requests (DSRs) from brokers, I collected the stockpiles of data and related classifications they’ve obtained on me. Such records are available under the California Consumer Privacy Act (CCPA), and not every broker will share them for people, like me, who live elsewhere (Ohio). But to their credit, some did. All of the data brokers are generalists offering a mix of data they collect directly, sometimes through surveys or tracking web behavior, and data they obtain through third parties, such as loyalty card programs, and sometimes, location tracking. The data is used for analyzing and targeting potential audiences with ads in digital media, TV and anywhere else it’s possible to match people with the data. Odd and contradictory Doing so, however, might be a fool’s errand. If one were to believe Experian records, for example, I am both a retiree born in 1938 with less than a high school education, as well as a Gen Yer/millennial with an 8-year-old student loan. Experian also showed me moving to a new address last year. Epsilon showed me as a Spanish speaker who owns a Mercedes and a house valued at $1 million. All of this is inaccurate. Interpublic Group’s Acxiom had me once living in Little Rock, Arkansas, a place I have never even visited, and also linked me to a few email addresses I’ve never used (including my daughter’s). It also gave me better than even odds of getting a colonoscopy (it’s unclear what that was based on, since Interpublic declined to say, though it’s probably based on age). Even where the data is accurate and from first-party sources, such as shopper loyalty programs and location records, it raises questions, particularly for Publicis Groupe’s Epsilon. That includes a set of location-based records that may have been covered by another company’s Federal Trade Commission deletion order and a set of data from its rival, Interpublic’s Acxiom (see below). The biggest question for marketers paying good money for this data and using it to reach advanced audiences at premium CPMs should be how accurate and useful it really is. Spanish-language mistargeting During the course of this investigation, I started noticing five Procter & Gamble Co. laundry and home care brands—including Tide—consistently serving me Spanish-language ads via MLB.tv’s app. Epsilon was the only one of the three brokers whose data I examined closely that had me classified as a Spanish-speaking Hispanic. A few weeks after I inquired with both P&G and Epsilon about the issue, I stopped receiving Spanish-language ads from P&G, and Epsilon deleted the entire collection of data on me from its Retargetly affiliate, which was responsible for their error. Publicis acquired Retargetly, which specializes in data from Latin America and the U.S. Hispanic market, in 2022 and merged it and its identity system with Epsilon’s. A visit at some point to NBCUniversal’s Telemundo.com, which showed up in my Epsilon record, may have been responsible for Retargetly pegging me as Spanish-speaking. But it’s unclear whether Epsilon was the source of the bad data that drove P&G’s targeting mishap. “The Spanish-language ad on MLB.TV had no connection to Epsilon’s digital DSR,” said a spokesperson for Epsilon. A P&G spokeswoman said the data behind it was corrected, but declined to specify the source. P&G buys connected TV via The Trade Desk. Epsilon’s sibling Publicis Media doesn’t handle P&G buying, but does work for consumer goods rivals such as Unilever and Kimberly-Clark Corp. Epsilon’s Conversant Media unit is a named reseller of MLB app inventory via three supply-side platforms, but it’s not clear that was part of P&G’s problem, either. In the Byzantine world of marketing data and programmatic media buying, many other potential sources could have been involved. Epsilon, ironically, was an early partner of TruthSet, which vets data from multiple brokers to improve accuracy among all of them by, for example, reducing misidentification of Hispanic audiences. Two years ago, that collaboration helped Heineken increase the number of Hispanic consumers who actually were Hispanic in its targeted campaign. TruthSet’s efforts also include powering TelevisaUniversal’s Hispanic data graph to help clean up widespread inaccuracies in classifications of Hispanic audiences. Epsilon’s strategic data reserve In all of this, there is also the question of how long brokers have the right to keep and distribute consumer data. Publicis Groupe’s Epsilon, for example, appeared to retain many years of supermarket purchase classifications, which include hundreds of packaged-goods buying (or lapsed buying) categories based on loyalty card purchase data that came via Circana, triggered by Kroger purchases. Making Epsilon’s data trove more interesting is its strategic importance to Publicis Media, by far the most successful media agency group globally in recent years and a big part of Publicis Groupe’s outsized stock performance. “The opportunity to connect data and media, as you have seen in our results, is delivering a lot of growth and allows us to win market share in a material way,” Publicis Groupe CEO Arthur Sadoun said during an investor call in July. Media measurement blog Tracking TV, social and digital updates Click here Sadoun was effusive about how the linkage of assets such as Epsilon with Publicis Media, tech platform Sapient, retail ad network CitrusAD and commerce analytics platform Profitero fuels growth. Epsilon, which also operates demand-side and supply-side platforms, has privacy disclosures to consumers giving it leeway to broadly move data acquired directly or via partners widely within Publicis. But that’s not the same as Publicis’ contracts with data partners and clients. And despite all those potential synergies, Publicis Media does not share Epsilon data from one client with other clients, spokespeople for the media agency and Epsilon said. Were it to do so, that would create a variety of potential conflicts, such as the possibility of sharing Kroger customer segment information with Publicis Media client Walmart or rival retailers in the CitrusAD network. Old client and a clean room Several data entries in my file appear linked to work Epsilon did for Kraft Heinz, such as one labeling me a buyer of Kraft Mac & Cheese and another labeling me a buyer of Kroger natural cheese. While Kraft left Publicis for Dentsu’s Carat last year, that data is still in my record. The data left over from work for Kraft was not first-party data from the marketer, but data segments pulled from Circana for Epsilon work on Kraft’s behalf, according to people familiar with the matter. That appears to differ from how another Circana partner, Experian, uses data under a 2023 deal for targeting connected TV ads for CPG clients. “Experian doesn’t license or retain Circana/IRI ProScores,” a spokeswoman said. “Experian leverages their identity products to facilitate a connection between IRI/Circana and select clients.” Circana (formerly IRI) gets sales data from supermarket checkout scanners and also from retailer loyalty card programs affiliated with the company. Circana also uses data from a consumer panel it shares with NielsenIQ, which tracks all of their purchases and uses the more in-depth data it gets from panelists to model the behavior of people who buy the same things. Mike Ellgass, executive VP of media, consumer packaged goods and retail at Circana, said buyers of the company’s data are required to delete it within 60 to 120 days of the end of the contract. But Circana and Epsilon declined to comment on when or if any of their contracts have ended or are set to end. And examination of my Epsilon records between February and September of 2024 shows no indication any data segments from Circana or IRI (the company’s previous name) had been deleted over a span of more than 200 days, while some are dated as far back as 2021. Also, Circana doesn’t sell data or analytics from the numerous loyalty card databases of supermarket or drugstore retailers it works with to Publicis Media client Walmart, nor are partners such as Epsilon permitted to transfer such data to Walmart, Ellgass said. While there’s no set contractual end date on how long Epsilon can store Circana data, older data is generally past its useful shelf life for CPG marketers after a year or two when they’re using it for media buying or promotional targeting, Ellgass said. Kroger’s terms for users of its loyalty card make it clear that the retailer can use such data with partners, but not that it can transfer data to them permanently. Technically, however, Circana and Epsilon aren’t storing raw transaction records, just classifications based on them. Epsilon didn’t have authorization from Kroger to label any of its audiences with a potential association with Kroger, said a spokeswoman for the retailer. “As we understand it, they intended to have this audience segment be ‘likely’ Kroger shoppers,” she said. Why Epsilon has Acxiom data Another question is why Epsilon has Acxiom data segments on its servers (Epsilon labeled the data as coming from Acxiom) given that Publicis and Acxiom’s owner IPG are rivals. Acxiom doesn’t currently sell data or analytics to Epsilon, said a spokesman for IPG, who directed questions to Epsilon about why and how it had access to the data. The Acxiom data in Epsilon’s database, which had me in various financial and credit card segments, was brought in by a client for their exclusive use in campaigns, said an Epsilon spokesperson. “Any data provided by a client remains for their exclusive use and will not be used for any other client,” according to an Epsilon statement. “All of these details are explicitly outlined in contracts with buyers. Our data security practices are verified on an annual basis by an external, third-party auditor since 2011.” In my data broker records, Publicis and Epsilon stood out as keeping the most data from the most places for the longest time. But agencies generally are like Star Trek’s Borg when it comes to data—assimilating everything they can, said one person familiar with the industry. An annual third-party review, such as what Publicis Groupe’s Epsilon does, provides limited visibility into how each client’s data is handled, and it’s paid for and controlled by the agency, not the clients, said Manuel Reyes, CEO of media audit firm Cortex Media. Auditing data governance requires a specialist firm beyond his firm’s scope, and he said client contracts vary widely in how much protection of data is written in, with bigger clients generally having more leverage to get better terms. Zach Srebnick, director of financial compliance at Cortex and a recent alum of Omnicom Group, said he believes agencies generally are abiding by solid data governance rules with clients and that clients generally are getting more sophisticated about how they protect data in contracts. “You’re not going to see clients’ data comingled with other clients publicly, and certainly not intentionally internally,” Srebnick said. “They might use it for modeling or put it into some sort of data-rama or proprietary dashboard. But rarely are you going to see first-party data received or shared with a client end up side by side with another client’s data.” Epsilon’s NinthDecimal data and the FTC Another interesting aspect from my February Epsilon data is from NinthDecimal, which compiled location-based data from mobile apps, before it was bought by InMarket in 2020. That data is linked to an enforcement action by the FTC earlier this year. In a section retained by Epsilon, NinthDecimal correctly pegged me as visiting multiple chain restaurants and retailers, for example, Subway, Papa Johns and Burger King. While I’m not a regular customer of any of these chains I’m sure I’ve been to each at some point, though I have no idea when or what app(s) tracked me when I was. Apparently, other consumers didn’t recall providing permission for such tracking either. In January, the FTC announced an enforcement action against InMarket for allegedly collecting consumer location data without proper consent; it finalized that order in May. The order, among other things, required retained location data to be deleted or destroyed. It’s not clear that the visitor segment data in Epsilon’s files was covered by the order. Regardless, by August, Epsilon had removed almost all NinthDecimal data segments from my file, except for one identifying me as a leisure traveler, though not to a specific place. “Epsilon does not have any data from NinthDecimal that was the subject of the FTC settlement,” said a spokesperson for the company. InMarket, in a statement, said, “All of InMarket’s solutions, including audience segments, are compliant with applicable industry standards and legal requirements, and our organization has implemented market-leading, privacy-forward standards for years.” An FTC spokesperson said the commission won’t comment on whether a company is complying with its order, including whether third parties that store NinthDecimal data are covered by it or whether the order treats segments differently than the raw data they’re based on. Complicating things further, InMarket is now closer to being a direct competitor of Publicis. In August it bought ChannelMix, a Kansas City, Missouri-based firm that handles analytics that shape media plans for a wide range of clients, with reported combined annual billings of $12 billion. More questions for brokers Epsilon’s data retention raises questions, but it also means the Publicis unit has a lot of accurate data, and it appears meticulous in labeling the partner sources of that data, while Acxiom and Experian don’t appear to do so in many cases. An Experian spokeswoman, on the apparent co-mingling of my record with that of an unknown 86-year-old, said: “Without being able to assess the record directly, the most likely scenario is that there is an error in the source data that contributed to this specific record.” Experian, which also runs a credit reporting business, is prohibited by law from co-mingling my credit report (which I believe is accurate) with data they sell for marketing. Good thing. Acxiom spurred a host of quality-related questions, including the wrong address in Little Rock, wrong email addresses and apparently modeled speculation about my sleep quality and interests in medicines and procedures (correct on allergy meds, wrong on the colonoscopy, and right at least in part on the likelihood of getting a flu shot, since I come up anywhere from 2 to 5 on a 5-point likelihood scale in different entries). An IPG spokesman said the incorrect Little Rock address and wrong email addresses, including my daughter’s, were not in the database used by clients. On how Acxiom knows about such things as my likelihood to get a colonoscopy or a flu shot, he declined to name specific sources but said the underlying data is “collected via licensing from vetted, reputable companies, and are not inferences created by Acxiom.” Acxiom’s data processing governance rules and logic, he said, “help us weed out the data points in the data we have collected that are no longer, or never were, relevant to/associated with you.” Wrong turns for marketers Even so, caveat emptor. One thing that stands out from all the providers is how much “data” they have about automotive history and intent, and how often it’s wrong. It’s all the more puzzling because all of them used historical vehicle registration records to create screening questions used when I accessed my data records in the first place. So at least they should know what I bought and own. On automotive data, I also found my LexisNexis file had an erroneous traffic conviction, which I went through the laborious process of correcting. That can cost you on insurance. My rate went down 2% after LexisNexis fixed the error. Other classification errors probably add up to a lot more wasted money for automotive marketers. Data brokers should know, for example, I’ve never owned an SUV, though none seem to. Despite numerous data broker inferences, I am never buying an SUV, a Lincoln or a Nissan Sentra or a Jeep (I’ve rented Jeeps, but one dropped a part in South Dakota and another stopped running on the road in California due to an overheated transaxle). Look back: Nearly half the data used for ad targeting is wrong Arielle Garcia, former chief privacy and responsibility officer of Interpublic’s UM and later director of intelligence for digital watchdog group Check My Ads, has done a similar investigation of her data records, and also found lots of erroneous and contradictory results. Her data profiles included a child she never had and classifications of her anywhere from below poverty level income to high net worth. From just one broker, she said she found herself in more than 500 audience segments, which she characterized as “a mix of wildly inaccurate, useless and questionable. I was in the market for just about every vehicle, even though I don’t drive. I was a defense contractor, but I also was in agriculture and foodservice.” She recommends marketers make similar requests for their own data records, particularly from vendors they or their agencies work with or intend to work with, before they spend money on data. Brokers or data providers that won’t comply should be ruled out, she said. Garcia said marketers often don’t know what data sources their agencies use, even though she believes that information should be specified in requests for proposal. Her main takeaway was that audiences built from such data are largely useless for marketers. And, while the pushback is that this will just empower big digital platforms Google and Meta due to their massive first-party data advantage over other media companies, that’s not the only possibility, Garcia said. For one thing, her look at Meta audience segments she fell into contained plenty of questionable conclusions, too. “We’re ignoring the incentive that they have to create as many inferences and put you in as many audience segments as they can,” she said. The better answer for marketers, she said, may be pulling back on using personal information in favor of drawing inferences where the signals are more solid and verifiable, such as media context. ~ ~ ~UPDATE AND CLARIFICATION: The story has been updated with comments from Epsilon in cases where the company previously had declined to comment on the record and to clarify the paraphrase of a comment from Circana Exec VP Mike Ellgass that Epsilon has the right to retain Circana data and consumer segments indefinitely.

I took a data journey—here are the disturbing things I found

Featured Stories