Anonym and the art of knowing less

In the spring of 2024, a football merchandise store had a familiar ambition: sell more shirts. UEFA’s Men’s Club Competitions Online Store, operated by Event Merchandising Ltd, was advertising on TikTok around the European club finals. It also had a condition. Fans’ personal information should not go directly to the platform. A campaign needs an answer to “Did this work?” A customer deserves some say over the route taken to get there.

Enter Anonym. The store used its Private Lift tool for a three-week European campaign. Staff uploaded data through a drag-and-drop interface; processing stayed in Europe. Results arrived within days. Mozilla’s case study reported 93% conversion lift. That measures the advertising’s effect. It does not mean buying Anonym produces a 93% sales improvement. The distinction is small enough to disappear in a pitch deck and large enough to matter to anyone paying the bill.

Two insiders change the terms

Brad Smallwood and Graham Mudd founded Anonym in 2022 after each spent more than a decade at Meta. Their experience covered marketing science, measurement and advertising products. They understood why marketers wanted behavioral data: it helped them make decisions. They also understood how that appetite had helped create an industry built around collecting and exchanging information about people.

Mudd’s explanation for leaving included entrepreneurial ambition and a market opening. In a later interview, he pointed to healthcare and financial services as examples of industries that adapted their use of data under stronger privacy constraints. Advertising could adapt too. In March 2023, he took that argument to Congress, supporting federal privacy legislation alongside technical protections. Better software, in his account, still needed incentives to change.

“We don’t create profiles or hold on to that data.”

Graham Mudd · AdExchanger Talks, March 2024

A vault with an exit policy

Anonym sits between advertisers and advertising platforms. Both contribute encrypted information. A trusted execution environment, or TEE, processes it inside protected memory. The company’s published architecture describes matching records using shared identifiers, running an approved advertising algorithm, releasing protected results, then destroying the temporary environment and its contents. Personal data is processed; the boundary concerns who can access it and what can leave.

The permissions are unusually concrete. Partners review documentation and code through a Transparency Portal. Before decryption, the system checks that the running code matches what they approved. Partners can revoke approval and inspect processing logs. In its December 2024 technical account, Anonym described using Intel SGX on Microsoft Azure. This is an infrastructure business, with a considerable amount of engineering behind a marketer’s upload button.

The answer can give you away

Encryption solves only part of the problem. A report can expose someone even when nobody sees the underlying spreadsheet. Anonym’s researchers give the example of overlapping groups: suppressing a small group does little good if a larger, visible group lets an observer work out the missing person’s behavior. Aggregation alone is an unreliable hiding place.

Anonym therefore adds differential privacy, which introduces calibrated statistical noise to protect individuals in the results. Its researchers tailor methods to the measurement task and the decisions an advertiser actually needs to make. Counts, purchase values and reporting dimensions all affect the design. Their expertise combines advertising algorithms with privacy mathematics. The practical lesson is refreshingly unfashionable: asking for every possible breakdown can make a report worse for privacy without making it more useful.

Three questions, three tools

Private Lift asks whether advertising caused additional action. Private Attribution connects ad exposures with outcomes, including view-through measurement. Private Audiences, announced in closed beta in April 2025, uses machine learning to find audiences resembling a brand’s valuable customers. These are familiar marketing jobs, offered through a different data-handling arrangement.

Mozilla reported an average 30% campaign performance improvement over broad targeting among early Private Audiences adopters. Treat that as a company-reported beta result, rather than a forecast for your campaign. Another public customer, German employment platform Zenjob, used Private Lift for a four-week TikTok campaign; TikTok reported a 38% increase in sign-ups. A useful business case starts with a defined outcome and a comparison you can explain.

Mozilla buys the middle

Mozilla initially discussed a partnership with Anonym. By June 2024, that conversation had become an acquisition. Laura Chambers told AdExchanger that shared missions and the opportunity to move faster drove the decision. The price remained private. LUMA Partners had previously reported an $8 million seed round in March 2023. Neither figure supplies a customer price: this is a business sold through commercial relationships, with no standard public tariff established here.

Distribution now supplies the next test. Snap announced a partnership in October 2025, followed by Pinterest in December. In April 2026, Reddit described expanding a limited pilot toward open beta for qualified advertisers. Its intended users have competitive, contractual or policy reasons for withholding first-party data from the platform. That is a specific buying problem: a brand may already want the advertising, but lack an acceptable way to measure it.

Start with the decision

Marketers choosing between Anonym, native platform tools and clean-room workflows should begin with the information they need to act. Attribution and incremental lift answer different questions. Data availability, platform participation and governance rules determine which workflow is feasible. The system cannot manufacture useful matching data or make a badly designed comparison persuasive.

There is also a deliberate limit. Anonym’s documented architecture rejects arbitrary queries and real-time data manipulation; operations must be programmed beforehand. That suits repeatable advertising jobs better than open-ended exploration. Its approach offers a principle other companies can copy: agree on the question, constrain the calculation, protect the answer, and discard temporary data. For a business built on measurement, knowing when to stop knowing is a useful skill.