Nestlé ran a mobile test with its agency WPP Media to find out which identity solutions actually control one of digital advertising's basic levers: frequency.
The work extends the advertiser's push to buy the open internet better, a program Julien Lamb, digital media lead at Nestlé, laid out in a recent episode of Media Buyers Club.
Nestlé wanted a clean picture of how several identifiers perform in live delivery. The question: which ones genuinely recognize a user, which ones lose that user along the way, and which ones can hold a cap of four exposures without drifting. Not the promise, the operational reality.
WPP Media built a mobile test campaign around five separate deals inside the curated marketplace of French adtech firm Equativ. Each deal carried one identifier – the third-party cookie, identity vendor ID5, French open ID First ID, telco-backed ID Utiq and The Trade Desk's Graph ID – and ran only on the French publishers already integrated with Utiq, roughly 100 domains.
"The goal was to compare how each type of identifier performs, whether it's a cookie, a deterministic ID or an ID mixing deterministic and probabilistic approaches, using Utiq as the yardstick, since it's 100% deterministic and tied to a mobile carrier contract," said Hoai Nguyen, data and tech director at Choregraph/WPP Media.
Close to 4 million impressions ran per line item, which makes the sample solid. No branding or performance goal was attached to the campaign, to avoid skewing it.
"The only objective was to deliver impressions while respecting a maximum cap of four exposures per user," Lamb said. Underneath sits a bigger question for an advertiser trying to rationalize its vendor list.
"It's hard to project with a solution that will sell me CTV retargeting when it already struggles to handle frequency capping across two browsers," Lamb said. For him, precision, not reach, is what should separate one identifier from another.
Look at the share of impressions delivered after the cap was hit and Utiq comes out well ahead, with 8% of impressions over the limit. No surprise, given it serves as the test's reference point. Still, the yardstick ID, and a deterministic one at that, also blew past its own ceiling.
"That can be explained by a slight lag between the moment the cap is reached and the moment the information reaches Equativ, or by ads showing simultaneously in the same auction," Nguyen said. In that window a few extra impressions get out, which accounts for a small share of overexposed users.
Identifiers with a probabilistic component, the hybrid IDs, struggle to manage repetition in cookieless environments. That wastes a quarter to a third of the advertiser's budget, in the sense that it keeps serving impressions to users who have in theory seen enough.
The clearest illustration is the number of users exposed to more than 15 impressions each, against a cap of four. "Because the observed frequency is averaged out, we even saw cases where the number of impressions hit a hundred for a single user," Lamb said.
A second finding: two of the three probabilistic IDs measured handle capping less well than the cookie in environments where the cookie is still available. "Maybe they'd be better off using the cookie when the signal is there," Lamb said. That is presumably what the third probabilistic identifier does, since it matches the cookie's performance exactly in cookie environments.
Next up: a test on Google's graph ID through DV360 and Amazon's through its DSP, and possibly an attempt to manage an average frequency of three or four rather than a hard ceiling, if that turns out to be technically possible.
Another option is the same test including Wi-Fi connections, which would challenge Utiq to tie a household-level identifier, built off the carrier contract, back to a targeted individual. That is no small task for the telco ID, and it is the only way to measure cross-device capping.
The fingerprinting-based solutions – everything measured except Utiq – were penalized in this first test because IP addresses are far less stable on mobile than on Wi-Fi. They will likely regain ground once desktop, and possibly CTV, enter the scope.
Their probabilistic approach and the restrictions around Utiq's ID may also penalize Utiq on reach, which is another factor when picking a shared ID.
This is only the start, according to Lamb. "We optimized campaign efficiency through SPO to take the shortest paths, and worked on attention to target the impressions that land hardest. We're now clearing the ground on useful reach: our ability to reach as many people as possible with controlled repetition," he said.
The next step, he added, is to "cross-reference those two dimensions to maximize both impact and coverage through efficient, useful reach." That tracks for an advertiser like Nestlé, which spends heavily on TV, where useful coverage is the core metric, and frequency control has been digital's historic promise.
Methodology
The setup had one goal: compare, under strictly controlled conditions, how the market's main identifiers perform on mobile.
The campaign ran on a single architecture: one media setup, deployed through five separate deals in Equativ's curation, each dedicated to a different identifier (Utiq ID, ID5, First ID, The Trade Desk's Graph ID, cookie).
Each line item carried an independent cap, so repetition could be measured per technology. For the cookie and the Graph ID, capping was handled on The Trade Desk's side; for the other three identifiers, on Equativ's.
RampID was originally in the plan and was dropped: The Trade Desk does not allow frequency control based on that identifier in the European Union.
To isolate each ID's own performance, every line item got a fully exclusive audience. No overlap was possible, which ruled out cross-delivery and accidental overexposure.
Activation was limited to Utiq identifiers from mobile connections, so that one ID equals one phone line and, with rare exceptions such as connection sharing, one individual. Utiq does not issue an ID where a user may have several IP addresses, such as VPN, Cloudflare or APR.
At the end of the campaign, the Utiq ID served as the deduplication key to measure real user-by-user repetition on each line item. Deterministic and fed directly by carriers, it rests on the IP address tied to the customer contract, which makes it a stable anchor.
The whole experiment ran exclusively on Utiq's partner inventory, around 100 domains that systematically carry a Utiq ID alongside the other identifiers. That removed bias tied to environment heterogeneity and ensured 100% of bid requests contained a common reference. Since the point was to judge precision rather than reach, it did not penalize the other identifiers.
