The other day, I was chatting to a Head of Trading who told me:

β€œWhat a nightmare all this audience fragmentation is!”

It's true we didn't have all these problems back in 2016!

Desktop ruled the roost πŸ‘‘
Cookies were flowing freely πŸͺ
No headaches with targeting / frequency capping / measurement! 😊
Those were the good old days! πŸ•š

But today, things are... a tiny bit more complicated. 🀯

Why has identity become so fragmented in programmatic advertising?

Particularly because of:

  • πŸ“± The proliferation of devices per person: mobile phones, connected TV (CTV), work computers, tablets, watches and even glasses.
  • πŸ›‘οΈ Browser privacy policies: especially Safari and Firefox, which restrict certain forms of cross-site tracking.
  • πŸ‘ Consent pop-ups and privacy rules: in France and Europe, the GDPR and ePrivacy-related rules complicate the lawful collection and reuse of tracking identifiers (the precise consent requirements depend on the operation).

Not exactly making life easy for our beloved digital advertising industry, is it?! πŸ’ΈπŸ’Έ

Different environments may use identifiers that do not naturally recognise one another. A browser cookie is not the same thing as a mobile advertising ID, a connected-TV identifier or an email in an advertiser's CRM. Without a permitted bridge, matching audiences or deduplicating exposure across those environments becomes difficult.

For a broader breakdown of disappearing addressability, see our analysis of the audience attrition chain.

What is an Identity Graph β€” and how does it connect advertising IDs?

And since nature abhors a vacuum, AdTech has found a workaround... the Graph ID. 🧩πŸͺͺ

What's that, then? πŸ€” An Identity Graph links different types of identifiers together β€” cookies, hashed emails, alternative IDs, phone numbers and other permitted signals β€” to build a unified profile where the available data supports that connection.

πŸ“Œ Why is it useful? It's the AdTech equivalent of the philosopher's stone. πŸͺ¨πŸ’°

It doesn't actually turn lead into gold, but it can link a third-party cookie or CRM identifier to a relevant in-app or CTV identifier β€” where matching is possible, permissions apply and the platforms support it. πŸ†”A β†’ πŸͺͺπŸ‘© β†’ πŸ†”B

That matters because cookies have traditionally been plentiful on parts of the Open Web, while other environments have different identity signals and constraints (the availability of third-party cookies is not uniform across browsers or users).

Signal or IDTypical environmentWhat a graph might do
Browser cookieWebLink browser activity to another authorised identifier where a matching signal exists.
Hashed email / CRM IDPublisher logins and advertiser databasesConnect first-party records across permitted partners (hashing alone does not anonymise a person).
Mobile advertising IDIn-appAssociate mobile-app exposures with other recognised identifiers when available and allowed.
CTV / household identifierStreaming and connected TVSupport household- or device-level links, which must not be mistaken for certain person-level identity.

What is the difference between deterministic and probabilistic identity matching?

Not all connections in an Identity Graph are created equal. Some come from directly observed relationships; others are inferred from patterns. And they do not offer the same level of confidence.

MethodHow it worksMain limitation
DeterministicMatches identifiers using a known link, such as the same authenticated account or a recognised first-party identifier.Coverage depends on available verified links, identity access and permissions.
ProbabilisticInfers that signals or devices may belong to the same person or household using a statistical model.It can make false matches or miss valid links; confidence depends on the model and signals.

For Media Traders, this distinction matters particularly when evaluating incremental reach, frequency capping and attribution. A graph's coverage is not the same thing as its verified match accuracy (and household reach is not automatically individual reach).

Who benefits from Identity Graphs in the advertising ecosystem?

Who's interested? Quite a few people, as it turns out. πŸ’΅

  • πŸ’΅ Advertisers: retarget CRM audiences and attempt to manage frequency capping across environments, where matching is supported and lawful.
  • πŸ’΅ Agencies: build a strategic asset around identity. A particularly striking example is Publicis, which announced the acquisition of Lotame in 2025 and completed its LiveRamp acquisition on 2 November 2026.
  • πŸ’΅ AdTech companies: especially bidders and DSPs, which can use eligible identity links to strengthen their proposition.
  • πŸ’΅ Publishers: monetise their first-party audiences through audience extension across growing formats such as CTV, audio and in-app inventory.

In short: the players that can responsibly link identities may have an advantage when audiences are scattered across technologies. It's easy to see why identity is becoming such a strategic battleground.

What are the limits of Identity Graphs β€” accuracy, transparency and privacy?

But every silver lining has a cloud. ☁️

⚠️ Like the Coca-Cola recipe, each graph's formula is often kept secret. The methods behind its matches, confidence scores and data coverage may be difficult for a buyer to audit.

⚠️ Graphs often mix deterministic and probabilistic signals. Mapping accuracy varies with input quality, freshness and which identifiers can genuinely be linked.

β†’ So there's a transparency problem. πŸ”’

There's another essential consideration in France and Europe: identity resolution does not sidestep privacy law. The CNIL's guidance on cookies and other trackers explains that advertising retargeting using device trackers generally requires prior consent, and rules on personal-data processing also apply. An email address does not become anonymous simply because it has been hashed.

Questions a Media Trader should ask an ID Graph provider

  1. Coverage: what proportion of my real audience is addressable in each browser, app or CTV environment?
  2. Accuracy: which matches are deterministic, which are probabilistic, and how are false matches assessed?
  3. Transparency: which data sources, TTLs and validation methods can you document?
  4. Privacy: which consents, lawful bases, purpose restrictions and opt-outs apply?
  5. Measurement: is deduplicated reach verified, or simply modelled?

Related reading: why publishers are considering Login Walls, and what a Bid Listener can actually observe.

How does GreyFox Ads tackle audience fragmentation without its own Identity Graph?

At GreyFox Ads, we're trying to solve audience fragmentation, but... we don't have a Graph ID. πŸ˜‚

Instead, we've taken a complementary approach: we use different targeting methods according to the environment β€” cookies + first-party data + alternative identifiers + contextual and semantic targeting. 🎯πŸͺπŸ†”πŸ“ (Each approach is used only where technically available and consistent with the relevant privacy permissions.)

Exposure frequencyΓ—2reported campaign result
Anti-cookie environments32%share of the campaign delivered

We've just finished a campaign which, using this approach, kept exposure frequency at x2 while delivering 32% of the campaign in anti-cookie environments. πŸ’ͺ

These are GreyFox Ads' reported campaign results, not a market benchmark. The term β€œanti-cookie environments” reflects the original campaign terminology; it does not, by itself, show that every impression was consentless or that all users were deduplicated across devices. Frequency Γ—2 is the reported campaign metric, not proof of graph-level person-by-person matching.

Not bad for France's smallest programmatic ad sales house, eh? πŸ˜‰

For a related operational tactic, read why DSPs may underbuy cookieless impressions and how to manage frequency more intelligently.

PS: πŸŽƒπŸŽƒ Happy Halloween, everyone! πŸŽƒπŸŽƒ

Original post, primary sources and further reading

This page adapts the original GreyFox Ads personal analysis. The original LinkedIn publication date has not been independently verified; this English HTML edition was updated on 12 November 2026. Product descriptions and company transactions are attributed to their respective sources.