Written and contributed by: Thuy Ho

During the webinar about Identity, Prebid walked publishers through how to properly measure the incremental value of identity solutions using Prebid’s Enrichment Lift Measurement (ELM) module. The session combined industry context, a live demo, and practical guidance on interpreting results, with a strong reminder that first-party relationships and privacy now define the future of addressability.

The webinar was hosted by:

Mike Racic: President of Prebid

Jeff Wieland: Identity PMC Chair & Principal Technical Product Manager at Magnite.

Mark Dixon: Senior Sales Engineer at Optable

Identity after third-party cookies: 4 key trends

• Third-party Cookie Loss is about Privacy, not Technology
  Third-party cookie deprecation is no longer just a browser change. It reflects a wider shift towards privacy compliance and changing media consumption patterns, where CTV and other closed environments are gaining share.
  
  The conclusion: the industry still needs first-party identifiers (EIDs) that work across environments.
  
• Publisher-provided Identity and First-Party Data
  As third-party signals disappear, publishers that invest in email logins, hashed identifiers and frameworks like LiveRamp ATS, UID2 and SharedID will own the durable signals.
  
  Those who build trust and authenticated user IDs are best positioned to win on both addressability and measurement.
  
• Privacy Regulation and Infrastructure Reshaping Identity
  With over 20 US states having privacy laws, intensifying GDPR enforcement in Europe and scrutiny of IAB TCF, publishers and vendors are under pressure to:
  

• Implement CMPs and honour Global Privacy Control
• Maintain consent logs and proof of compliance
• Prepare for audits and stricter enforcement
  

This complexity is pushing the ecosystem towards privacy-by-design infrastructure and transparent, auditable systems.

• Server-side Identity: Building Control, Speed, and Compliance into the Future
  Moving identity operations server-side improves latency, centralises governance and reduces privacy risk. Server-side setups:
  

• Reduce client calls and speed up auctions
• Make it clearer what data is shared, with whom and under what conditions
• Make it easier to integrate consent and maintain cleaner data flows

Publishers that invest early in server infrastructure gain long-term advantages in performance, compliance and trust.

What is ELM and why was it built?

ELM (Enrichment Lift Measurement) is a free Prebid module that lets publishers measure the incremental lift from identity or enrichment modules in real auctions, rather than relying on vendor claims or lab tests.

It was developed in response to two clear pieces of feedback from the Prebid community:

• Identity vendors wanted commercial models based on percent of lift driven by their identifiers. That requires credible, publisher-controlled measurement.
  
• Publishers had no simple way to A/B test identity modules. They could add IDs to their wrappers, but had little visibility into whether those modules actually improved monetisation.
  

ELM fills that gap by providing a practical, standardised way to run controlled experiments and push the results into existing analytics setups.

How ELM works

ELM works in 3 steps:

1. Suppression Decision
   For each auction, ELM randomly decides whether to suppress enrichment for that user or not.
   

• The sampling percentage is configurable (e.g. 50/50 true A/B test).
• The decision can be stored in a cookie or in memory, depending on requirements.
  

2. Prevention of Enrichment
   If a user is selected for suppression, ELM blocks the mapped provider (for example, an identity module or real-time data provider) from populating user IDs or segments. This simulates a “no enrichment” experience for that user.
   
3. Control Group Labelling
   ELM labels each auction as test (enriched) or control (suppressed). Those labels are then sent to an analytics adapter.
   

ELM can be wired up to the generic analytics adapter or any existing analytics adapter. The only requirement is that the adapter supports the ELM events, such as bid-won arguments that expose metrics like:

• Revenue per impression by cohort
  
• Bid CPM by cohort
  
• Win rate by cohort
  

In the example conducted by Jeff, SharedID was enabled 50% of the time. Live metrics showed:

• A clear split between control and test auctions
  
• Incremental lift in absolute CPM (e.g. test impressions showing around $0.17 higher CPM than control)
  
• Percentage lift (for example, around 7.5% uplift in the demo environment)
  

Importantly, ELM is not limited to a single ID. Publishers can test combinations of user IDs, suppress any module they choose and explore which configuration of identity signals and enrichment performs best.

Correlation is not causation: how to avoid being misled

Mark Dixon used a fictional “Mark Dixon ID” to illustrate the point:

• Suppose 60% of users have this ID and show a $3.30 eCPM.
  
• The remaining 40% without the ID show a $3.00 eCPM.
  
• On the surface, that looks like a 10% uplift and an easy rev-share pitch.
  

The problem: user value is not uniform. Many other factors affect CPMs, including:

• Traffic source
  
• Device and browser
  
• Ad formats, sizes and positions
  
• Floors, traffic shaping and other yield strategies
  

If the “Mark Dixon ID” simply mapped to Chrome third-party cookies, and the non-ID group was mostly Safari, then higher CPMs would reflect user mix, not the ID itself.

Run a true split test that equalises Chrome vs Safari across test and control, and the apparent uplift disappears. The conclusion: the ID is adding no value.

ELM’s controlled suppression is designed to avoid exactly this kind of mistake.

CPMs are NOT the correct metrics

Mark also argued that CPM on filled impressions alone is not the right metric for evaluating identity solutions.

Identity can create value in several ways:

1. Higher bids in existing channels
   The same bidder in Prebid may simply bid higher when an ID is present.
   
2. Shifts in share of voice across channels
   An SSP in the ad server or another auction may lose to a Prebid bidder that starts winning impressions at a higher CPM when enriched. That changes the mix of revenue sources.
   
3. Higher fill where impressions were previously unfilled
   Some users may previously have received no ad because bids did not meet the floor. With better identity, those requests now receive bids that clear the floor, adding new monetisation.
   

Looking only at CPM on already filled impressions can hide this uplift. In Mark’s simplified example:

• CPM on a specific channel may appear flat or even lower
  
• Total eCPM on an opportunity/request basis, and total revenue, can still increase significantly (for instance, 50%) once fill and channel shifts are factored in
  

The key message: publishers should focus on request-level eCPM and total revenue, and analyse identity performance across all channels and auction paths.

Practical considerations when testing identity solutions

• Consider which channels see the ID
  Other environments outside Prebid (direct, ad server SSPs, video players, etc) may also be enriched. This can shift demand and outcomes in ways that are not obvious from a single channel’s CPM.
  
• Drill into bidders and advertisers
  Using Prebid events, publishers can see which bidders and even which advertisers transact on IDs in test vs control. Some may ramp up spend while others reduce it, shifting share of voice.
  
• Account for deal and open auction mix
  The ratio of direct deals, preferred deals and open auction will influence how much impact an ID can have and where uplift shows up.
  
• Watch for coincidences and unrelated changes
  Changes in inventory profile, geo mix, device distribution, traffic acquisition strategies, or “unrelated” deployments can all distort test results. Time-based analysis down to the hour can help pinpoint when a change really occurred.
  
• Protect first-party value and prevent data leakage
  Publishers should understand:
  

• Who inserts identifiers and where
• What data sources are used for matching
  
• How match methods are represented in the EID provenance fields
  
• Above all, they should ensure identifiers are not used to pool and aggregate their audience data across publishers in ways that erode the value of their own first-party relationships.
  

The role of the Identity PMC and what comes next

• The Identity PMC increasingly acts as a policy and advisory group, helping other committees interpret identity and privacy challenges.
  
• The group also builds and maintains modules, such as ELM and new privacy modules like the TCF controller and TCF consent module.
  
• The goal is to provide publishers with easy-to-integrate tools for:
  

• ID measurement
• Cookie-less attribution
• Regional privacy compliance
  

Looking forward, the team sees ELM evolving into a general A/B tester for Prebid modules, not just user IDs and RTD modules. In theory, the same approach could be applied to:

• Privacy modules
  
• Bid adapters
  
• Other parts of the wrapper where configuration changes need rigorous testing
  

Become a Prebid member to contribute to the market with us!

Q&A

1. When capturing the analytics adapter data back, is that something that anyone could call and get the results outside of the organisation?
   Answer: Access control is handled by whoever maintains the analytics adapter, since it sends events to its own server-side endpoint that shouldn’t be publicly accessible. So, you generally can’t just tap into someone else’s adapter, and detailed questions should go to that adapter’s maintainer.
   
2. Do the analytics/ID control test settings also impact the EID identifiers, too?

Answer: When suppression is triggered, SharedID is simply removed from the EIDs array before it enters the bid stream.