The measurement infrastructure that agencies have relied on for 15 years is breaking apart. Third-party cookies are functionally dead in most environments, mobile identifiers are opt-in at single-digit rates, and platform-reported attribution is so inflated that comparing numbers across channels has become an exercise in fiction. Agencies that wait for a drop-in replacement will be waiting indefinitely; the only viable path forward is rebuilding measurement from the ground up using frameworks that do not depend on user-level tracking.
Most agencies hit this problem sideways. A client asks why Facebook is claiming credit for 300 conversions while Google also claims credit for 280, and the actual checkout count is 310. The ops lead pulls the numbers into a spreadsheet, applies some deduplication logic they invented at 11 p.m., and presents a "blended" view that everybody quietly knows is wrong. This ritual plays out monthly at thousands of agencies, and it has only gotten worse as signal loss accelerates. Apple's App Tracking Transparency rollout in 2021 cut Facebook's observable conversion pool by an estimated 30% to 40% overnight. Google's Privacy Sandbox, while delayed repeatedly, signals the same direction. The writing is not ambiguous.
The first thing agencies need to accept is that the era of deterministic, user-level, cross-platform attribution is over for the vast majority of campaigns. Last-click and even multi-touch attribution models that relied on stitching together cookie-based journeys are operating on increasingly incomplete data. In some verticals, particularly mobile-heavy categories like gaming, food delivery, and retail apps, the observable portion of the conversion funnel has dropped below 50%. You are making budget decisions on half the picture, and the half you can see is biased toward platforms that own their own login ecosystems.
"If your measurement framework requires seeing every user touchpoint to function, it already does not function. You just have not admitted it yet."
So what replaces it? The answer is a layered approach that combines three methods, each compensating for the others' blind spots: media mix modeling, incrementality testing, and first-party data enrichment. None of these are new ideas. Media mix modeling, or MMM, dates back to the 1960s when consumer packaged goods companies needed to measure TV and print without clickthrough data. Incrementality testing is just controlled experimentation applied to marketing channels. First-party data enrichment is what every direct-response marketer did before the ad tech ecosystem existed. What is new is the urgency, and the tooling that makes these approaches viable at agency scale rather than only at enterprises with in-house data science teams.
Media mix modeling works by analyzing the statistical relationship between marketing spend (broken down by channel, creative, timing, and geography) and business outcomes (revenue, leads, store visits) over time. It does not require user-level data at all. It requires historical spend data, outcome data, and enough variation in spend patterns to isolate each channel's contribution. The classic weakness of MMM was latency: traditional models took 8 to 12 weeks to build and refresh quarterly, making them useless for in-flight optimization. Modern implementations using Bayesian statistical methods and automated pipelines can refresh weekly. Google released its open-source Meridian MMM framework in 2024, and Meta has maintained its own open-source Robyn project since 2021. Neither is plug-and-play for a typical agency, but both have lowered the barrier considerably from the days when MMM required a six-figure consulting engagement.
The operational challenge for agencies adopting MMM is data hygiene. The model is only as good as the inputs, and most agencies have spend data scattered across platform dashboards, manual reports, and inconsistent naming conventions. Before you build a model, you need a single source of truth for spend by channel, by day, with consistent taxonomy. This is grunt work, not glamorous work, and it is where most MMM initiatives stall. Agencies that run disciplined sprint-based operations have an advantage here because the data collection and normalization work can be scoped, assigned, and tracked the same way you would manage any other deliverable. Platforms built for continuous operational workflow, like ScrumRithm, help ops teams treat measurement infrastructure as ongoing work rather than a one-time project that drifts off the priority list.
Incrementality testing fills a different gap. Where MMM tells you the average contribution of a channel over time, incrementality testing tells you what would have happened if you had not spent on that channel right now. The most common format is a geo-holdout test: pick a set of matched markets, suppress spend in the holdout group, and compare outcomes. This sounds simple, and conceptually it is simple, but the execution details matter enormously. You need enough geographic units to achieve statistical power, you need to run the test long enough to capture the full purchase cycle, and you need a client willing to leave money on the table in holdout markets for four to eight weeks. That last part is the hardest sell.
| Method | Data Required | Refresh Speed | Best For | Biggest Weakness |
|---|---|---|---|---|
| Last-Click Attribution | User-level cookies/IDs | Real-time | Nothing, anymore | Ignores 60%+ of the funnel |
| Multi-Touch Attribution | User-level cross-platform IDs | Near real-time | Logged-in ecosystems only | Data gaps render models unreliable |
| Media Mix Modeling | Aggregate spend and outcomes | Weekly (modern) | Budget allocation across channels | Cannot optimize creative or audiences |
| Incrementality Testing | Geo-level spend control | Per test cycle (4-8 weeks) | Validating channel value | Requires spend suppression |
| First-Party Data Enrichment | CRM, email, purchase data | Continuous | Customer-level insights | Limited to owned data scope |
First-party data enrichment is the third layer, and it is where agencies can differentiate most for their clients. Every business has data about its actual customers: email engagement, purchase history, support interactions, product usage patterns. The problem is that this data usually lives in five different systems with no unified view. When you connect CRM data, email engagement data, and purchase data into a single customer record, you can build cohort-level measurement that does not depend on third-party signals at all. You can answer questions like: "Do customers who engage with our email program within 14 days of first purchase have higher 90-day LTV?" That is a measurement insight no amount of ad platform data could give you, and it is immune to cookie deprecation.
The practical work of connecting these data sources is where agencies earn their keep. Most mid-market clients do not have a data warehouse. They have a CRM, an email platform, an e-commerce backend, and maybe a customer support tool, each with its own customer identifier. Building even a basic identity resolution layer that matches email addresses across these systems unlocks measurement capabilities that the client could not access before. This is also where AI-driven customer intelligence tools add real value. Platforms like Aigotchu that track customer interactions and surface health signals can contribute behavioral data that enriches the measurement picture without requiring any third-party tracking.
One pattern I have seen work well at agencies going through this transition: start with a measurement audit sprint. Spend two weeks documenting every measurement tool, every attribution model, every reporting workflow currently in use across the agency's client portfolio. Catalog what data each one depends on, what percentage of that data is still reliably available, and what decisions get made based on each report. You will almost certainly find that 30% to 40% of your current measurement stack is producing numbers that nobody trusts but everybody still references because there is nothing better. Killing those reports is step one. Replacing them with something honest, even if less precise, is step two.
"An honest answer of 'we believe this channel drives between $80K and $120K in monthly revenue' is more useful than a false-precision answer of '$97,432.18' built on data that is 60% modeled by the platform selling you the ads."
The agencies that will win this transition are the ones that reframe measurement as a service line rather than a reporting function. When your client's CMO asks "is our media working," the old answer was a dashboard with green numbers. The new answer is a quarterly measurement program that combines MMM budget allocation, incrementality validation for the two or three biggest channels, and first-party cohort analysis that ties marketing activity to actual customer value. That program has its own scope, its own deliverables, and its own pricing. It is not a freebie bolted onto media buying.
This is hard. It requires skills that many agency teams do not have today, specifically statistical literacy, data engineering basics, and the ability to communicate uncertainty to clients who want certainty. It also requires operational discipline to maintain data pipelines, run tests on schedule, and update models regularly rather than treating measurement as a quarterly fire drill. But the agencies that build this capability now will own client relationships for the next decade, because measurement is becoming the bottleneck that determines whether marketing budgets grow or shrink. The signals are disappearing. The need to prove ROI is not.
What is the best replacement for third-party cookie-based attribution?
There is no single replacement. The most effective approach layers media mix modeling for budget allocation, incrementality testing for channel validation, and first-party data analysis for customer-level insights. Each method covers the blind spots of the others, and together they produce a more honest measurement picture than cookie-based attribution ever did.
How long does it take an agency to implement media mix modeling?
A basic MMM implementation using open-source tools like Google's Meridian or Meta's Robyn can be stood up in four to six weeks if your spend and outcome data is clean. The real time investment is in data normalization, which can take two to four weeks on its own depending on how many platforms and naming conventions need to be reconciled.
Do agencies need data scientists to rebuild their measurement frameworks?
You need at least one person with statistical literacy, but they do not need to be a PhD-level data scientist. Open-source MMM tools have lowered the technical bar significantly. More important than modeling expertise is operational discipline: maintaining clean data inputs, running tests consistently, and communicating results with appropriate confidence intervals rather than false precision.
How should agencies price measurement services for clients?
Measurement should be scoped and priced as its own engagement, not bundled invisibly into media management fees. A typical mid-market measurement program, including quarterly MMM refreshes, two incrementality tests per year, and monthly cohort reporting, might run $5,000 to $15,000 per month depending on complexity. Pricing it separately forces both the agency and client to treat measurement as a first-class deliverable.
What first-party data sources matter most for post-cookie measurement?
Email engagement data, purchase or conversion history, and CRM lifecycle stage are the three highest-value first-party data sources. Email engagement is particularly valuable because it provides a direct, consented signal of customer interest that can be correlated with downstream revenue without any third-party tracking dependency.
