Avinash Tripathi is a VP Analytics at University of Phoenix, thought leader and keynote speaker with over 20 yrs of experience in the field.
The better YouTube works, the worse it looks on a dashboard from a last-touch standpoint. That is not a glitch in your reporting. It is the design flaw at the center of modern marketing measurement.
For years, marketing leaders have operated under a comforting illusion: that performance can be measured precisely, cleanly and in near real time. AI-driven discovery and platforms like YouTube have broken that illusion for good.
The measurement problem didn’t start with AI.
Even before AI reshaped consumer behavior, measurement was structurally misaligned. Today’s ecosystem is deeply fragmented:
• Linear TV and streaming are measured in one system.
• Digital conversions are measured in another.
• Platforms operate on different attribution logic.
• View-through windows vary widely.
• Even the definition of a “conversion” is inconsistent.
This creates an invisible distortion to the marketer’s understanding of the value of their own marketing.
What makes this moment more interesting is that the platform serving an ad is no longer the platform defining its value. In emerging setups, media buying, audience definition and measurement are increasingly decoupled across platforms. A marketer might reach a LinkedIn-defined executive audience on a Roku or FOX stream, buy that inventory through Google’s DV360 and measure the results using a framework completely separate from both LinkedIn and Google.
This is the architecture of the new marketing, not a flaw. What is needed is a transition from platform-level intelligence to system-level intelligence, where all attribution models, all types of incrementality tests and all types of media mix models should be combined into a unified reporting system rather than reported out separately.
YouTube changes everything.
YouTube does not behave like a traditional performance channel. It operates upstream, shaping intent, influencing consideration and driving demand that materializes later across search, direct traffic and even offline channels.
As organizations scale investment in YouTube and other upper-funnel channels, the risk is no longer just inefficient spending. It is systematically flawed decision-making. Without a unified measurement system:
• High-impact channels get underfunded.
• Lower-funnel channels get over-credited.
• Strategic bets are cut prematurely.
• Growth is constrained, not by budget, but by mismeasurement.
Organizations end up optimizing for what is easiest to measure, not what drives outcomes.
Utilize the YouTube Measurement Framework.
Solving this does not require a new dashboard; instead, it turns the patchwork of existing YouTube measurement tools into a system that works as one cohesive unit, which I call the YouTube Measurement Framework. It consists of three core elements that together answer three core questions of any business: Where is the demand coming from? What is the real impact of marketing? And where should a business invest next?
Together, attribution modeling, incrementality testing and media mix modeling (MMM) provide a far more complete analysis of marketing performance than any of a single method.
1. Attribution provides direction, not truth.
YouTube does not simply capture existing demand; it influences the moments where intent is formed. Traditional attribution models are built to measure clicks, while YouTube’s value lies mostly in shaping consideration and future behavior long before a conversion happens. Relying on last-touch attribution for a channel like this significantly understates its impact, because the model is answering the wrong question.
A better approach is to evaluate multiple attribution models. Because they can all be calculated from the same underlying conversion data, the goal is not to pick one “correct” model. The goal is to evaluate all of them against your own data to see the range of plausible values a channel like YouTube may be creating.
Multi-touch attribution is not a perfect measure of incrementality, but it is a practical bridge between attribution, experimentation and MMM. That is where the next two parts of the system come in.
2. Incrementality reveals causality.
Attribution tells you what happened. Incrementality tells you what would have happened anyway, and the gap between those two answers is where the real strategic insight lives.
Experiments like Google’s Meridian GeoX, which is currently open source and available to anyone, introduce genuine causal measurement into YouTube reporting. By varying investment across geographies, they isolate the actual lift a campaign produced from the demand that would have shown up regardless, including whether some regions are simply more valuable than others to begin with.
These tests are slower and less convenient than dashboard reporting. But they answer the one question attribution can’t: Did this investment drive incremental outcomes, or just take credit for demand that already existed?
3. MMM turns measurement into strategy.
MMM has been used to evaluate marketing at the mix level for decades. What has changed is that modern data and AI have turned it from a backward-looking explanation tool into a forward-looking decision system.
The biggest shift in MMM is not a new algorithm; it is the convergence of attribution, experimentation, identity resolution and AI-driven consumer behavior into one measurement ecosystem. Traditional MMM was built to explain what happened last quarter. The next generation is built to recommend what should happen next, using AI-powered scenario planning and causal calibration from GeoX and incrementality testing as inputs.
This convergence is also why MMM increasingly needs to move beyond impressions and clicks to account for attention, engagement quality and incremental reach. That is especially true for YouTube, CTV, streaming and retail media, where traditional metrics were never built to capture upstream influence.
The bigger shift, though, is the rise of AI-mediated discovery. The task now is to understand how demand is created and shaped in an AI-first world, long before that demand appears as a measurable website click.
System intelligence is strategy.
Each of these three methods is incomplete on its own. Attribution shows direction, but it can mistake correlation for causality. Incrementality proves causality, but it moves too slowly to run everywhere. MMM helps optimize the broader system, but the quality of its recommendations depends on the attribution and incrementality inputs that feed the model.
Together, they stop being three separate reports and become one system, capable of telling marketers not just what happened, but what’s working and what to do about it next.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?