From The Trenches Part 10: Media Buying Just Came Full Circle, But Smarter

By /

More than once in my career, I’ve stepped into environments where paid media ROI wasn’t measured.

Finance would look at the quarterly results, decide how much of the P&L could go into marketing, and then give the marketers the money to spend.

Having built my performance marketing career in a space where funnel metrics on last-touch attribution were constantly monitored, it was hard for me to adapt to and influence within spaces steeped in the “here’s your budget, go spend it” culture.

But by my fifth year of marketing leadership, I got it. I realized that budget size reads as “growth in responsibility, career progression” to a tactical performance marketer and getting more budget is a strong morale booster. Eventually, I got to the point where channel manager morale became the second most important criteria for reallocating small to medium size budget shifts, provided the primary one – ROI – wouldn’t be consequentially affected.

That last part – that ROI governance must always be primary – is what falls by the wayside if we don’t have stewardship of the bottom line. And that was why the “go spend it” culture worked only when Finance was managing the purse strings.

In The Days Before Paid Media And Big Data, We Didn’t Have An Alternative

It’s hard to imagine now that paid media didn’t always exist, but even after e-commerce started to scale in the mid-1990’s, it would be five years before paid ads appeared. The availability and richness of data to inform bidding decisions didn’t come about until circa 2005, which was also the time when businesses started adopting relational databases.

Prior to the advent of Google Adwords (now Google Ads), the main channel where a business might spend ongoing sums of money for marketing was TV. And the effect of TV advertising is not real-time. Think back to the days of your childhood. How many times did you see an ad before it got imprinted in your memory? A certain amount of saturation is necessary to meaningfully drive awareness. That is probably an even bigger reason than the lack of real-time data capture for why marketing ROI couldn’t be measured in isolation.

Big Data Changed The Game – And Clicks Made Things Measurable

Click-based attribution rose in popularity because Google Adwords was the first paid channel to bring Internet advertising to life. And the paid search model is simple: customers search for what they want, sponsored listings appear at the top of Google search engine results, and the people who don’t stop to scroll down to the organic results click on them. It’s the darling of the analytics community because everything is measurable: if someone is interested, they will click because they are already searching for the product in question, and because people are already searching for that product, by nature they are ready – or close to ready – to buy.

Google wasn’t the first platform to use big data in advertising. Programmatic advertising involving real-time bidding (RTB) had been operating since 1994. But click-through rates on display banner advertising are extremely low: less than a tenth of a percent in most cases. Therefore, paid search scaled much more quickly because it was easier to directly associate with ROI, almost in real time. An entire discipline got built around paid search bidding, supported by machine learning algorithms, with click-based attribution at its core.

Widening Adoption of Social Media Drove New Concerns About Privacy

After the advent of Google Adwords, Facebook Ads started running in 2007. YouTube had been acquired by Google in October 2006, leading to the emergence of social media as a paid advertising channel. And as Facebook adoption scaled through the early 2010’s, concerns about privacy emerged, first in Europe where the General Data Protection Regulation (GDPR) was adopted in 2016 and became fully enforceable in 2018.

Click-based ROI measurement hasn’t ever truly gone away. Advertisers have worried that the loss of data signal from the adoption of the GDPR and the release of Apple’s iOS 14.5 in 2021 would substantially impact the measurable ROI, and hence the viability, of performance marketing. Yet, 2022 came and went, and running advertising on click-based attribution remained alive and well. For as long as Google and Meta remained the two primary means that consumers researched purchase decisions or discovered new products and services on the Internet, ROAS measured on click-based attribution remained a viable way to scale growth marketing.

Generative AI Could Shift Digital Touch Points Away From Click-Based Attribution

With the emergence of ChatGPT, a new power came about that could influence purchase decisions. And its impact can’t be fully measured by clicks.

As at the time of writing (November 2025), ChatGPT doesn’t have direct ad placements yet. However, OpenAI can capture commissions from the revenues of sellers through the Etsy and Shopify integrations in ChatGPT, announced in early October 2025.

What can’t be measured, though, is the impact of Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO) when recommendations in AI chat output can influence consumers to make purchases without direct click-through.

Furthermore, engendering customer loyalty and ensuring that customers discover products in ways that are not subject to the vagaries of Google and OpenAI is a key focus for advertisers today. That is resulting in the resurgence of offline tactics such as experiential and billboard marketing. A salient (and ironic) example of this is the use of billboard advertising in New York’s subway to promote the Friend.ai pendant – which turned out to be an example of advertising gone wrong.

Marketing Mix Modelling Isn’t New, But The Proliferation of Touch Points Makes It Relevant

As the number of marketing channels continue to proliferate, and marketing becomes a guessing game about the pattern of consumer interactions in both online and offline environments, click-through attribution is no longer enough to measure marketing P&L holistically.

Visual ad formats in social media, such as Facebook and Instagram Reels and TikTok Shop, may inspire clicks and purchases. But the broader impact to consumer awareness can only be quantified through incrementality testing.

Also, with the multiplicity of channels across search, social, TV, streaming, audio, and out-of-home, measuring each channel in a silo doesn’t take possible inefficiencies from cannibalization into account.

Marketing mix modelling (MMM) isn’t a new technology. The concept dates all the way back to the 1970s. But when the efficiency of measuring channels in silos comes into question, predicting the holistic ROI of the entire marketing portfolio becomes an important component of quantifying the impact of paid marketing to the business.

With MMM’s, we would come full circle. These are models that predict the channel mix that would maximize Return on Ad Spend (ROAS), based on machine learning from historical data. They are, in fact, a form of AI since all machine learning is a subset of AI.

That means we might take the spend mix from the MMM to determine budgets for tactical marketing teams, then tell them to “go spend it”. And harvest the data for the next round of model training. Media buying just came full circle, but smarter.

Why Is Adopting an MMM So Scary?

Even with the market conditions driving MMM integration into Martech roadmaps, there’s still trepidation among advertisers in using them to govern business-as-usual. Here are my theories on why this is so:

  • For an MMM to predict the most efficient channel mix, its training data needs to include the points where each channel achieves diminishing returns. Companies need to make the brave decision to spend inefficiently on a temporary basis to drive better efficiency later.
  • Diminishing returns curves only make sense when all points on the curve (revenue vs. spend) are achieved in similar business conditions. With the current uncertainty surrounding channels and consumer touch points, it’s hard to achieve a long enough period of stability to gather clean training data.

In Part 6, I argued that pay-per-click (PPC) bidding algorithms are declining in importance as an AI-driven discipline to drive marketing ROI, but MMM implementation gives data scientists another area to pivot their attention to.

Coming up in Part 11: the robo-specialist already exists, and his name is Albert. He’s been around since 2010, in fact. But he hasn’t taken the jobs of the hundreds, or thousands, of Marketing Specialists in the field. Why is this? I dig into agentic digital campaign management solutions that are already in the market in my next piece. Stay tuned!

References:

Leave a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Scroll to Top