Data overlords or data poverty: A stark choice for advertisers
Opinion
Big Tech’s outcome models are powerful but pricey. Advertisers’ meagre brand-level models just don’t cut it anymore. If we want to know what really drives business growth, we have to work together, says ITV’s Sameer Modha.
If you’ve ever tried to buy a kidney, you’ll have an idea what it’s like purchasing pricing data from a major vendor.
A few years ago, I did precisely this. Though when I say “bought” pricing data, I mean I begged, negotiated, and wore down two account managers, only to be quoted a price somewhere between Luxembourg’s GDP and that of a black-market organ.
I was trying to tackle a big problem: we knew TV advertising had pricing power, but couldn’t show it in a way that stood up to platform dashboards. To prove marketing drove margin, not just volume, we needed category-wide models.
And to build them, I needed that data.
The result? Our first estimates suggest up to 30% of value is missing from conventional marketing ROI, which is a strong result.
But the bigger issue is how hard it was to get there.
Platforms have privatised marketing learning
In the pre-digital age, we shared the work of measurement. Media owners funded joint industry systems to count eyeballs, and advertisers paid to link those measurements to their business results — usually through a long process of market mix modelling.
Then along came Meta and Google, offering fast, robust measurement for free. All advertisers had to do was hand over their outcome data. The platform would do the rest.
But what else were they doing?
Everyone today is familiar with large language models. We know tech companies need vast amounts of data to train something like ChatGPT. Similarly, to train a model that predicts and measures advertising’s impact on business outcomes, you need vast datasets, spanning countless campaigns across every market condition.
But unlike Language models, no one is talking about licensing Outcome data. In fact, millions of advertisers have been donating it for free for over a decade.
Large Outcome Models
The Large Outcome Models, which Google and Meta have built into their ad products, are one of the glories of modern, planet-scale data science. But they serve only one master. Platforms use them to extract as much money as possible, leaving advertisers only marginally better off than they would be otherwise.
The result? Data serfdom. Advertisers can only see their own campaigns, not how their media mix compares to others or how channels work together.
But if they choose life outside the walls, even the biggest brand can only scrape together a few hundred campaigns across a handful of years. That’s like trying to train GPT on a few Wikipedia pages. Basic models built with first-party data just can’t cut through the complexity of today’s media mix.
What we need are shared models capable of measuring the actual business impact from different media, without extorting the platform tax. But how?
Put more R in the ROAS
A Google colleague once told me: “In a ROAS world, no one will care about audiences.”
As a former planner, I hated the idea at the time. Still, I realise now he was right: incentives were becoming reality, Return On Ad Spend, the lived experience: challenging targets, quarterly reviews, spreadsheets full of KPIs. If it looked good, it was good.
And thanks to their private, planet-scale models, platform ROAS looked very good indeed.
How did classical media respond? We turned our eyes to the gold at the end of the rainbow. “The Long and the Short of It” framing created an unhelpful binary: chase short-term activation or wait patiently for brand effects to pay off later.
That set up, conflict where there should have been coordination. Worse, it made TV look incompatible with near-term performance.
TV delivers measurably positive value within the quarter, but while econometrics, attribution and experiments measure what’s easy to count, they struggle to detect the big payoffs: perceived value, mental availability, margin growth and risk mitigation.
To redress the balance, we need to use the tools of modern data science to fight back by increasing the breadth of return we detect; growing the R of ROAS, which brings us back to mismatched incentives and costly data.
A collaborative approach
You see, measurement and data vendors face their own incentive trap. In the short term, it’s easier for them to keep selling bespoke models and data, one punter (and kidney) at a time. After all, pursuing shared learning would mean smaller individual contracts.
But this approach ultimately weakens everyone’s ability to compete with the Large Outcome Models. Which is why, as media owners, we have to help kickstart a different approach.
Just as our predecessors did when they created the JICs decades ago, we believe in ecosystem economics. We know measurement credibility drives investment. And investment keeps the market healthy; when advertisers feel confident in the value of media, everyone wins.
Do we have a vested interest? Of course we do. But the same one as advertisers. This is why we’ve spent the last few years at ITV building a collaborative approach, rooted in shared learning.
Through tools like Outcomes Planner, Lantern, and now Pricing Power, we’re helping advertisers see what moved real business levers like margin, market share, and price elasticity. We’re even sharing our models with measurement partners to help improve their work and make it more scalable.
But we need to get a move on.
The Tok is ticking
AI is driving huge improvements in platform ad products. Its models are about to get even better, and advertiser choices even starker.
This work isn’t about proving TV is always the answer; it’s about ensuring fair, credible outcome measurement for all media and aligning it with how businesses actually grow. If you’re looking to do the same, let’s work together to build a new era of ecosystem economics.
Wait too long, and we might have to sacrifice more than a kidney to get the full marketing picture.
Sameer Modha is measurement innovation lead, commercial, at ITV