Stories That Matter: AI in media planning - the race nobody wins

Seven AI announcements landed in advertising this week featuring seven different companies. But there’s only one pattern. And it's one the industry has seen (and ignored) three times.

“The fact that millions of people share the same mental pathology does not make these people sane”. Erich Fromm; The Sane Society (1955)

There are so many reasons why you should read Fromm’s important psychological takedown of modern consumerism and how, as Marx expertly diagnosed through an economic historical lens, capitalism estranges individuals from their work and others. And so, people become ‘marketing characters’ who treat their own personalities as commodities to be ‘bought and sold’ for professional or social success.

Such mania must explain this relentless, wolfish, and overbearing attitude that AI will (or even must) be revolutionary. That must be why I persist with the crazy-making ordeal of sifting through all the trade media stories, and newsletters about advertising, despite this never-ending and unchallenged ‘AI is changing everything’ refrain.

From the last week, here are seven different stories, featuring seven different companies, containing seven different announcements.

OpenAI turns on CPC ads. (Digiday) StackAdapt leaks a ChatGPT pitch deck. (Adweek) Jellyfish wants you to use LLMs for media planning. (Adweek) Stagwell backs Trade Desk’s AI agents. (Campaign) Omnicom launches an agentic influencer tool. (Adweek) Hershey bets on AI agents for MMM. (Adweek) Goodway Group partners with Optable to feed cleaner data to AI agents. (Adweek).

And as for LinkedIn… if your feed is not currently being clogged up with incessant know-it-all AI chat, then please screenshot and send it to me so I can a) verify AI didn’t create it for you; b) have some faith restored.

Why am I so down on AI, you ask? Ah well, read my recent piece for The Media Leader. Actually, read this one too.

Now, let’s zoom out beyond the breathless headlines and ask what any good journalist should be doing as a matter of course.

What is actually going on here?

The honest answer is: a land grab. Every one of those seven announcements is a move in the same game — a scramble for position in the emerging AI advertising stack. But, as usual, very few out there are trying to join the dots.

So let me try.

What “commoditising” advertising actually means

Advertising has always had two valuable layers.

The thinking layer — strategy, planning, creative judgement, cultural insight — where the interesting work happens. Where someone looks at beer sales data among men in their forties on winter Saturdays and understands that what they’re really seeing isn’t thirst, it’s ritual.

Then there’s the execution layer — buying, trafficking, optimisation, reporting. This is where the margin has quietly lived; not because it was glamorous, but because it was complex enough to require specialists with expensive tools.

Let’s be clear: it is this second layer that AI is genuinely dismantling. Not the thinking, but the procedural scaffolding around the thinking. That’s certainly important, but it’s not revolutionary.

The media plan as a document, the channel allocations, the budget splits, the reach and frequency targets: yes, AI can accelerate all of that.

But here’s the thing I keep asking in response: so, bloody, what?

“People with blindfolds running a race”, an AI-gen image created by Substack. Don’t look too closely!

In all my time talking to agency planners and buyers, this part of advertising was never cited as ‘the hard part’.

The hard part was always the question underneath it: why this channel, this audience, this moment, at this weight, in a way this particular brand can credibly own?

How you answer that is generally called “craft”. Great media planning or media strategy is artisanal, much like consulting: it requires taste, judgment, experience, and — crucially — can not be optimised because it is, to a point, subjective. At some point you just have to accept, as a matter of faith, that your advertising was planned by someone who knew what they were doing and gave you a really great plan. You accept it’s simply not possible to mine the world’s data with the world’s best AI to come up with the greatest or the most optimal plan. Because life doesn’t actually work like that, even if the AI does what you think it can (not for this article, but buyer beware!)

So there’s an execution ‘problem’ here for the agencies whose advice big advertisers have relied on for decades. When execution becomes cheap and easy, it stops commanding a premium. The activity doesn’t disappear (if anything it will explode if the cost of trading lowers), but it stops being a source of advantage because anyone can do it.

The question is: who captures the value that’s left?

Three players are competing for the answer. They can’t all be right.

The platforms: because they’ve done this before…

Google, Meta, and their newest rival OpenAI are making the same bet: if you own the surface where advertising is executed, you capture the margin regardless of who does the planning.

But ‘own the inventory, own the targeting, own the measurement.’ is not a new bet. In fact, it’s the only bet these companies have ever made. And it has worked, repeatedly, in ways the industry keeps failing to learn from.

Search advertising. Google built the infrastructure, agencies built practices around managing it, and so brands paid for the clicks. Google captured the margin. Today, Google and Meta together account for roughly half of all global digital ad spend.

The agencies that “managed” search are now structurally dependent on the platforms they were supposed to be advising clients about. Is that why Meta and Google keep doing awful things and this industry keeps doing deals with them instead of treating them as the rogue actors they are? (Tune in next time on No Shit Sherlock…)

Programmatic display. Years ago as a tech editor, I remember constantly hearing about ad tech was supposed to democratise media buying. But instead it created a supply chain so complex — with bendy arrows bumping into the jargon language of DSPs, SSPs, DMPs, ad exchanges — where value concentrated at the nodes controlling data and inventory. Of course, Google owned most of these nodes. The independent ad tech layer got squeezed or acquired.

Social. Facebook built targeting infrastructure so sophisticated that brands couldn’t replicate it internally and agencies couldn’t meaningfully improve on it. The strategic layer became largely a matter of feeding the algorithm correctly. Meta captured the margin.

Three transitions. Same pattern. A new technology layer arrives, multiple parties compete to own the value, and the entity controlling distribution wins.

…but here’s where ChatGPT is different

But the platform advantage goes deeper than distribution. What the platforms actually captured, alongside the pipes, was measurement.

Advertisers spent years feeding their outcome data into Meta and Google, freely and at scale. The platforms used it to build what Sameer Modha, writing in The Media Leader last year, called Large Outcome Models: proprietary systems trained on data given to them by millions of advertisers (including their competitors!) Those models now tell advertisers what’s working. They’re also built to extract maximum spend, not to serve advertiser interests.

The advertiser trained the model. The model now works against them.

That’s not distribution control. That’s knowledge control. The platform doesn’t just own the pipe. It owns the definition of what counts as a good result.

OpenAI is now trying the same move, but faster and with a new flex: ChatGPT isn’t just an ad platform, it’s a demand-generation environment. The user arrives with intent already formed. If OpenAI can prove that ads served inside a chat interface drive outcomes, it doesn’t just take a cut of existing budgets — it claims to create a new budget category that didn’t exist before.

Meanwhile, Brian O’Kelly — the man credited with inventing programmatic — is building an operating system for AI agents that buy and sell media to each other with no humans in the loop. The roads are being paved; toll booths, and all.

Is OpenAI certain to win this? No. Any company led by Sam Altman, given everything we know about him, should be treated with extreme caution.

And anyway, Google’s moat was a decade of search intent data at a global scale. ChatGPT generates conversation data, which is valuable, but not the same thing.

So the attribution question remains unanswered: can OpenAI prove that an ad served mid-conversation drove a sale? CPMs have already fallen from $60 at launch to around $25 nine weeks in. If they can’t solve measurement, the premium inventory narrative collapses before it’s even started.

But the direction of travel — platform, owns stack, captures margin — is familiar enough to take seriously.

The agencies: the same trap, one level up

Agencies are falling over themselves to remind you, meanwhile, that they are not standing still.

Jellyfish (Brandtech Group) wants you to use LLMs for media planning. Stagwell is backing Trade Desk’s AI agents. Omnicom has launched an agentic influencer tool. The pitch is consistent across all of them: AI handles the execution, which frees us to focus on the strategic thinking machines can’t do.

It’s a reasonable pitch. But it has a big flaw.

I spoke recently on my podcast Believe It Or Not with co-host Hamish Nicklin — former CEO of Dentsu Media UK, now an AI consultant — about the extent to which AI will genuinely replace human judgment in media planning. His argument was more unsettling than the agency talking points suggest.

The planning knowledge base that agencies draw on is, he argued, already largely codified. We have 60 years of media effectiveness data, decades of IPA case studies, Binet and Field’s long/short framework, and MMM outputs across every major category — the patterns are well understood, well documented, and already partially embedded in agency planning tools. What a planner does — often instinctively — is operationalise those patterns.

Yes, sometimes brilliantly. But the patterns themselves are learnable. And if they’re learnable, they’re automatable.

The flaw in the agency AI pitch is this: if every agency is using broadly similar tools, trained on broadly similar data, optimising toward broadly similar signals, the planning layer homogenises too. You don’t get differentiated strategy from undifferentiated tools. You get a new kind of commodity, one level up the stack.

We have seen this before. When DSPs ‘democratised’ media buying, agencies initially sold access to them as a competitive advantage. Within three years, every agency had the same DSP access. The tool became table stakes.

I’ve seen some very clever people literally roll their eyes when, in conversation, I compare AI tools to email. Everyone has had email for decades now, where is the advantage?

Then the sanguine myth surfaces about “unlocking human potential” based on some cheery hypothetical which ignores our natural urge to be lazy (driven by a very important biological drive to conserve energy!)

The one genuine agency advantage that could survive this logic? Proprietary data at scale.

Decades of multi-client, multi-category campaign performance data that no individual brand can replicate. That’s a real moat. Hamish made this point explicitly — the agency can point its AI at thirty years of cross-industry data; the brand can only point it at its own history.

But the question hanging over that advantage is duration. How long will agencies have until AI tools become more accessible and data availability increases across the ecosystem?

The brands: efficiency is not advantage

Brands are being sold something rather different: efficiency.

“Lower CPMs!” shout the bazaar hawkers. “Faster creative iteration!” “Reduced headcount!” The Hershey story this week — AI agents fixing a $2 billion marketing blind spot — is a version of this frenzied pitch in press-release form.

Some of the efficiency gains must be real… but efficiency is not the same as competitive advantage. It’s a cost reduction, not a revenue driver. And if every brand captures the same efficiency gains from the same tools, no individual brand is better positioned relative to its competitors than before.

The deeper problem is measurement. Outside the walled gardens, the infrastructure required to prove what actually drives growth is too weak to do the job. Brand-level models — even sophisticated ones built on strong first-party data — are structurally insufficient. Modha’s framing in The Media Leader was pointed: isolating ROI one advertiser at a time is like shouting into a data storm. Too little signal, too much noise.

You need category-wide, cross-brand data to build models robust enough to withstand platform dashboards. Individual brands can’t build that alone. Only platforms, or serious industry collaboration, can.

Which means the brands chasing efficiency through AI tools are doing so largely blind to whether it’s working. Because the measurement infrastructure that would tell them is either controlled by the platforms or doesn’t yet exist outside the walled gardens.

So who wins?

The honest answer is: we don’t know yet.

But we do know that three times in the last 25 years, a technology transition in advertising has produced the same outcome.

Agencies and brands were told they’d capture the value, so they built practices, tools, and pitches around the promise. But the platforms captured the value; not just because they owned distribution, but because they owned measurement too. Measurement is always a huge story in media and advertising because it’s a battle over what literally ‘counts’ as winning.

There are genuine reasons to think this time might be different at the margins. Hamish made a point I haven’t been able to shake: the serendipity layer — the Channel 4 executive who mentions a series that hasn’t been announced yet, or the media partnership that emerges from a conversation that no dataset could have predicted — this remains stubbornly human. The insight that emerges not from pattern recognition but from genuine cultural judgment, the ability to make a brand punch above its weight in ways that no optimisation loop would generate, that still requires a human. Probably.

And so the uncomfortable question underneath all of this — the one I haven’t seen anyone in the trade press ask plainly this week — is not whether AI can replace human judgment in advertising. It’s whether the humans who do the thinking get paid for it, or whether the platforms that own the room take the money instead.

That question doesn’t have an answer yet. But the industry is currently behaving as though it does — and as though the answer is flattering to the people making the announcements.

It probably isn’t.

---

Welcome to Ad-verse Reactions

I’m Omar Oakes: journalist, industry critic, and founding editor-in-chief of The Media Leader.

I’ve spent a decade watching smart businesses in media and advertising say less than they mean. The thinking is there, but the language isn’t. I call that gap narrative debt — and this newsletter is where I pick it apart.

Every Tuesday: a Stories That Matter edition dissecting the week’s big story. Plus deeper dives, things I’m reading, and the occasional provocation.

Alongside this, I run oomph. — a narrative consultancy for media and advertising leaders whose story hasn’t kept pace with their business. More at oomphoomph.com.

📩 omaroakesmedia@gmail.com 🔗 linkedin.com/in/omaroakes