The AI-in-advertising conversation is running almost entirely on one track right now: generation. Produce more creative, faster, in more variants, personalized to more segments. It is a real capability and it demos beautifully, which is exactly why it dominates the discourse. It is also, I would argue, the least durable part of the story. Generation diffuses quickly and commoditizes just as quickly. The part of the AI-and-advertising intersection that will still matter in five years is quieter and more structural, and in out of home it is happening while most of the attention is pointed elsewhere.
Here is the structural shift, stated plainly. out of home is becoming machine-readable. Physical inventory that used to live in rate cards and sales inboxes is being modeled into structured data with real-time pricing and availability. That sentence sounds unglamorous next to a model that writes a hundred headlines, but it is the thing that decides whether a channel can participate in automated, AI-driven buying at all. You cannot algorithmically plan, target, buy, or measure inventory you cannot query. Structure is the precondition for everything the industry keeps saying it wants from AI in this medium.
Once inventory is structured, the genuinely interesting applications become possible, and notice that they are architectural rather than cosmetic. A DOOH DSP can plan across physical screens, optimize allocation against goals, forecast outcomes, and pace spend the way demand-side platforms have long done online. Machine learning enters as forecasting, allocation, audience modeling, and incremental measurement, all sitting on top of a clean representation of the physical world. The value is in the system that reasons over structured inventory, not in the generator that decorates it.
It is worth being precise about where generation actually fits, because the honest version of the story is more useful than the hype. There is real precedent for AI doing operational work in this medium once the plumbing is in place. A widely cited campaign used generative tooling to produce thousands of hyperlocal headline variations in little more than a day and drove a meaningful lift in store visits. That is impressive, and it is easy to file under a generation win. But look at what actually made it deployable. The generation was the visible layer. What made it work at all was the underlying ability to target and vary messaging across thousands of specific physical locations, which is a structured-inventory and targeting problem. Strip away the structure and the thousand clever headlines have nowhere to go. The generation rode on top of the architecture, not the other way around.
This is the distinction I want trade readers to hold onto, because it changes where you should be looking for advantage. Generation is a feature. Structure is infrastructure. Features get copied within a quarter. Infrastructure compounds. The advertisers and platforms that win the physical channel will be the ones that treated its inventory as a data asset to be modeled, versioned, and served, not as a blank canvas to be filled with more variants. The creative layer sits on top of that layer and always will. The strategic question is who owns the layer underneath, because that is the layer that does not get commoditized next quarter.
There is also a measurement dimension that gets less attention than it deserves. Structured inventory is what makes rigorous, AI-assisted measurement of physical advertising possible, tying exposure to outcomes through location data and controlled comparisons rather than a laminated audience estimate. The same structure that enables automated buying enables honest attribution, and honest attribution is what lets sophisticated, measurement-driven advertisers keep increasing their allocation with confidence. The two capabilities are the same capability viewed from two ends, and both depend on the inventory having a queryable shape.
None of this requires believing anything dramatic about the far future of screens or agents. It only requires noticing something concrete and already underway: a large, growing, high-trust physical medium is finally getting the same data treatment every digital channel received years ago, and the real AI opportunity there is mostly about reasoning over structure rather than generating on top of it. The medium was never behind because it was weak. People still move through physical space, still look up, and still cannot skip a wall. It was behind because it had no machine-readable interface, and that is precisely the gap being closed now.
So if you cover AI in advertising, my suggestion is to spend less of your attention on the next generation demo and more on the unglamorous question of which players are actually structuring physical inventory into queryable, real-time data. That is where the durable advantage in this channel is being built, quietly, while everyone else argues about how many headline variants a model can produce before lunch. Generation grabs the headline. Structure keeps the position. In a medium that is only now becoming programmable, the position is the thing worth having.
