Contextual Targeting Is Just the Beginning

At VideoByte we built scene-level contextual targeting for CTV. Not “this is a cooking show” but “this scene, right now, is about this,” and the ad slot next to it could know that. It worked. The thing I keep thinking about since is how small a piece of the puzzle it turned out to be. Solving it mostly revealed how many problems were standing in line behind it. The acquisition came later; this post is about what the technology didn’t finish.

Contextual answered where an ad should go. It said nothing about what the ad should be. The creative was still fixed months in advance, made for a demographic average, dropped identically into every context we so carefully identified. We could tell you the scene was a road trip at golden hour, and then serve the same spot that ran against a cooking segment that morning. That mismatch is the tell for where this goes next: creative that adapts to context in real time. Not a different ad chosen from a pool of five, but an ad assembled for the moment it runs. The pieces exist now. Generation is cheap, and I’m building some of this at Adwave, which is why I can be concrete where most “future of advertising” pieces stay hand-wavy.

Past creative, the list gets longer. Interactive formats that TV has promised and failed to deliver for twenty years become plausible when the creative can respond instead of play. Measurement is still weaker than the industry pretends: attribution models that everyone uses and few fully trust. AI may help less by measuring better than by making experiments cheap enough that you can simply test instead of model. And identity resolution across platforms remains the swamp it’s always been, where the constraint isn’t technical capability. It’s privacy, regulation, and the platforms’ unwillingness to share, none of which a better model fixes.

Now the caution that has to be in this post, because I’ve watched this pattern repeat: every one of these capabilities is currently being described as inevitable, and TV advertising has a long history of inevitable things not happening. I sold into this industry for years. The gap between what’s possible and what a media buyer will actually adopt is the most reliable fact in it. Scene-level targeting itself was possible long before anyone bought it. The buyer’s workflow, the agency’s incentives, the brand’s risk tolerance, those move on a clock that has nothing to do with the technology’s clock. Betting a company only on the technology clock is how you get surprised.

The open ending, because the honest state of things is open: the interesting problems in TV advertising are no longer about knowing things. We know plenty. We know the scene, the audience, the moment. The problems are about making things, at the speed and specificity the knowing already supports. That inversion, from knowledge-constrained to production-constrained, is new. Nobody has fully priced it in, including, probably, me.