Is AI a threat to DSPs?

Short answer: Absolutely.

Longer answer: In multiple ways.

As is my wont, I’ll start with the conventional wisdom. Implementation of AI throughout the creative→planning→execution→measurement loop will be transformative, and will significantly benefit parties that control the entire process and collect data accordingly. Google and Meta are names that come to mind. While a point solution like AI creative might be awesome, without the feedback loop of a complete media execution it will be out competed by larger data panopticons (see Is AI killing creative tech?).

That’s cool stuff and I think its largely correct. But in the same way the Holy Roman Empire survived for a thousand years after the fall of Rome, the “open web” in my view will remain the bastion of independent tech and messy non-conforming AI stories for some time (see Trimming the walled gardens ).

For the sake of argument, let’s assume I’m right that the whole open web thing doesn’t go under or get absorbed by big tech. How does AI potentially upend the current ad tech market structure? I think the answer has a lot to do with curation (see: Curation is the anti-SPO, plus Audigent interview. ).

While I was running Beeswax I liked to tell people that the modern DSP was one of the most complex categories of software ever invented. It both requires massive scope of features, as well as massive scale for those features. A DSP might have to support some form of ad serving, creative management, data management, reporting and attribution, fraud detection, and more, all at millions of transactions per second.

The role of the DSP might be represented something like this:

This was cool for a while. But since the mid-2010s and the advent of header bidding the supply side has been rife with duplication. As a result, in the web world you could treat the leading SSPs as interchangeable. This was bad for SSPs as a first order effect, but also bad for DSPs since it removed generic inventory access as a competitive advantage. SSPs then started adding data directly, which resulted in the current paradigm of “curation” (highly simplified below):

What does this have to do with AI? AI, and especially “Agentic AI” (which is just AI that actually does something), is a shot across the bow from DSPs. Because the promise is that Agentic AI agents (is that redundant?), do the targeting and optimization for the client, and more so, do so within any touch point the client desires. This could lead to a diagram like this:

You’ll also note that the boxes that are still left on the right side of this diagram (attribution, creative) are generally ones where specialist vendors offer better products than most DSPs already.

Whenever I write one of these types of articles folks in the comments come to what I would call the “Taco Bell Defense.” This is a riff on the joke I like to tell: “Ad tech is like Taco Bell, the same seven ingredients served in hundreds of ways.” Essentially the feedback is that I’m just moving boxes around on a diagram and none of it matters; life will find a way. OK sure, here’s why I think Agentic might be different:

The signs to look for in this transition would be:

Let’s keep our eyes out.