A hard truth is emerging from every AI conversation happening in programmatic right now: more models trained on the same flawed inputs will only produce more versions of the same flawed outcomes. In fact, the speed and confidence with which AI now optimizes makes the consequences of bad inputs much harder to see and much more expensive by the time you notice them.

Programmatic didn’t get here overnight. The intermediaries that built this ecosystem were optimizing long before we called it AI. They were just optimizing for the wrong thing: Not outcomes or advertiser performance but the margin in the middle.

The ANA put a number on it: $26.8 billion in annual programmatic media value lost to non-working spend. That’s far from a rounding error. It’s bigger than the annual revenue of roughly two-thirds of the Fortune 500.

For a decade, adtech sold complexity rather than resolving it. Now, AI is forcing a shift.

The model is shifting

For years, the structure was simple. DSPs represent buyers and SSPs represent publishers. SSPs competed on scale, access, and favorable pricing. Optimization happened downstream—after the buy, after the auction, and, most importantly, after data had already degraded through multiple hops.

By then, you’re optimizing toward the wrong things: Fraud, request duplication, signals weakened by distance and regulation. No algorithm fixes that; it just scales the wrong optimizations.

That's why the center of gravity in programmatic is moving closer to supply. Because that’s where signals are cleanest, where identity is strongest, and where decisioning is most intelligent. Buyers who recognize this shift earliest will have a structural advantage over those who keep treating the supply side as an interchangeable pipe.

What better inputs actually look like

Moving intelligence closer to supply isn’t abstract. In practice, it means three things:

1. Proximity. The closer you are to inventory, the more complete and real-time your signals. No QPS caps means an unfiltered view of supply, which means you can define quality on your own terms, not someone else’s pre-filtered universe.
2. Data fidelity. Today, a lot of identity is stitched together downstream, after data has already been lost or modeled away. Identity resolved at the supply layer—built directly from publisher relationships and real user signals—is more accurate, more addressable, and more durable.
3. Supply integrity. Two nodes in the supply chain or fewer improve performance. Every additional intermediary is a point where signal degrades and auction integrity erodes.

The thinking behind OpenXSelect is a control layer where buyers define what quality means for their business before the auction, not after. And with OpenXBuild, buyers can bring their own data, models, and decisioning logic directly into the supply layer—shaping how supply is evaluated and how audiences are constructed upstream.

Start earlier

Performance isn’t determined at the optimization layer or after during reporting. It’s determined at the point where supply is defined and intelligence is applied.

You can’t optimize your way out of bad supply, weak signals, or lost data. You have to start earlier. That’s where intelligence lives and where the next phase of programmatic is being built.