Programmatic efficiency in the age of AI
In the programmatic ecosystem, efficiency means getting more out of every dollar, second, and interconnected platform. In short: less waste and better results.
Concretely, that means two things.
- More of every campaign budget should go to working media—the part that drives reach and results—rather than getting lost in fees and transaction costs, or what is sometimes referred to as the ad tech tax.
- Each step in the path should add value, not friction. Every unnecessary call or duplicate request adds up. It’s like paying a toll on a road that you didn’t need to take.
Programmatic efficiency: the infrastructure cost
Every second, billions of real-time auctions determine which ads appear on screens around the world. Each auction pulls in thousands of data points, multiple decision-makers, and split-second responses.
The volume of digital activity only continues to grow, which creates more pressure for everyone. Buyers expect more outcomes for the same dollar, media owners are optimising every impression to protect yield, and tech platforms are being pushed to do more with less.
That pressure exposes a cost we don’t always talk about: the infrastructure cost. Every extra hop, every duplicate request, and every missed impression create an opportunity cost.
It’s easy to gloss over because it’s the cost of doing business. But in aggregate, it’s lost value that could have gone to working media, increased product investment, or simply lowered prices in the market.
Historically, the programmatic model has worked a bit like an airport currency exchange—changing conversion rates, hidden spreads, and high fees on top. You start with a dollar, but then you get back far less on the other side.
As the market matures, the goal is for it to look more like Visa—a unified, globally scaled system that translates to low, predictable fees. As budgets are facing more scrutiny, expectations are shifting toward transparency and a clear value exchange.
The race for space
AI is excellent at creating efficiency and improving outcomes. So, it’s no surprise today that there are several companies bringing next-level modeling and custom bidding algorithms into the ecosystem, training neural nets on vast datasets across the entirety of the open internet to drive smarter decisions and better outcomes.
The reasoning capabilities of these models allow them to predict which impressions are most likely to achieve a desired outcome. This helps ensure that brands reach the right audience, in the right moment, with the right message, without losing budget to inefficiencies.
We’re already seeing that access to AI models is getting easier, and in some cases, less expensive. But, AI also increases compute demand. The surrounding costs of that—electricity, GPUs, and the cost of cooling data centers—are rising.
AI’s momentum has kicked off a 21st century space race. Rack space, power, and cooling are now strategic resources.
A decade ago, the public cloud was enticing: incredible convenience, elastic scale, attractive pricing. Today, convenience remains, but AI-era workloads have made compute and storage meaningfully more expensive.
Every request you ship to the cloud and back pays a toll, whether or not it clears a transaction. AI essentially turns a software problem into a physics problem—everything has a cost in terms of energy consumed, so data transit results in real work cost implications.
And when you multiply that cost across hundreds of billions of auctions a day, it’s not a rounding error. It’s a business model. That’s why inefficiency is less tolerable than ever.
For many businesses, it’s simply becoming too expensive to continue scaling this way.
What AI means for ad tech’s future
So, what does this mean for the next chapter of ad tech?
1. Sell-side decisioning: More businesses are choosing to apply algorithms, models, and decisioning on the sell side. This is closer to the impression where higher-fidelity signals are freshest, and noise is lowest. SSPs already ingest signals from across the open internet. Adding the capability to decision and curate across that full universe of inventory means supply is strategically activated before it’s even sent to buyers.
That shift can reduce the QPS, or queries per second, that DSPs have to process, allowing them to make better use of resources. It also drives better win rates and outcomes for marketers because supply is smarter from the start.
2. Containerization: There’s now an opportunity for businesses to containerise their algorithms or code. Instead of shipping it off to a public cloud, they can run it right on top of an exchange or SSP—just like how collocating servers and infrastructure at a stock exchange allows traders to execute orders faster.
Containerisation eliminates the cost of redundant infrastructure by making better use of what already exists. It moves these algorithms closer to the exchange, speeding up auctions and cutting down the latency and the costs associated with sending signals back and forth to the cloud.
3. Programmatic best practices: Finally, it’s as important as ever to master the programmatic basics. Supply-path optimisation improves performance. So, be sure to prioritise direct paths and transparent economics and look to strengthen quality and fraud controls to protect working media. These fundamentals compound with AI: Cleaner inputs make models smarter, and smarter models make clean paths even more valuable.
An efficiency-first framework
If we want more of the next dollar to reach high-quality media owners and deliver better outcomes for marketers, we need to be smarter about taking advantage of the resources and utilities so that resources are prioritised towards working media.
AI can help us do that by boosting programmatic efficiency, improving predictions, and letting us operate with fewer, faster, more efficient queries.
But to capture that benefit, we all must operate using an efficiency-first framework and be intentional about how we design for proximity, performance, and scale.
Learn more about programmatic efficiency and why every millisecond matters.
Thank you to Josh Prismon, who also contributed to this video.
