The Future of Programmatic Is Being Decided Right Now
And the industry continues to focus on the wrong fights.
By Navneet Singh VP, Managing Director of Programmatic | Product Contributor: Chris Gum, VP Data Solutions
The programmatic industry thrives on a new obsession every few months, and the latest battleground is the DSP arms race. Who’s winning and who’s not, has been thrown into the spotlight by Publicis Groupe’s recent decision to no longer recommend The Trade Desk to its clients. Before that it was Amazon vs. The Trade Desk. While this is today’s news, the trade press is full of others: Are SSPs undercutting DSPs? Is OpenPath a genuine innovation or another revenue lever? Can IAB Tech Lab’s initiatives co-exist with AgenticAdvertising.org and those popularizing AdCP?
While the industry remains fixated over these questions, they are fundamentally the wrong ones. We are arguing over the plumbing while ignoring the architecture.
This isn’t about which platform wins the DSP arms race. The real question is who is in control. The focus on DSPs is a dangerous distraction, because the entity with the most to lose—the one whose budget is on the line—is not the platform, but the buyer.
The real question is: Who Controls the Intelligence Layer?
The question is not “who has the best platform or the lowest cost or take rate” but who owns the decisioning engine layer on top. The focus should be on the layer that determines which signals matter, which models to deploy, which platforms to activate against, and how to create a feedback loop that compounds learning over time. That is the fight that matters.
The DSP is not the Strategy. The Intelligence that Commands it Is.
DSPs deserve more credit than the current discourse gives them. The engineering required to process millions of bid requests per second, evaluate inventory across thousands of advertisers simultaneously, and clear auctions at sub-hundred-millisecond latency is a huge asset for the buyer. Think of a DSP’s processing power as a massive radar – the scope of what it can see and evaluate in real time is unlike anything else in the advertising ecosystem.
But that radar has a structural constraint. A DSP is not just representing you; it is representing thousands of advertisers at once, allocating QPS across all of them, pacing budgets, managing competing auction dynamics. When the system identifies a high-value impression, it has to decide which of its many clients gets the shot – and that decision is governed by the platform’s economics, not yours.
For years, this was the best arrangement available. The platform you picked determined your data strategy, supply path, optimization methodology, and measurement framework. The entire agency operating model in digital orbited around it. That era is ending – not because DSPs have stopped being useful, but because the strategic decisions are migrating above them. Which audiences to pursue, what outcomes to optimize toward, how to score users, how to learn from one cycle and feed that back into the next – those decisions increasingly belong to a layer the DSP does not own.
Meanwhile, the lines between demand side and supply side are collapsing. SSPs are bringing ad servers, bidders, and containerized execution environments to the table – FreeWheel’s Buyer Cloud, Magnite’s SpringServe and ClearLine product suite, PubMatic’s Activate, and the containerization model Index Exchange pioneered. Buyers now have more surfaces to deploy decisioning than ever before. That’s a good thing – but only if you have an open orchestration system that connects all of those surfaces rather than treating each one as another disconnected silo. The proliferation of execution endpoints only reinforces the central point: the bidder is not where value compounds anymore. The intelligence layer above it is.
Everyone Has a Platform. Not Many Are Building the Layer That Matters.
In practice, the intelligence layer means building what we think of as an open operating system: a cross-platform signal engine that includes a platform, data architecture, identity spine, feedback mechanisms, human intelligence, and trained talent activating it. It trains predictive models on real business outcomes – actual conversions, revenue events, lifetime value – and distributes those scores into whatever execution surface makes the most sense. Amazon’s commerce engine for retail brands. Google’s intent infrastructure for search-driven performance. Microsoft’s open web supply. Magnite or FreeWheel’s curated CTV inventory. A direct SSP integration through containerized workloads. The platform is secondary. The intelligence is primary.
This is the work being done with HorizonOS and its Blu platform. It is not a DSP replacement – it is a decisioning engine, designed to plug into big tech rather than compete with it. The result: a client’s data is portable, their models are interoperable, and their learnings compound from one platform to the next – because the system is built to be open for the benefit of the marketer, not the agency.
With AI, there is a real opportunity to build smarter, more tailored radars on top of this. Intelligence systems can shape and filter what gets evaluated before it ever reaches the auction – scoring inventory at the supply level, sending more qualified bid requests, making decisions that sharpen demand-side execution by default. The intelligence layer determines the quality of what the bidder even gets to see.
Top-Down AI vs. Bottom-Up Operating System
The predominant AI narrative in the market is top-down automation – summarize the brief, generate the plan and compress the workflow. The pitch is: we are going to use AI to make everything we already do faster and cheaper. In theory, yes. But it focuses almost entirely on the planning layer while never touching the infrastructure underneath. Faster audience queries, cohesive forecasting, and rapid media planning are efficiencies, not solutions to core strategic problems
The agencies approaching this correctly are building from the bottom at a foundational level. They are mapping connection points, ensuring data portability, and building legitimate feedback loops: we ran with this algorithm partner on this platform, here is what we learned, here is how that feeds back into the next model. A client needs to drive new-to-brand acquisition but lacks the data so a commerce data partner feeds in the conversion signal set that gets matched with the platform’s audience graph. The first campaign delivers measurable lift, and those learnings get stored in a persistent data environment, and the next campaign starts with a richer model. Every cycle makes the next one smarter.
The gap between “we have agentic solutions” and “our infrastructure enables AI to make better decisions continuously” is enormous. If you do not have the open operating system – the pipes, the data architecture, the identity spine, and the feedback mechanisms – then all you have is a fun demo with concept slides used over and over again. You may lock in some early wins, but clients will soon realize there’s a mess behind the curtains.
What AI actually does is reduce the gap between testing and risk. Being able to pivot flexibly in decisioning goes a long way. The foundation must come first.
Stop Picking Sides. Start Co-Developing with an Ecosystem of Partners.
When the industry sees an agency scaling investment into Amazon or Google, the immediate read is that they are picking sides in the DSP war. That framing misses the point.
Amazon and Google matter in this next phase because of engineering depth, signal density, and infrastructure that can accept externally built intelligence and operationalize it. We have seen this firsthand. Amazon has dedicated technical engineers working alongside our product teams – co-developing custom bidding solutions, integrating commerce signals into predictive models, and building measurement frameworks that connect media investment to business outcomes. For the advertiser, that translates into models that are trained on real purchase data, measurement that ties media spend to actual revenue, and optimization that improves every campaign cycle – not just within one platform, but across them. That level of co-development is not something a traditional pure-play DSP is structured to offer – not because it lacks ambition, but because its business model serves a different purpose.
Google is on a similar trajectory. Gemini integration powers our internal agent tools. Vertex AI provides infrastructure for predictive model deployment. Custom bidding development on DV360 is accelerating. The partnership extends into the engineering and data science layers where intelligence actually gets built.
Owning the Stack vs. Owning the Decisioning Engine
The legacy agency playbook has been consistent: acquire the technology, internalize the capability, package it as differentiation, turn client adoption into dependency. That playbook is antiquated. Every acquisition adds tech debt. Every proprietary system adds integration cost. Every closed stack limits the speed at which new innovation can be absorbed.
The deeper question is not whether you build technology internally – it is what you choose to build. The organizations that have invested in owning entire execution stacks – proprietary DSPs, proprietary data platforms, proprietary identity graphs – are discovering that the rate of change in the market is punishing that bet. Building a decisioning engine that plugs into the best available infrastructure is fundamentally different from building a closed stack that forces clients into your infrastructure. One compounds in value as the ecosystem evolves. The other compounds in maintenance cost. The product the advertiser pays for should be intelligence that makes their media smarter everywhere it runs – not a proprietary seat that limits where it can run in the first place.
No agency is going to build a world-class AI engineering team from scratch. Those engineers have better options. The talent goes where the problems are most compelling, and the compensation is highest. So the question becomes: do you try to own the engineering and accumulate the tech debt that comes with it, or do you co-develop with partners who have the depth and focus your own people on what genuinely differentiates – understanding the client’s business, governing the models, and driving growth?
The agency industry is going through another transition on the value chain – from creative shops to deal-makers to billable-hours models, and now being forced to behave more like technology companies. The largest organizations can only meet that challenge by leveraging buying power. Independent agencies have flexibility, speed, and freedom. Accessibility is the new power. And in an era where innovation cycles are measured in months, the ability to move fast without asking permission from a tech stack acquired in 2019 is worth more than the stack itself.
Programmatic’s Next Frontier: Built on Architectural Commitment, Not Just DSPs
Programmatic experts have always been the frontier builders of this industry. They connected supply and demand at scale. They built the audience-targeting infrastructure that made addressable advertising possible. They developed the measurement frameworks that tied media investment to business outcomes.
This AI-driven era is the next frontier – and programmatic experts are, once again, the ones best positioned to build it. Not because they will be replaced by agents, but because they understand the systems well enough to deploy agents intelligently. Agents will become increasingly powerful decisioning engines – capable of communicating across platforms, negotiating in real time, and operating across diverse data environments. But the effectiveness of those agents will always depend on the humans who design their objectives, set the guardrails, and know when the machine is wrong.
The programmatic industry has spent a decade optimizing around platforms. The next decade will be defined by who optimizes around intelligence – and by the experts who build and run the systems that make it real.
The bet we are making is on the decisioning engine and the open operating system that powers it. On infrastructure that makes data portable, models interoperable, and learnings compoundable. On big tech partnerships that go beyond media buying into genuine co-development. On an ecosystem that is genuinely open – adaptable, responsive, and built for the benefit of the marketer. Doesn’t matter how many times anyone labels their system “open” in the press release – it needs to be an architectural commitment our clients can test, build around, and hold us accountable to – and in the belief that the industry’s proven builders will define its next frontier