Adtech Tax: AI agents reshaping the path of media money spends

The economics of digital advertising is being quietly rewritten. For over a decade, marketers have operated within a system where a significant share of programmatic budgets never reaches the publisher. What was once rationalised as the cost of automation is now being scrutinised as a deeper structural inefficiency that affects profitability, control and trust.

That scrutiny is intensifying as the ecosystem itself is undergoing rapid change. Privacy shifts have weakened signal clarity. Supply paths are being rationalised. Platforms are consolidating influence. And now, agentic AI is entering the system with the promise of making media buying smarter, faster and more autonomous.

Read more: Big Tech earnings signal AI expansion

But beneath that promise lies a more uncomfortable reality. The same technology that claims to remove inefficiencies could also introduce new layers of cost and opacity. The question facing marketers today is no longer whether programmatic works. It is whether their money is working as hard as they think it is.

The shifting shape of adtech leakage

The long-cited estimate that nearly a third of programmatic spend is lost in the supply chain is not disappearing. It is evolving. The nature of inefficiency is changing from visible intermediaries to less visible issues around inventory quality, duplication and signal loss.

Rajiv Dingra, Founder and CEO at ReBid, points out that the narrative around 30 to 40 per cent leakage remains directionally accurate but requires nuance. “The more defensible 2025–26 benchmark is around 25 to 30 percent inefficiency depending on how you define waste. If you include invalid supply, signal loss and misattribution, the real effective waste still trends closer to 30 percent plus in many large advertisers,” he says.

Read more: AI taking over human roles in creativity?

The shift is not necessarily towards a cleaner ecosystem but a more complex one. Earlier, inefficiencies were easier to identify across multiple intermediaries such as SSPs, exchanges and resellers. Today, they are embedded in the quality of supply itself. Low-quality inventory, MFA environments and AI-generated content are adding a new layer of dilution that is harder to detect.

Gopa Menon, COO and Co-founder at TheBlurr, offers a sharper distinction between different types of buyers. “Verifiable waste meaning spends with no traceable path to a legitimate impression sits closer to 18 to 25 percent for buyers with even moderate hygiene practices in place. The bigger story is the widening gap between sophisticated buyers and those who have not done the work on supply path optimisation,” he says.

This divergence is creating a two-speed market. While on one end are advertisers actively managing supply paths and reducing leakage, on the other are those still exposed to legacy inefficiencies that continue to push losses closer to 35 to 40 per cent.

Read more: Brands rethink adtech as FY26 closes

Agentic AI adds cost before it removes it

Into this already complex system comes agentic AI. Its promise is compelling. Autonomous systems that can plan, optimise and execute campaigns in real time without human intervention. The expectation is that this could reduce inefficiencies by eliminating redundant layers and improving decision accuracy.

The reality, at least in the near term, is more layered.

Agentic systems introduce new cost structures that marketers are only beginning to understand. These include compute and model inference costs, data unification and storage, access to premium data environments such as clean rooms and identity graphs, and orchestration layers that enable agent-to-agent workflows.

Dingra estimates that early deployments add roughly 5 to 10 per cent incremental cost to media. “Agentic AI does not eliminate cost. It restructures it. Brands are effectively trading 25 to 30 per cent waste for around 10 percent structured AI and data cost with higher signal fidelity and better decisions,” he explains.

Menon’s assessment is closely aligned. "Buyers should expect an incremental 4 to 8 per cent added to their total media cost if they are running genuine agentic systems. Whether that is worth it depends entirely on the performance lift, and the measurement frameworks to answer that are still evolving,” he says.

This creates a transitional phase where costs may rise before efficiencies are realised. The key variable is not the technology itself but how it is implemented.

Compression or complexity depends on execution

The debate around whether agentic AI will simplify or complicate the supply chain does not have a single answer. Both outcomes are already visible in the market.

In the short term, complexity is increasing. AI layers are being added on top of existing stacks, creating additional cost without necessarily removing inefficiencies. In such cases, AI becomes another participant in the supply chain rather than a replacement for it.

Over the medium term, the potential for compression becomes clearer. Dingra highlights three ways in which this could happen. AI agents can reduce redundant decision layers, enable supply path optimisation at machine speed, and shift budgets towards curated marketplaces and direct publisher relationships.

Menon reinforces that the outcome is determined by structural choices rather than technological capability. “Buyers who use agentic systems to rationalise their supply chain are seeing real efficiency gains. Those who add AI as another layer are spending more for a similar outcome,” he says.

This distinction is critical. The industry is not moving uniformly towards efficiency. It is fragmenting between those redesigning their ecosystems and those layering new tools onto old models.

Budgeting models are being rewritten

The financial implications of this shift are beginning to show up in how marketers allocate budgets. The traditional distinction between working and non-working media is becoming harder to define. A growing share of spend is being redirected towards technology, data infrastructure and AI capabilities. These investments are necessary to operate in a signal-constrained, AI-driven environment, but they also reduce the proportion of budgets directly spent on media.

Vijay Shenoy, Deputy Vice President at LS Digital, points to the inevitability of this shift. “The moment you involve more tech stacks or layers onto your existing ecosystem, there is an added cost and that cost is ultimately borne by the brand. In the initial quarters, it may not be clearly visible in returns,” he says. This introduces a lag between investment and measurable outcomes. Brands may need to absorb higher upfront costs before efficiencies materialise. Over time, Shenoy expects this to stabilise. “From a long-term perspective, they would break even eventually, but in the short term it is an added cost,” he adds.

Pankaj Sharma, CEO at MGID India, frames it as a trade-off rather than a cost escalation. “Nothing comes for free. If you use agentic AI, there will be some form of subscription or cost involved. The idea is to replace certain services and reduce inefficiencies, but there will still be a cost layer,” he says.

This shift is forcing marketers to rethink how they evaluate ROI. Efficiency is no longer just about CPM or CPA. It is about the total cost of decisioning, including data, technology and execution.

The changing role of intermediaries

One of the most closely watched questions is whether agentic AI will eliminate intermediaries or simply reshape their role.

The expectation that agencies could be disintermediated is not playing out straightforwardly. Instead, the ecosystem is undergoing a structural reorganisation where fewer players are expected to do more, even as their roles evolve from execution to advisory and orchestration.

Vijay sees a clear transition underway. “The agency ecosystem will not disappear but its role will change. Media buying and selling as we know it will evolve, and agencies will increasingly act as consultants helping brands navigate this complexity,” he says.

At the same time, the number of partners in the ecosystem is reducing, but their strategic importance is increasing. Rajiv points to a clear shift in how global advertisers are restructuring partnerships. “We are seeing consolidation over fragmentation, with brands moving from managing 10 to 15 vendors to working with 3 to 5 strategic partners across DSP, data, measurement and activation,” he says. This consolidation is not just about efficiency but also about reducing duplication and improving accountability across the supply chain.

Full in-housing, however, is not emerging as the dominant model. Instead, a hybrid approach is taking shape. “Brands want control of data and measurement but still rely on partners for execution, AI and optimisation layers,” Dingra explains. This reflects a growing recognition that while ownership of signals is critical, execution in an AI-driven ecosystem still requires specialised capabilities.

Alongside this, platform concentration is intensifying. More budgets are flowing into a smaller set of scaled ecosystems such as Google, Amazon, retail media networks and connected TV platforms. The open web, in contrast, is becoming increasingly curated rather than truly open, as advertisers prioritise quality and signal integrity over scale.

Perhaps the most significant shift is the emergence of what Dingra describes as “agentic partners.” These are not traditional agencies or standalone tools, but integrated entities combining platform capabilities, AI-driven decisioning and execution. “It is a new category that sits at the intersection of technology and services,” he notes.

The result is a paradox. While the number of intermediaries may reduce, the control exerted by those that remain is likely to deepen. Intermediaries are not disappearing. They are being redefined, fewer in number, broader in scope, and more deeply embedded in the decision-making layer of advertising, with greater influence over how media money ultimately flows.

Transparency improves or becomes harder to decode

There is a competing narrative emerging around transparency. On one hand, automation and agent-to-agent communication could reduce human bias and improve clarity in decision-making.

Shenoy is optimistic about this possibility. He believes that increased automation could lead to greater transparency and trust in programmatic systems.

On the other hand, the rise of AI-driven decisioning introduces a new form of opacity. Algorithms make decisions at a scale and speed that are difficult to audit in real time. The system may become more efficient but less interpretable.

This tension defines the current moment. The industry is moving from a system that is complex but partially auditable to one that could be efficient but harder to fully understand.

A new equilibrium is still forming

What emerges from these shifts is not a clear reduction or increase in the adtech tax but a redistribution of it.

Some traditional inefficiencies are being reduced through supply path optimisation and consolidation. At the same time, new cost layers are being introduced through AI, data and infrastructure. Sharma suggests that even in a more efficient future, costs will not disappear entirely. “Maybe costs reduce to around 20 percent, but there will always be a layer of cost involved,” he says.

The more important shift is qualitative rather than quantitative. The industry is moving from visible fees to embedded costs. From human-managed inefficiencies to machine-driven trade-offs. From fragmented ecosystems to more concentrated ones.

For marketers, this changes the nature of control. It is no longer just about negotiating rates with partners. It is about understanding and influencing how algorithms allocate spend.

The next phase of programmatic advertising will not be defined by whether AI works. It will be defined by whether marketers can keep pace with how it changes the flow of money.

The adtech tax is not going away. It is becoming more sophisticated. And in that shift lies the real challenge. Not just to reduce waste, but to ensure that in a system increasingly run by machines, every rupee spent still has a clear path to value.