Less than 60% of CTV buyers have high confidence in knowing where their ads actually ran. That number comes from IAB's 2026 Digital Video Ad Spend & Strategy Report, and it should alarm anyone signing off on programmatic video budgets. We are spending more on connected TV than ever before, trusting AI systems to make millions of bidding decisions per second, and yet the majority of buyers cannot verify the legitimacy, source, or placement of their impressions with confidence.
This is not a measurement problem. It is a governance problem. And it will land on your desk the moment your CFO asks why you cannot explain what the algorithm actually did with the budget.
The Glass Box Imperative
The industry has started calling this shift glass box versus black box. Melissa Burdick, President at Pacvue, framed it clearly after CES 2026:
AI should remove the friction. It should not remove the ownership.
Melissa Burdick, President at Pacvue
Amazon's VP of Ad Sales echoed the same principle: automation plus visibility, not automation instead of visibility.
The distinction matters because programmatic video has become the default infrastructure for premium inventory. US programmatic spend reached $134.8 billion in 2024, up 18% year over year, with CTV absorbing an increasing share. Nearly 90% of CTV ad spending is already programmatic, according to eMarketer data cited by Mediasmart. The scale is no longer experimental. The accountability expectations should match.
Yet the systems making these decisions remain opaque. As Lemma Technologies noted in their analysis of explainable AI in programmatic, marketers see what works but not why it works. Why did the system bid $X for one impression and skip the next? Why was one user shown a premium automotive ad while another was not? These are questions most systems cannot answer because they operate inside a black box.
What Buyers Actually Need to See
The IAB's 2026 report identifies a specific trust hierarchy that programmatic video buyers are missing. Fraud is the leading driver of mistrust, but buyers are less likely to pay a premium to prevent it. What they will pay for is campaign performance and accountability, which means sellers need to invest in tools that explain which audiences drove outcomes, not just that outcomes occurred.
This is the operational gap: outcome delivery alone does not guarantee streaming partner success. Targeting capabilities and audience delivery now rank above business outcomes in why buyers cut streaming partner spend. Sellers who cannot pinpoint which audiences drove results will lose budget to those who can.
The measurement challenge compounds as AI takes over more of the execution. Experian's analysis of agentic AI in programmatic describes systems that interpret signals, suggest next steps, and enable action within defined parameters. That sounds efficient until you realize the parameters themselves are often invisible to the humans accountable for the spend.
The Explainability Layer
Explainable AI, or XAI, is the technical response to this visibility gap. Lemma's framework describes XAI as AI that can justify its decisions, identifying which data points influenced a choice and to what extent. Techniques like LIME and SHAP allow marketers to trace exactly which signals justified a bid price, turning optimization from guesswork into auditable strategy.
The practical applications are straightforward:
- Bidding transparency means explaining every dollar spent
- Creative optimization means data-backed storytelling, not magic
- Audience selection means showing which behavioral signals triggered which decisions
None of this requires abandoning automation. It requires instrumenting automation so humans can verify, course-correct, and learn.
IAB's AI-Powered Video Outcomes series, launched in March 2026, addresses this directly. The first edition breaks down how agentic AI works, where it is already appearing in the ecosystem, and the operational changes organizations must prepare for as autonomous systems begin to influence media decisions. The framing is explicit: AI-driven execution with human oversight and guardrails, not AI-driven execution instead of human oversight.
The Measurement Infrastructure Gap
The visibility problem extends beyond individual campaigns to the measurement infrastructure itself. IAB's State of Data 2026 report examines how marketers are navigating a measurement ecosystem that is, in their words, fundamentally broken. Privacy changes, fragmented proprietary platforms, and inconsistent cross-channel approaches limit the ability to connect media exposure to business outcomes.
Braze's analysis of attribution challenges puts it bluntly: attribution is less reliable because journeys are fragmented, privacy reduces observable signals, and identity breaks across devices. The practical shift is toward first-party data, journey-level measurement, and experimentation to prove lift.
For programmatic video specifically, this means the AI making bidding decisions often operates on signals that cannot be independently verified. The system reports success, but the success cannot be triangulated against external data. That is not a technical limitation. It is a governance risk.
What a Pilot Looks Like
If you are running programmatic video at scale, here is a two-week pilot to surface the visibility gaps in your current stack:
First, request decision logs from your DSP for a single campaign. Ask specifically: which signals triggered which bids, at what price, for which impressions? If the answer is we do not provide that level of detail, you have identified your first gap.
Second, compare reported placements against your brand safety and fraud detection tools. The IAB report found that even within the most trusted CTV buying methods, less than 60% of buyers have high confidence in placement legitimacy. Your own data will tell you where you fall on that spectrum.
Third, run a holdout test on a single audience segment. Suppress AI optimization for that segment and compare performance against the AI-optimized cohort. The delta tells you what the AI is actually contributing. If you cannot run this test, you cannot verify the AI's value.
The risks are straightforward: your DSP may resist providing decision logs, your holdout test may show the AI is not adding as much value as reported, and your placement verification may reveal gaps you did not know existed. All three outcomes are better than not knowing.
The Board Question
The CFO question is coming. It will sound like this:
We spent $X million on programmatic video. Can you explain what the algorithm did with that budget and why?
If your answer requires trusting the system without visibility into its decisions, you have a governance problem. If your answer includes decision logs, holdout tests, and independent verification, you have a defensible position.
ExchangeWire's analysis of the 2026 ad tech landscape frames the stakes clearly: transparency is now a non-negotiable mandate. The industry is moving from raw identity to intelligent signals, and AI is no longer just an optimization tool but the operating system of the industry. Operating systems require visibility. They require audit trails. They require the ability to explain what happened and why.
The programmatic video systems that win budget in 2027 will be the ones that can answer the CFO's question. The ones that cannot will find their budgets reallocated to channels where accountability is possible. Model or it did not happen.
