Earlier this year, one company ran up a $500 million AI bill in a single month after management forgot to set a usage cap. Uber burned through its entire 2026 AI coding budget by March. On the agency side, Digiday reported that holding companies are routing AI infrastructure costs through principal media deals, meaning some CMOs are already paying for AI they cannot see on any invoice. The spending story and the returns story have been running on separate tracks for two years. They are starting to diverge in ways that are hard to ignore.
AI Spending Has Outpaced AI Accountability
Marketing AI budgets have expanded without a parallel expansion in how those budgets get measured. A significant share of marketing AI investment is evaluated on faith, not evidence.
Part of the reason is that AI is often sold as infrastructure rather than as a specific capability with a specific job. “We use AI across our entire platform” is a common claim. It is also nearly impossible to audit. When AI is described that broadly, there is no defined output to evaluate and no accountability mechanism if performance does not improve.
Most companies are deploying their most powerful and expensive AI models on tasks that do not require that level of sophistication. Matching the model to the task, rather than defaulting to the most capable one available, is where AI spending starts to make economic sense. For marketing organizations, that means evaluating an agency on whether they can match specific AI capabilities to specific jobs, and show what each one produced.
Three Questions Every CMO Should Ask About AI Performance
Measuring AI return in paid media comes down to three questions:
- What decision did AI inform?
- What would the decision have been without it?
- What did the difference produce in performance terms?
Answering them requires granular, unified data across every channel, organized in a way that traces a specific AI-driven action to a specific outcome. Without that data infrastructure, AI accountability becomes a conversation about process rather than results. You can describe what the AI is doing. You cannot show what it produced.
An AI tool that flags a budget pacing issue is useful. Whether it is worth what it costs depends on how much faster it flagged the issue compared to a human analyst reviewing a dashboard, how much spend was protected as a result, and whether that happened consistently across accounts and campaigns over time. None of that is visible in a weekly performance report.
How TrueIntelligence Makes AI Accountable
TrueIntelligence is the AI platform True Interactive built to run underneath everything we do across client accounts. Each of its capabilities is designed for a specific job: anomaly detection, budget optimization, creative performance monitoring, audience discovery, cross-platform reporting, competitive intelligence, conversion tracking validation, and growth forecasting.
Individually, none of these capabilities are surprising. Together, they cover the full span of a paid media program, from the first data point collected to the next budget decision made. Because each capability has a defined job, each one has a defined way to measure whether it is doing that job. When Anomaly and Trend Detection flags a performance issue, we can track how many hours earlier that flag came compared to a human-reviewed dashboard, and what the spend impact of that timing difference was. When Predictive Budget Optimizer recommends a reallocation, we can compare outcomes against the prior allocation method. The accountability is built into how each capability is designed.
Clients do not pay extra for TrueIntelligence. The platform is part of what every True Interactive account gets, which means the ROI question is embedded in the client relationship itself. If TrueIntelligence is working, it shows up in results. If it is not, that is visible too.
How to Hold Your Agency Accountable for AI Spending
AI spending in marketing is not going to decrease. The tools are proliferating, the capabilities are expanding, and the pressure to deploy AI is coming from every direction. What is not keeping pace is the discipline around measuring what it returns.
Ask your agency something specific: for each AI capability running on my account, what is the defined output, how is it measured, and what has it produced over the past 90 days? A clear answer means the accountability is there. A general answer about how AI is improving performance across the board means it probably is not.
If you want to understand how True Interactive measures the performance of TrueIntelligence on your account specifically, reach out to us here.
Lead photo source: Immo Wegmann on Unsplash
