AI marketing budgets face scrutiny as costs outpace ROI

By late 2026, the problem was not whether marketers had embraced AI, but whether their organizations had built the human capacity to absorb it. Duke’s business school survey of several hundred CMOs found that AI use in marketing “has more than doubled in the last two years,” even as “technology adoption is actually moving faster than organizational execution and capability”; at the same time, training budgets declined to only 3.8% of marketing spending, and marketing headcount growth fell by more than 50% compared to the previous year.

That readiness gap mattered once AI became embedded in everyday work: in Basis’s August 2026 survey of 184 marketing and advertising professionals, 61.4% said they use generative or agentic AI daily, yet just 6.5% of organizations measure AI ROI against defined benchmarks, while 65.1% either do not measure it at all or are unsure whether anyone is doing so. With 88.6% now paying for premium AI tools and agent use surging 155.5% year over year, the budgeting implication for 2027 was clear: support people, not just technology — “a really critical thing,” as the Duke survey summary put it.

The pricing architecture behind marketing AI has changed in ways that make costs harder to ignore and harder to predict. As AI Business Models Shift From Subscription To Usage Based noted, “in 2026 85% of entropic revenues came from usagebased models rather than subscription models,” a sign that enterprise AI is moving toward consumption billing because marginal costs are real, not near-zero; once token, API, and inference charges scale with use, marketing teams no longer buy a tool once but absorb an expanding operating expense that can swell with every campaign, workflow, and prompt.

That cost volatility is pushing prioritization toward AI work that can justify itself financially and operationally, while preserving human checks where quality and trust are at stake. The Agile Brand’s discussion of “predictable AI” argued that humans still need to define workflows and keep “checking to make sure that the agents are performing as expected,” while other pricing analyses warned that providers are now passing through token consumption and changing prices frequently, making oversight not just a governance preference but a budget discipline against expensive misuse, hallucinations, and low-value output.

Finance is no longer treating marketing AI as a strategic free pass; it is asking the same hard ROI questions it asks of any other spend. MarTech reported that “AI is getting funded faster than marketers can prove its value,” while “40.2% of martech leaders still can’t show a clear financial contribution from their technology investments,” even though “AI may get special treatment during the investment phase, but it still ends up in the same budget conversation,” and only 10.8% of agent initiatives are fully scaled in production, leaving most programs exposed when budget reviews turn from experimentation to proof. That pressure fits the broader direction of travel Forrester’s Chief Research Officer described for 2027: gaining control over AI costs without limiting growth, a framing that puts finance-led scrutiny at the center of AI budgeting decisions.

That scrutiny is redirecting money toward AI uses that can survive human and financial accountability tests, not just generate activity. MarTech found “Just 1.6% allow fully automated AI-generated customer-facing content,” while “Most (43.3%) allow external use of AI-generated content only after it’s been reviewed, edited, and verified,” and warned, “There’s no increase if time saved on production is spent on verifying the output”; Snowflake made the same case operationally, saying “AI tools like Coco and Cork are really great at generating the first 80%… But, what about the last 20%?... That last 20%, that’s still completely us,” while teams that could prove value were far likelier to win budget growth. The same accountability logic is surfacing in labor decisions too: “We've seen Salesforce laid off over 4,000 people specifically citing AI,” but the speaker adds that “when you've got the… customers complaining… you're going to have to bring some of these people back in,” underscoring how quickly cost-cutting claims can be tested against real business outcomes.