A survey of 250 marketing and creative agencies ($1M to $50M in annual revenue, split across the US, Europe, and Asia-Pacific) found that 41 percent have shipped at least one AI agent to production, up from 9 percent a year earlier. Another 58 percent are still piloting. The adoption curve moved fast. The economics underneath it did not move together.
The return gap is wider than the adoption curve suggests. Among agencies running agents in production, median return on investment is 3.2x. The top decile of deployments returns 11x. The bottom quartile sits at 0.7x, below the point where the agent pays for its own token spend. Returns also vary sharply by task: SEO audits return 11.4x, code generation and refactoring return 8.3x, lead qualification returns 5.8x, and client-report drafting returns 1.6x, the lowest of the four workflows the survey measured directly.
The most common production use isn’t one of the four anyone measured. Content brief and outline generation is the single most widely deployed agent workflow in production, used by 64 percent of the agencies surveyed, ahead of SEO audits (51 percent), ad-copy iteration (48 percent), client-report drafting (39 percent), and lead qualification (27 percent). It is also the one workflow the survey’s ROI breakdown does not cover. Median token spend per agent runs $1,800 a month, with a tenfold gap between the top and bottom spending quartile, so agencies are not spending evenly either. The workflow nearly two-thirds of agencies run most is also the one with no published number attached to whether it pays back.
The same split shows up on the brand side, under different numbers. A separate industry roundup puts AI agent usage among marketing professionals at 90 percent, with 93 percent using agents specifically for content generation. But only 34 percent of enterprise marketing teams let an agent act without a human approving the output first, the difference between a team that has adopted AI and a team that has handed an agent an unsupervised budget decision. A third data set, built from roughly 15,000 businesses, puts marketing-campaign automation adoption at 45 percent among enterprise and mid-market companies, with real measured lift where it has taken hold (27 percent faster campaign builds, 19 percent lower cost per lead). Three separate surveys, three different populations, and the same shape each time: the headline usage number is roughly double whatever number describes agents actually running unsupervised in production.
What that means for anyone reading an adoption stat. A claim that 90 percent of marketers use AI agents describes who opened the tool, not who trusts it with money. The number worth planning around is closer to a third, and even inside that smaller group, a quarter of deployments have not earned back what they cost. Before copying a competitor’s AI-agent rollout, the more useful question is which specific workflow they automated, since the most heavily adopted one in this year’s data is also the one nobody has published a return on yet.
Sources referenced in reporting: Digital Applied, Agentic AI Adoption: 250-Agency Survey 2026, The Stacc, AI Agent Adoption in Marketing 2026, First Page Sage, Agentic AI Adoption Statistics.