Adoption numbers for agentic AI in marketing look impressive until you check what is actually running underneath them. Supermetrics’ 2026 Marketing Data Report, which surveyed 435 marketers between October and December 2025, found that 13% of marketers currently use agentic AI, and 82% of those using or planning to use agents expect major or moderate ROI improvement from it. McKinsey’s November 2025 research lines up with that: 23% of organizations say they are already scaling agents in at least one function. On paper, agentic marketing has moved past the demo stage and into real deployment.
Then look at what the same research says about the data those agents are supposed to run on.
The plumbing problem
52% of marketing teams told Supermetrics they don’t own their data strategy. 37% said a lack of integration between their analytics and activation tools is actively blocking data activation. Only 33% feel confident they can activate the data they already have, and barely 41% feel confident analyzing it in the first place. An agent can only act as fast as the pipes feeding it, and for more than a third of marketing teams, those pipes, not the model, are the actual limit.
Figures compiled by Omnibound’s 2026 agentic AI statistics roundup describe the same gap from another angle. Marketing organizations need an average of 7 data sources to support agentic marketing, yet fewer than six in ten marketers report complete access to service data (58%), sales data (56%), or commerce data (51%). Only 1 in 5 companies have a mature governance model for autonomous agents at all, according to Deloitte’s State of AI in Enterprise 2026 survey of 3,235 organizations. Gartner puts a number on where that leads: it expects more than 40% of agentic AI projects to be canceled by the end of 2027, and names unclear ROI, rising costs, and weak governance as the reasons, not model performance.
What is actually working
Chief Content Marketer’s account of agents already running production marketing work backs up that diagnosis from the opposite direction. Where agents deliver real time savings, lead enrichment agents saving 15 to 20 hours a week, analytics and reporting agents 12 to 18 hours, campaign optimization agents 10 to 15, the teams behind them share one habit: they documented the process before they automated it. The piece singles out “delegating before documenting” as the most common failure, along with teams that pull human review before error rates drop below 5%, and teams that tune agents for cost-cutting instead of output.
None of that points to a model limitation. The same agents already produce 22% higher lead-to-opportunity conversion and 15 to 30% better cost per acquisition when the process and the data behind them are sound. The agent is not the constraint. The documentation, the access controls, and the seven-source pipeline feeding it are.
The actual buying decision
That changes what a marketing team should be evaluating in 2026. The question for an agent vendor is not whether it can run a given workflow. Most can, at least in a demo. The question is whether a team’s own data strategy, who owns it, which systems it touches, how clean the handoffs are between them, is in a state an agent can actually run on. The 82% of marketers expecting ROI from agentic AI are making a bet. The 52% who still do not own their data strategy are the ones who may never collect on it.