The AI Mirage: 96% Believe, 8% Deliver
Why most "AI-transformed" marketing functions are still run by humans
Stat of the week — 8%: the share of CMOs who run campaigns where multiple AI agents operate autonomously, against the 96% who say AI is already transforming marketing end-to-end.
Most marketing leaders have watched similar AI demos. In these, an agent reads a brief, creates a campaign, sets the budget, chooses channels, and reports back while the marketer observes. The idea is that marketing can mostly run on its own, with people overseeing instead of doing the work themselves.
Survey results show a different reality. This year, 96% of CMOs told BCG that AI is changing their work from start to finish. But only 8% actually run campaigns where several agents work on their own. Another 42% use generative AI mainly to help with specific tasks. In short, expectations are about a year ahead of what’s really happening.
The gap between what people believe and what they actually do is the main story this year. The teams that succeed won’t be the ones who talk the most about AI or spend the most money. Instead, they’ll honestly assess where they are and improve step by step. We’ll look at how you can figure out your group and what to work on next.
The data agrees: belief is ahead of operation
Deloitte’s research shows that about 38% of organizations are testing agentic AI, but only 11% have it in full use. McKinsey found that almost 90% of Fortune 250 CMOs are experimenting, but less than 10% are seeing value across all their workflows. Nearly everyone is trying AI, but true autonomy is still uncommon. AdExchanger summed up this issue at recent industry events, asking if the agentic 'oasis' is real or just an illusion.
This doesn’t mean the technology is overhyped. It just shows that adoption is still in the early stages.
Why the gap exists: data, orchestration, and talent
The gap between 96% and 8% isn’t because of a lack of ambition. Instead, it’s due to a few key structural challenges that need attention.
Three forces are affecting full adoption:
Agents are limited by the data they can access. Many organizations still need weeks to move a single audience from their marketing database to a digital channel. This delay makes real-time agentic work impossible. Demos use clean, connected data, but most real systems still need to fix their basic data issues.
Orchestration needs clear ownership. One agent writing copy is just a tool, but when agents work together, it becomes a new way of operating. This setup needs a supervisor, clear rules, and a way to handle problems. McKinsey calls the goal a hybrid human–agentic workforce, where a marketer oversees agents instead of doing all the work. Many teams have added agents but haven’t assigned a human supervisor.
Teams are still learning new skills. About 80% of CMOs say they are investing in AI training, which is up 10 points from 2025. The investment is real, but teams are still building their expertise. You can’t supervise what you don’t fully understand yet.
Taken together, these are actually good signs. A structural gap can be closed. Data connections can be made, orchestration can be assigned, and teams can build the needed skills over time. This is more about careful planning than anything magical.
From 42% to 8%
Knowing how ready you are for AI is the first step to moving forward. There are four stages between belief and full autonomy:
Stage 1: Assisted (about 42%). Generative AI helps people with specific tasks. This is helpful, but not yet a big change.
Stage 2: Piloting (about 38%). Agentic workflows are tested in a small part of marketing, with close supervision.
Stage 3: Agent-led. Agents handle multi-step workflows from start to finish, with people checking in at certain points.
Stage 4: Autonomous (the 8%). Multiple agents work together on a campaign with very little human involvement.
One global retailer called its campaign engine 'autonomous.' But on closer look, a person was approving almost every step, from budget to creative to audience. This isn’t a failure—it’s a good example of stage two being mistaken for stage four. The real risk was believing they were further along than they actually were.
Early leaders take a different approach. Instead of saying they’re at the top, they choose one workflow, move it up one stage, and show the results before expanding. This steady progress adds up as AI spreads through the organization.
How to get started
Ask yourself these questions to help shape your plan.
Which stage are you really at? Share the four stages with your leadership team and agree on where you stand. Usually, the honest answer is one stage lower than most people think. First step: Agree on your starting point and stick with it.
Which single workflow will you improve? Pick one that is high-volume and low-risk, like audience building, creative variation, or budget pacing. First step: Choose the workflow and decide which stage you’ll move it to by next quarter.
What needs to happen for an agent to work in your system? Usually, it’s about making sure the agent can access the right data and knowing when a person should step in. First step: Connect the data for that workflow and set a clear point where a decision goes back to a person.
Who is responsible for managing the workflow on your team? First step: Choose that person before you add more agents.
Tech Watch. The “oasis or mirage” question ran through this spring’s industry panels, including POSSIBLE 2026, with practitioners divided on whether agentic media buying is shipping or still on slides. The clear bottleneck most often named isn’t the models. It’s the data and identity plumbing the agents depend on.
This year, the marketing teams that do best will be the ones who can clearly say where they are now and take the next real step forward.
Where are you on the 8-to-42 spectrum? Let us know your stage.
Sources
BCG, How CMOs Are Moving Agentic Marketing from Illusion to Reality (June 2026, n=300 CMOs + 50 interviews) — bcg.com
Deloitte, AI agents are scaling faster than their guardrails / State of AI in the Enterprise (2026, n=3,235 leaders) — deloitte.com
McKinsey, Reinventing Marketing Workflows with Agentic AI (Apr 2026, n=35 Fortune 250 CMOs) — mckinsey.com
AdExchanger, Is Agentic Commerce An Oasis Or Mirage? (2026) — adexchanger.com