AI adoption in marketing is decided by org design, not procurement. McKinsey's latest State of AI survey counts 88% of organizations using AI regularly in at least one function, while its AI high performers, the group seeing significant value with meaningful profit impact, represent about 6% of respondents. My read: the 6% redesigned the work around AI. The rest just installed it. Three redesigns decide which group your marketing organization lands in, and all three sit with the CMO, not the CIO: the work, the people, and the guardrails.
This article expands on a point I made in a recent LinkedIn post on AI and org design. We got all three of these redesigns wrong at least once at Improvado, guardrails most expensively, so what follows is written from scar tissue rather than theory.
Key Takeaways
- The AI adoption gap in marketing is structural: 88% of organizations run AI somewhere, but McKinsey's high performers, the roughly 6% seeing significant value and meaningful profit impact, are the ones that fundamentally redesigned workflows, which McKinsey found is among the strongest contributors to business impact from AI.
- An agent added to an existing workflow produces a demo. A workflow rebuilt around the agent produces pipeline. If your process map still looks like 2024, the agents are decoration.
- The scarce skill is not prompting. It is writing the business down: naming rules, channel logic, and what good looks like, in a form an agent can act on. The World Economic Forum found skill gaps are the top barrier to transformation for 63% of employers.
- Guardrails belong in the harness, not the prompt. Our agents once spent $1,768 in one night on an API key that one polite line of prompt told them not to touch. Removing the key cut that spend by 99.4%.
- None of this needs a bigger budget. It needs the org chart to admit the agents exist: named owners for workflows, documentation, and spend limits.
The 88/6 Gap: Installed AI Is Not Adopted AI
The numbers describe two different realities. In McKinsey's 2025 State of AI survey, 88% of respondents report regular AI use in at least one business function, up from 78% a year earlier. Yet the high performers, respondents attributing 5% or more of EBIT to AI and reporting significant value, make up only about 6%. Nearly everyone has AI. Almost nobody's P&L can tell.
The same survey points at why. McKinsey found high performers are nearly three times as likely as others to have fundamentally redesigned individual workflows, and that intentional workflow redesign is one of the strongest contributors to meaningful business impact from AI. That is an organizational finding, not a technological one. The tools are broadly the same across the 88%. The org design is not.
For marketing teams this lands directly on the CMO's desk. Procurement can buy an AI marketing agent; it cannot rebuild the workflow the agent sits in, reassign who documents channel logic, or set what an agent is allowed to spend. Those are organizational decisions, and they are the three redesigns that separate the 6% from everyone else.
Redesign 1: The Work
An agent added to an existing workflow produces a demo. A workflow rebuilt around the agent produces pipeline.
The distinction sounds subtle and is not. Bolting an agent onto your current reporting process means a human still assembles the brief, still checks the numbers, still routes the output, with an AI step wedged in the middle. The workflow's shape is unchanged, so its economics are unchanged. That is why so many marketing AI projects plateau at the demo stage: the agent performs, everyone nods, and the pipeline number does not move.
Rebuilding around the agent means asking what the workflow would look like if the agent were a permanent team member: which steps disappear entirely, which decisions the agent makes inside defined limits, and where humans move to review and exception handling. If your process map still looks like 2024, the agents are decoration.
A practical test: pick one reporting or campaign workflow and count the handoffs before and after the agent. If the count did not drop, you added a step instead of removing any.
Redesign 2: The People
The scarce skill in marketing AI adoption is not prompting. It is writing the business down: naming conventions, channel logic, budget rules, what good looks like, in a form an agent can act on. The World Economic Forum's Future of Jobs Report found skill gaps are the top barrier to business transformation for 63% of employers, and in marketing the specific gap is exactly this documentation skill, not model expertise.
In our own go-to-market the tell was simpler: every re-brief was a senior marketer explaining the business to software. Again. Each session re-taught the same naming rules, the same channel quirks, the same definition of a qualified lead. The knowledge lived in people's heads, so the agents woke up ignorant every morning and payroll quietly absorbed the cost of re-teaching them.
The fix is to treat business context as infrastructure. Documented taxonomies, metric definitions, and channel rules become assets an agent loads, not folklore a marketer recites. That is the same principle behind giving agents persistent organizational memory: extract the knowledge once, integrate it, and retrieve it in every session instead of re-briefing from zero.
Redesign 3: The Guardrails
Our agents once spent $1,768 in one night, on an API key that one polite line of prompt told them not to touch. An agent hit a blocker, found the key in its environment, and obediently kept working toward the goal we set, billed per token. Removing the key cut that spend by 99.4% overnight.
The durable lesson: you cannot patch governance holes with prompt guardrails. An agent is tireless and literal, and it will pursue the goal you set with whatever resources you leave within reach. Spend limits, permissions, and key scoping belong in the harness, enforced by infrastructure, not requested by prompt. The same logic applies as agents get wired into ad platforms and CDPs: ask your team what the agents' overnight spend limit is, and if nobody knows, it is infinity.
This is the least glamorous redesign and the most expensive one to skip. We cover the operational side in depth in our guides to AI agent governance and AI agent security: capped keys per task, environment hygiene, cost limits at the platform level, and audit trails for every agent action.
What This Means for the CMO
None of this needs a bigger budget. It needs the org chart to admit the agents exist. Concretely, that means three ownership decisions any marketing leadership team can make this quarter:
Name a workflow owner. One person owns rebuilding one high-volume workflow around an agent, with the handoff count as the metric. Demos do not count; removed steps do.
Name a documentation owner. One person owns writing the business down: the taxonomy, the metric definitions, the channel rules an agent needs. This is a senior marketer's job description changing, which is precisely why it is an org-design decision and not a tooling one.
Name a guardrails owner. One person owns agent permissions, key scoping, and spend limits, enforced in the harness. The role sits naturally with marketing ops, and its first deliverable is an answer to the overnight-spend question.
This is the operating shift we describe across our work on AI business transformation for marketing and the emerging AI CMO role. Improvado's platform is built for the resulting structure: an agentic layer that reads governed, unified marketing data with your taxonomy and definitions applied, and acts inside permissions and limits your team controls. We got the three redesigns wrong at least once ourselves before the structure held. If your version looks messy mid-rebuild, ours did too.
Frequently Asked Questions
Why is AI adoption in marketing not producing ROI?
Because most organizations install AI into existing workflows instead of redesigning the workflows around it. McKinsey's research shows the small group of high performers seeing meaningful profit impact are far more likely to have fundamentally redesigned workflows. Without that redesign, the agent adds a step instead of removing any, and the economics of the process stay the same.
Who should own AI adoption in a marketing organization?
The CMO, not the CIO. The three decisions that determine impact, which workflows get rebuilt, who documents the business rules agents act on, and what agents are permitted to spend, are marketing org-design decisions. IT provisions the tools; only marketing leadership can change the work, the roles, and the limits.
What is the most important skill for marketing AI adoption?
Writing the business down. Agents act reliably only on documented naming conventions, channel logic, metric definitions, and quality standards. The World Economic Forum found skill gaps are the top barrier to transformation for 63% of employers, and in marketing the gap is this documentation capability far more often than prompting ability.
How should marketing teams set guardrails for AI agents?
In the infrastructure, not the prompt. Use scoped, capped API keys per task, keep credentials out of agent-reachable environments, enforce spend limits at the platform level, and log every agent action. A prompt instruction is a request; a harness limit is a control. Only the second one holds overnight.
