Buying an AI capability and becoming capable of using it are different managerial acts.
Gartner's 2026 CMO Spend Survey covered 401 marketing leaders, primarily at companies with more than $1 billion in revenue. The numbers expose a readiness gap:
- Respondents allocated an average of 15.3% of marketing budgets to AI.
- 70% described becoming an AI leader as a critical goal for 2026.
- 30% reported mature or fully developed AI-readiness capabilities.
The survey does not show that large organizations are incompetent, nor does it compare enterprise performance with small firms. It documents a gap between investment and reported readiness. Gartner's interpretation is organizational: data foundations, processes, governance, talent, budget agility, and operating discipline determine whether purchased capability becomes useful.
Scale changes the coordination problem
A large organization carries dependencies that a small team may not: regulated data, security review, customer contracts, multiple business units, established technology, and consequences that compound at scale. Some delay is the cost of taking those obligations seriously.
A smaller team may be able to run a bounded experiment with fewer handoffs. That can create a learning-speed advantage. But speed becomes an advantage only when the team can judge the output, protect customer trust, assign responsibility, and turn the result into a repeatable process.
The useful unit of AI maturity is not the license. It is the organization's ability to learn safely from a consequential use case.
What readiness looks like close to the work
For a small or midsized marketing organization, readiness can be made concrete through three operating practices:
-
01
Bound the experiment. Name the task, the customer or business consequence, the data the system may use, and the condition that would stop the test.
-
02
Assign a decision owner. Specify whether the system drafts, recommends, or acts; what evidence a person must review; and who has authority to approve, challenge, or override the output.
-
03
Close the learning loop. Compare the result with an agreed standard, record what changed, and decide whether the lesson should alter a workflow, policy, prompt, data source, or role.
These practices are modest by design. Readiness is not a permanent state conferred by a large platform or transformation program. It is the repeated ability to connect a new capability to a responsible decision and an observable lesson.
Expertise is changing roles, not disappearing
The Gartner survey shows investment and readiness separating. A different kind of evidence helps explain why operating knowledge matters.
In the NBER study “Generative AI at Work,” access to an AI assistant raised productivity by 14% on average across 5,179 customer-support agents. Gains were larger for novice and lower-skilled workers and limited for the most experienced workers. This was one customer-support setting, not a universal model of marketing work. It nevertheless shows why leaders should ask how AI changes the distribution of expertise, not merely whether average output increases.
The counterargument: speed can amplify fragility
A small team can automate a weak process, expose sensitive information, or become dependent on a tool nobody understands. Fewer approvals can reduce coordination cost; they can also remove useful challenge. The advantage is not the absence of governance. It is the opportunity to make governance proportional, legible, and close to the work.
The implication for leaders: measure AI maturity by the organization's ability to learn safely—how quickly it can run a bounded test, evaluate the output, assign responsibility, and incorporate the lesson—not by licenses purchased or content produced.
Readiness is the ability to turn an experiment into responsible operating knowledge.
A smaller team may be able to complete that loop faster. Whether it does is a question of judgment, ownership, and learning—not company size alone.
Sources: Gartner 2026 CMO Spend Survey (401 respondents, fielded January–March 2026) and Brynjolfsson, Li, and Raymond, “Generative AI at Work,” NBER Working Paper 31161.
