AI Can Work Inside a Box. Marketing Mostly Happens Outside It.
AI clearly belongs in modern marketing. That should not be controversial by now. AI can handle a meaningful portion of marketing operations: the structured, repeatable, output-oriented work where speed, scale, and consistency matter more than original interpretation. In campaign operations, content adaptation, personalisation, testing, and framework-driven communications work, the use case is real.
The mistake begins when that reality is stretched into a larger claim than it can support. Because AI is effective in some highly visible parts of marketing, people start talking as if it is absorbing marketing itself. It is not. What it is absorbing is a narrow formalised subset of marketing-adjacent work, the part that can be boxed into a legible input-output system and improved through computation. And the businesses that confuse adopting a tool with redesigning their operating model are the ones most likely to discover the gap between capability and strategy on their own numbers.
That matters. But the box is not the discipline.
The Visible Layer Gets Mistaken for the Whole
One reason this confusion keeps happening is that the most measurable and repeatable layer of any discipline tends to get mistaken for the discipline as a whole. In marketing, that means campaign outputs, copy variants, comms workflows, response structures, and optimisation loops are treated as if they are the centre of the craft rather than one downstream slice of it. They are highly visible, highly measurable, and relatively easy to audit, so they are mistaken for the main event.
They are not the main event. They are the part of the field most amenable to systematisation. And as the existing Esbee article on the cost of efficiency argued in a different context, visible efficiency and genuine value are not the same thing. A function can become impressively productive at the layer that is easiest to measure while quietly degrading the layer that actually creates commercial advantage.
What AI Is Actually Good At
AI is strongest precisely where reality has already been simplified into something computable. The parts of marketing that AI can automate most effectively are those with bounded inputs, known constraints, and a clear output objective: drafting, adaptation, A/B testing, personalisation at scale, structured reporting, and optimisation within established campaign frameworks. That is why marcomms-style work is especially exposed. Much of it depends on producing plausible, coherent, framework-friendly language from already available context. Once machines became good at linguistic output, a portion of that work became compressible almost overnight.
None of that diminishes the value of communications. It just places it properly. Communications is often about operating well within a frame. Marketing, in the deeper sense, is about deciding what frame matters, what tension in the market is live, what symbolic position can be occupied, what latent demand is building, and what the market is not yet saying clearly about itself.
That is a much larger field.
Marketing Is the Larger Open Field
Marketing is not mostly output generation plus a little insight around the edges. It is mostly interpretation, synthesis, and contextual shaping, with output generation as one downstream expression of that wider activity. The real commercial environment is open, unstable, and socially entangled. Meaning shifts. Attention shifts. Competitors react. Customers do not simply reveal stable preferences. They signal, imitate, hesitate, contradict themselves, and respond differently depending on timing, identity, status, mood, and context. That is the field in which real marketing works.
This is the same pattern that appears when businesses build product roadmaps around solutions to problems their customers do not have. The elegant answer to the wrong question is worse than no answer at all, because it consumes the resources and the attention that should have been directed at the question that matters. AI can produce the elegant answer with astonishing speed. It cannot tell you whether the question was right.
AI is strongest where reality has already been simplified into something computable. Marketing lives mostly in the wider field where meaning, attention, and demand are still in motion.
Closed Systems Win on Calculation. Open Systems Require Synthesis.
The useful distinction is not really human versus machine. It is closed systems versus open ones. In closed systems, calculation wins. If the environment is sufficiently concrete, more computation usually does mean better performance. In open systems, the problem changes. The bottleneck is no longer just processing power. It is whether the relevant reality has been made legible in the first place.
That is where heuristic synthesis matters. The computable layer is not the core of the discipline. It is the compressed subset that becomes visible once the harder interpretive work has already happened. Humans do not beat machines by out-calculating them inside clean frames. In those conditions, they usually should not even try. Humans become valuable when the world is too large, too messy, too unstable, or too underdefined to be cleanly rendered into data. They can connect partial cues, weak signals, experiential fragments, and social context into workable inferences before those inferences become neat enough for the model to treat as decisive.
Data Science Is Strong. Markets Are Larger Than Datasets.
Data science should be respected, not caricatured. It is extremely good at extracting structure from what has already entered the dataset in usable form. That is a serious strength. But markets are larger than datasets. Some of the most commercially valuable signals begin as anomalies, vague tensions, behavioural contradictions, symbolic vacancies, or cross-domain cues that do not yet qualify as strong evidence inside the model. They may look like noise until they do not. And the market does not wait for formal proof before rewarding the firm that read the signal early or punishing the one that waited for the dataset to confirm what was already happening.
This is the same principle that runs through Esbee’s work on reading grievance patterns as organisational intelligence: the messiest, most inconvenient signals often carry the most important information, precisely because they sit outside the neat frameworks that the reporting system was designed to capture. The market equivalent is the gap a skilled marketer sees before the dataset proves it. That gap may show up first as a peculiar clustering of attention, an under-served emotional territory, a fatigue with the language of the category, or a mismatch between what customers say and what they actually move towards. None of that arrives with a tidy confidence interval attached. But the gap is there all the same.
The act of marketing is not merely to wait until the market is statistically obvious. It is often to infer significance from the negative space before the evidence is clean enough to reassure everyone else.
When the Frame Is Wrong, AI Scales the Wrong Answer
This is where AI reaches its real limit in marketing. Not because it is weak, and not because marketers are mystical, but because AI works within a frame. If the relevant context sits outside that frame, the system either infers it imperfectly through secondary and tertiary associations or misses it altogether. And if the frame itself is subtly wrong, the problem gets worse rather than better. AI can then optimise and scale the wrong answer with impressive efficiency.
That should sound familiar beyond marketing. It is the same reason businesses often mistake a performance problem for a people problem when it is actually a management problem. A system can become brilliant at solving the wrong problem. In fact, that is often its most dangerous form. The better it gets inside the box, the easier it is to forget that the box may be the issue.
The Expression Layer Is Not the Insight Layer
There is an implication here that deserves more honesty than it usually gets. Generative AI does not flatten all white-collar work equally. It is most disruptive where value depended heavily on producing polished, plausible outputs from already legible inputs. That is why some highly visible knowledge work suddenly looks more exposed than it did two years ago. The machine has become very good at producing confident language inside known frames. Work that derived much of its value from polished expression inside known frames now looks more compressible than it once did.
The difference between marketing and marcomms becomes visible here with particular clarity. Marcomms is largely the production of structured communications from established positions: messaging frameworks, campaign copy, channel-specific adaptation, and stakeholder materials. Marketing is the broader commercial discipline of sensing what is happening in the market, interpreting what it means, and shaping the organisation’s response. One is downstream of the other. AI compresses the downstream work. It does not replace the upstream thinking that gives it direction.
That is not the same thing as saying real marketing has been flattened. If anything, the opposite is true. The more generic linguistic output becomes, the more valuable genuine market interpretation becomes. The more firms can generate polished copy, strategy-shaped prose, and best-practice framing on demand, the more clearly the distinction emerges between expression and diagnosis, between articulate output and actual commercial insight.
None of this is entirely new. Systems have long been powerful when given clean, specific, and constrained environments. Humans have long been useful in synthesising the larger mess of the real world into action. Generative AI extends machine competence into language, which is significant, but it does not erase the broader symbiosis between structured systems and human synthesis. It mostly extends automation into the expression layer once some slice of reality has already been made legible enough to compute.
The Uncomfortable Conclusion
The right response is not to resist AI. It is to use it aggressively where the work is genuinely boxable and to stop pretending the box is the field. Use AI for adaptation, variation, drafting, structured communications, optimisation, and repeatable analysis inside known frames. But do not confuse that with the broader work of marketing, which remains the larger commercial task of sensing, interpreting, and moving the world before it becomes neat enough to model.
The firms that win will not be the ones that assume AI has absorbed most of marketing. They will be the ones that understand that AI is excellent in the narrow parts of the discipline that can be formalised, while marketing itself still lives mostly in the wider open field. AI can generate inside a box. Marketing mostly happens outside it.
The question for boards and leadership teams is whether the decisions being made about marketing capability, headcount, and operating model are informed by this distinction, or whether they are being driven by a capability demonstration and a cost-per-output calculation that mistakes the visible layer for the whole. One of those is strategy. The other is a slide deck with a savings number on it. The difference will show up.
If you are rethinking your operating model in light of AI capability and want an independent view on where the real value sits, talk to us. Esbee’s management consultancy team works with boards and senior leaders on organisational design, and our HR MOT provides the diagnostic baseline that any structural change should start from.