The easiest way to make an AI program appear mature is to count things, since licences, pilots, prompts and use cases all provide reassuringly visible signs of activity. However, they reveal much less about whether an organisation has improved how work gets done.
This distinction is particularly important in marketing because, although an AI tool can produce a campaign concept, customer email or product description in seconds, the organisation must still establish whether its claims are accurate, its evidence is sufficient and the communication is appropriate for its audience.
AI therefore reduces the cost of producing another version while leaving the importance of getting that version right unchanged.
Three pieces of 2026 research examine this problem at different levels. The Office for National Statistics’ analysis of AI use in UK businesses measures the spread and intensity of technologies across the economy, while a US Census Bureau working paper looks inside adopting firms to identify which functions and worker tasks are using AI. BCG’s global survey of 300 CMOs, meanwhile, examines how marketing leaders are changing workflows, investment and operating infrastructure.
Together, the studies describe adoption at business, functional and leadership level while exposing the same underlying tension: organisations can accumulate a great deal of AI activity before the work surrounding it changes very much.
Drafts may arrive sooner, yet evidence remains scattered. More variants can be produced, while review reasoning continues to disappear into inboxes. Teams acquire increasingly capable tools even though briefs, approvals and handovers retain their old shape.
For marketing leaders, this is an uncomfortable but useful observation because the challenge is becoming less about finding a plausible use for AI and more about deciding which parts of the operating model should now be reconsidered.
Adoption is spreading faster than depth
The ONS found that the proportion of UK businesses with at least 10 employees using an AI technology rose from approximately 12% in late 2023 to 35% in June 2026. As we explored in AI adoption is rising faster than organisational capability, the accompanying indicators are more modest: the average number of technologies used rose from approximately 1.4 to 1.6, only 10% of adopters described their use as extensive and 15% said more than half of their employees used AI in their daily work. The technology count is only a proxy for intensity and the latter two figures are early indicators, but the overall pattern is one of participation advancing faster than depth.
The US research adds something different because, rather than counting technologies, it considers where AI is being used inside the organisation.
Using nationally representative data collected between November 2025 and January 2026, the Census Bureau working paper found that 18% of US firms used AI in at least one business function. Among those adopters, 57% used it in no more than three functions.
Sales and marketing was the most commonly reported function, selected by 52% of adopting firms, followed by strategy and business development at 45% and IT at 41%.
The activity was similarly concentrated at task level because, among firms reporting worker use, 65% limited AI to three or fewer tasks. Writing, document analysis and information search were the leading generative AI applications, while augmentation was the dominant reported effect: among firms experiencing an AI-driven effect on tasks, 66% reported augmentation alone.
These details are more instructive than a headline adoption rate because they show how AI tends to enter organisations through familiar, bounded activities. It helps someone write, search, analyse or prepare, yet it can be present in several parts of a business without connecting those parts or changing the route through which work is completed.
Marketing has the mandate before it has the operating model
Marketing’s position in the US data is understandable because much of its work involves language, imagery, research, analysis and variation. General-purpose AI is already capable enough to assist with each of these activities and the barriers to beginning are relatively low.
A marketer does not need to redesign a department to summarise research, explore propositions or draft five versions of an email. Individual productivity can therefore improve before anyone has decided how AI fits into the wider operating model.
BCG’s global survey of 300 CMOs makes that gap unusually clear. Some 96% of respondents said AI was driving significant end-to-end transformation of their function, yet 42% were still using generative AI only to assist with discrete tasks in a handful of workflows.
The contrast between declared transformation and task-level reality formed the starting point for our earlier article, The missing layer in agentic marketing. However, BCG’s research contains another finding that is especially relevant here: roughly half of CMOs said marketing now leads AI investment decisions within the function.
As a result, marketing is no longer simply receiving technology chosen elsewhere. Many CMOs are gaining the authority to decide where AI should be applied, how far it should reach and which supporting capabilities deserve investment.
Marketing is often an entry point rather than an operating model
The 52% finding could be read as evidence that marketing is unusually advanced. A further part of the Census analysis suggests a more interesting interpretation: marketing is frequently where firms begin and where some of them remain.
The authors grouped firms according to their patterns of functional AI use rather than asking them to select a maturity label. Within this model, “marketing specialists” accounted for 31% of firms using AI in at least one business function and approximately 9% of all firms. They showed a high probability of AI use in sales and marketing, but considerably less use elsewhere.
These businesses were relatively small, averaging 18 employees, while the researchers suggest that they were using generative AI primarily for outreach. At the other end of the analysis, comprehensive adopters represented only 4% of firms using AI in a business function.
This helps explain why marketing can appear both prominent and immature in the same dataset. The tools are accessible, the tasks are well suited to generation and the commercial application is easy to see, yet adoption may still consist of a narrow specialism rather than a connected organisational capability.
Regulated teams should pay attention to how adoption spreads
The Census paper also found evidence of both top-down and bottom-up diffusion. Workers sometimes used AI for tasks where the firm did not report formal adoption, while some firms reported adoption without corresponding worker-task use.
This distinction matters because an organisation can approve technology without changing day-to-day work, just as employees can alter their own tasks before the organisation has developed a common approach. An adoption figure can therefore conceal very different operating realities.
Controls on tool access and confidential data remain necessary, although safe adoption also requires organisational practices around the technology. The UK government’s guidance on scaling and de-risking AI tools, for example, emphasises staff engagement, training, risk management and monitoring as part of sustained implementation.
BCG’s finding that marketing increasingly controls its own AI investment makes the sequence especially important. In regulated sectors, decisions about permissible data, evidential standards, human responsibility and monitoring need to inform the choice of use case rather than follow it. Compliance is most useful here as a contributor to the operating model, not as an additional queue created after the technology has been selected.
Useful governance should distinguish low-consequence assistance from work that changes a customer communication, substantiates a claim or influences a regulated decision, then apply oversight accordingly.
Five AI findings for marketing and compliance teams
Sales and marketing leads among US AI adopters
Among US firms already using AI, 52% reported its use in sales and marketing. Although this does not reveal how sophisticated that use is, it places the function close to the centre of business adoption rather than at its edge.
Marketing specialists form a substantial adopter group
The Census researchers classified 31% of functional AI users as marketing specialists, equivalent to approximately 9% of all firms. Their AI use was concentrated in sales and outreach rather than spread widely across the organisation.
Most adoption remains concentrated
Some 57% of US adopters used AI in three or fewer functions, while 65% of firms reporting worker-task use limited it to three or fewer tasks. Consequently, presence across an organisation should not be mistaken for depth within its workflows.
Augmentation remains the dominant model
Among US firms experiencing an AI-driven effect on tasks, 66% reported augmentation alone. Although autonomous systems attract considerable attention, assistance remains a more accurate description of current business use.
Marketing increasingly owns the decision
Roughly half of the CMOs surveyed by BCG said marketing leads its own AI investment decisions, yet 42% remained at the level of assistance in a handful of workflows. The question for marketing leaders is therefore shifting from gaining permission to deciding what deeper adoption should look like.
The useful work is still ahead
The research describes rapid adoption, a strong concentration in marketing and predominantly task-level use. It also shows that a firm can be an adopter while remaining highly specialised or can approve AI at an organisational level without corresponding use in employees’ work.
For marketing leaders, the immediate agenda is to decide which parts of the function should change, which decisions require human responsibility and what evidence will show that adoption has progressed beyond isolated assistance. In regulated sectors, compliance has useful expertise to bring to those choices before the operating model becomes difficult to change.