AI has moved quickly into the work of US bank marketers, although the most useful new data shows more than a rise in tool use. It reveals how marketers are gaining access, where the technology is being applied and why applications with high expected impact also demand more from the organisation around them.
The pattern is encouraging without being straightforward. More bank marketers are using AI to produce and adapt content, while many are assembling access through individual or team subscriptions rather than an enterprise platform. At the same time, respondents see particularly strong future potential in personalisation and customer-facing applications, where product accuracy, customer context, evidence and oversight matter most.
That leaves bank marketers at an interesting point. The accessible uses are already spreading, while the higher-value work depends on capabilities that cannot be bought with a general-purpose AI subscription alone.
Marketing is an early destination for AI
A 2026 US Census Bureau working paper gives the broadest view of where AI is entering American business. Drawing on nationally representative survey data collected between November 2025 and January 2026, the researchers found that 18% of firms used AI in at least one business function.
Among those adopters, sales and marketing was the leading function at 52%, ahead of strategy and business development at 45% and IT at 41%. Yet most use remained concentrated: 57% of adopters reported AI in no more than three business functions.
The authors also identified a sizeable group of “marketing specialists”. These firms represented 31% of functional AI adopters and showed a high probability of use in sales and marketing with limited use elsewhere. They were typically small, averaged 18 employees and appeared to use generative AI mainly for outreach. By comparison, comprehensive adopters operating across many functions accounted for only 4%.
For bank marketers, the distinction is useful. Marketing can be one of the first functions to make practical use of AI without the organisation having established a broad or deeply integrated capability. That is consistent with the work itself: language, imagery, research and variation are relatively accessible to current models, so experimentation can begin without a major change to systems or process.
Adoption among bank marketers has accelerated
The American Bankers Association’s 2026 survey brings the picture closer to the function. Its 116 respondents reported a sharp rise in the use of AI-powered marketing tools, from 16.9% in 2024 to 29.9% in 2025 and 50.4% in 2026.
The change is visible in specific types of work. For external content, 16% of respondents described their AI use as extensive and a further 47% as limited. Extensive use for internal content, meanwhile, rose from 1.5% in 2024 to 17.4% in 2026.
These figures describe a function moving beyond occasional curiosity. However, they do not imply that most marketing workflows have been redesigned. Limited use remains more common than extensive use for external content and the survey records tools and applications rather than whether work now moves through a connected, repeatable process.
Assistance is more common than organisational change
The Census research helps place the ABA findings in context. Among firms reporting worker use of AI, 65% limited it to three or fewer tasks. Writing, document analysis and information search were the leading generative AI applications, while augmentation was the dominant effect. Among firms reporting an AI-driven change to worker tasks, 66% reported augmentation alone.
Institutional changes were less common. Almost two-thirds of AI-using businesses reported making none of the adjustments measured by the survey. Training and the introduction of new workflows were each reported by roughly 15%, while changes to data collection, management or storage appeared in only about 7% to 8%.
That does not make current use trivial. Assistance with research, drafting and analysis can remove meaningful work. It does suggest that a bank can accumulate a large amount of individual AI activity while the route from brief to approved communication remains largely unchanged.
A marketer might generate a set of paid social variants more quickly, for example, but still search across documents for the current rate, track down substantiation for a claim and reconstruct the conditions attached to an earlier approval. The output arrives faster, yet much of the effort and risk sit in the surrounding work.
Access is often being assembled by teams
The ABA survey also shows how bank marketers are reaching these tools. Microsoft Copilot and ChatGPT were each used by 69% of respondents, while Canva was used by 58%. More revealingly, only 38% of AI-using respondents said their bank provided an enterprise-wide generative AI platform. Individual or team subscriptions accounted for 47%, while 15% had no employer-provided subscription.
This is a practical sign of bottom-up adoption. Teams are finding useful tools and creating ways to use them before a common institutional environment is in place. The immediate benefit is speed, but the arrangement can also fragment knowledge, permissions and working practices. One team may have an approved enterprise account, another a collection of paid seats and another no formal provision at all.
The useful response is a clearer operating picture. Leaders need to know which tools are in use, which information is suitable to enter them, how generated material reaches review and where the reasoning behind a decision is kept. An enterprise licence can support that work, but it does not complete it.
Today’s use cases are not tomorrow’s capabilities
The ABA findings contain a revealing gap between current and expected impact. Respondents rated AI’s present impact on personalisation at 2.44 out of five, but expected that to reach 4.09. Customer-facing applications moved from 2.14 currently to an expected 4.05.
Those are not simply more advanced forms of copy generation. Personalisation requires reliable audience data, a defensible basis for the variation and controls over what changes from one customer or segment to another. Customer-facing applications need accurate product information, appropriate explanations and a clear route for exceptions or uncertainty.
Consequently, the highest-value applications are also more organisational. They depend on data quality, system access, policy, review and monitoring. A team can begin drafting with AI in an afternoon, whereas it cannot safely create meaningful personalisation merely by writing a better prompt.
Marketing ambition is running ahead of data readiness
Cornerstone Advisors’ 2026 What’s Going On in Banking report reinforces this tension. The survey covered 416 executives, 89% of whom worked at financial institutions with between $250 million and $50 billion in assets.
Among bank respondents, 49% said their institution had already deployed or invested in generative AI and another 29% planned to do so during 2026. Marketing was one of the functions attracting attention: 52% reported current or planned use, up from 30% in the previous survey.
A separate 2025 data readiness assessment cited in the report found that sales and marketing was the lowest-scoring function. Two-thirds of the financial institutions analysed scored below 50 out of 100 in that area.
This matters because the applications bank marketers expect to become most valuable are precisely those that rely on dependable data. If product attributes sit in multiple systems, consent and preference data cannot be used consistently or approved evidence is difficult to locate, the model is not the main constraint.
Data readiness can sound like an initiative that belongs elsewhere in the bank. In practice, marketers can make it more concrete by identifying the information a particular workflow needs, who owns it, how often it changes and what evidence must remain attached to the output. That turns an abstract data problem into a defined piece of operating work.
A more useful way to describe adoption
Much of the disagreement about AI maturity comes from using the same word for different things. A bank may have adoption in the sense that employees can access a tool, even though only a small part of the work has changed.
Access is the starting condition: people have an approved route to a capable model and understand the basic boundaries around its use. The ABA figures suggest many marketing teams at banks are still reaching this point through a mixture of enterprise provision and local subscriptions.
Assistance is the next observable layer. AI helps with a task such as research, summarisation, drafting or analysis. This is where much current use sits and it can produce genuine productivity gains, although the person using the tool still carries the context between disconnected steps.
Embedded use changes the workflow itself. The relevant product information, evidence and policy are available at the right point, routine checks happen consistently, human responsibility is explicit and the outcome can be audited and reused. At this level, performance can be assessed through cycle time, rework, consistency and the quality of escalations rather than licence activation or prompt volume.
This distinction echoes a broader pattern explored in AI adoption is rising faster than organisational capability. Access spreads quickly because it is easy to buy. Organisational capability grows more slowly because it requires decisions about information, responsibility and process.
Start with a workflow worth improving
The data does not suggest that every process used by bank marketers needs an AI redesign. It points towards a more selective approach: choose work where the surrounding context can be made reliable and the improvement can be observed.
A promising candidate usually has recurring volume, a visible queue and a reasonably stable definition of acceptable output. Product page updates, campaign adaptation across channels, evidence gathering for recurring claims and the preparation of review-ready variants may fit those conditions, depending on the bank.
The unit of analysis should be larger than the prompt but smaller than the department. Follow the work from request to approval and ask where time is being lost. If the main delay is a missing source, unclear ownership or repeated interpretation of the same requirement, generating the first draft faster will have limited effect.
Instead, the workflow can be improved around the model: structure the brief, make authoritative inputs available, separate settled checks from issues needing judgement, preserve the approval conditions and feed that reasoning into the next comparable piece of work.
This also gives compliance teams a more useful role in adoption. Rather than reviewing a growing volume of disconnected AI output, they can help define which decisions are routine, what evidence is sufficient and where a new judgement must be made. The objective is not to remove human responsibility, but to spend it where it adds the most value.
Five questions for bank marketers
Are we measuring access, activity or embedded use?
A licence shows access and a prompt count shows activity. Embedded use means a repeatable workflow has changed, with reliable inputs, defined human responsibility and an auditable outcome.
Are our access arrangements consistent with how marketers actually work?
The ABA data shows more respondents using individual or team subscriptions than enterprise-wide platforms. Leaders need a clear view of which tools are used, which data enters them and how outputs reach approval.
Is the obstacle model capability or data readiness?
The desired applications in personalisation and customer-facing work depend on accurate product, customer and evidence data. A stronger model cannot compensate for missing context or uncertain ownership.
Which part of the workflow has actually improved?
Faster drafting matters only if the surrounding work also improves. Examine research, evidence assembly, review, revision, approval and reuse rather than measuring the first output alone.
What knowledge survives the approval?
The reasoning behind a qualification, rejected claim or approved exception should become reusable context. Otherwise each new campaign begins by reconstructing decisions the bank has already made.
The numbers point to more deliberate adoption
US bank marketers are not waiting for a complete enterprise transformation; they are already using AI through tools often selected close to the team and can see credible applications beyond content production.
The research also shows why the path becomes harder as the ambition rises. The most valuable applications need better data, clearer context and a more deliberate workflow than the tools used for ad hoc assistance. That is where marketing and compliance leaders can now be most useful: choosing a worthwhile process, defining what good looks like and making sure each completed decision leaves the next one easier to make.