Monday, 7 September 2026
Artificial intelligence in banking is moving beyond experimentation and into a more practical phase focused on operational value, customer intelligence and faster decision-making. At AI4 2026, a KeyBank executive responsible for consumer-bank AI strategy described how the institution is identifying the parts of its business where AI can produce the clearest commercial return, while avoiding the temptation to apply the technology everywhere at once. The presentation, “When the Rubber Meets the Road: How the Consumer Bank is Realizing Value With KeyBank,” focused less on model capabilities and more on where AI is actually changing day-to-day banking.
The central argument was that banks should begin with their core business functions and identify where intelligence, automation and better use of data can directly improve service, reduce friction or strengthen competitiveness. The presentation identified technology and operations as two of the most promising areas for AI deployment because these functions sit behind much of what determines whether a bank can serve customers quickly, understand their needs, process documents efficiently and respond to competitors.
One of the most important use cases is the long-standing ambition to create a complete view of the customer. Banks possess large amounts of transaction, deposit, lending and behavioural data, but assembling that information into a usable picture has historically been difficult. AI does not remove the need for clean and well-organised data, but it can accelerate the process of connecting that information and making it accessible to employees. A bank that understands how money moves through a customer’s accounts can identify changes in behaviour more quickly. If customers begin transferring significant funds away from the institution, for example, the bank can recognise the change sooner and determine whether there is a relationship issue, a competing financial product or another reason behind the movement.
The same intelligence can support what banks increasingly describe as the next best action. Instead of employees approaching customers with generic product offers, AI can potentially help bankers understand the circumstances surrounding each relationship and identify which conversation is most relevant at that moment. The objective is not simply selling more products but improving the timing and relevance of customer interactions. This moves customer analytics towards a more continuous model in which behaviour can be interpreted as it changes rather than relying entirely on broad and relatively static customer segments.
Document management represents another major opportunity because banking remains heavily dependent on paperwork. Regulatory disclosures, trust agreements, mortgage documentation, customer statements, internal policies and procedures create enormous quantities of information that employees must locate, interpret and process. AI can reduce the amount of time spent navigating that material. One example discussed involved trust documentation within wealth management, where multiple agreements can change over time and employees need to determine which document governs the current relationship. AI-powered search can help locate the appropriate material and surface the relevant clauses without requiring employees to work manually through large repositories.
Home lending provides another obvious application. Mortgage processes involve relatively standardised documentation but large volumes of information. AI can extract data, analyse documents and accelerate initial processing before a human makes or approves the final decision. This combination of machine processing and human oversight is particularly suited to banking, where automation can reduce administrative work without removing accountability from regulated decisions.
Internal procedure search is another comparatively straightforward but potentially valuable use case. Bank employees regularly need guidance on how to complete specific transactions or resolve unusual customer requests. Traditional keyword search may return numerous documents that employees still have to read before finding the correct procedure. Natural-language AI can narrow that search and provide the relevant information more directly, reducing the time spent navigating internal systems.
The presentation also highlighted competitive uses of document intelligence. Merchant-services teams, for example, may receive pricing statements from competing providers. AI can analyse those documents more quickly and help a bank generate an alternative proposal, shortening the time between an initial conversation and a competing offer. In this type of use case, the commercial benefit is not abstract productivity but the ability to respond to potential customers faster.
Data engineering is another less visible but potentially important area. Banks depend on large numbers of pipelines connecting customer information, financial systems, risk platforms and reporting tools. AI-assisted development can help engineers build and maintain those pipelines more quickly, improving the speed at which useful information becomes available to business teams.
Data visualisation could change as well. Large organisations have traditionally relied on dedicated business-intelligence platforms that require time to design and maintain dashboards. Generative AI makes it increasingly possible to create smaller, purpose-built web interfaces or visualisations rapidly, potentially allowing business teams to obtain precisely the information they need without commissioning large dashboard projects. This points towards a broader transformation in enterprise software in which employees increasingly create targeted applications around specific problems instead of adapting every task to large standardised systems.
KeyBank’s experience suggests this is already beginning informally. Employees are increasingly able to prototype applications themselves using AI coding tools. The challenge shifts from building the prototype to determining which applications are robust enough, secure enough and important enough to be placed into production across a regulated organisation. Modern AI makes creating software easier, but running production systems inside a bank still requires cybersecurity, data governance, testing, integration and ongoing maintenance. A useful tool developed for a small team is not automatically an enterprise application.
The same principle applies to broader AI deployment. Rather than building every solution centrally, the presentation described a layered approach in which some employees use generally available enterprise AI tools for everyday productivity, existing software vendors provide AI features within systems employees already know, and more complicated business problems are escalated to specialist teams developing custom solutions. This allows AI adoption to spread beyond the relatively small number of people working directly in machine learning or software engineering.
The productivity implications can become meaningful even when individual time savings appear small. The speaker used KeyBank’s workforce as an example, arguing that if every employee saved only one hour per week through AI tools, the cumulative effect across an organisation employing more than 17,000 people would be substantial. The example illustrates why large companies are concentrating increasingly on relatively simple productivity tools alongside more ambitious AI projects. A small improvement repeated thousands of times can have more financial impact than a sophisticated demonstration used by only a handful of employees.
However, the presentation repeatedly returned to the importance of solving actual business problems rather than developing technology for its own sake. The consumer-bank AI team has accumulated a large backlog of potential applications, but not every idea can or should be developed. The more effective approach is to begin with the operational problem and work backwards towards the technology. Banks increasingly need to ask whether an application reduces processing time, improves customer retention, increases conversion, lowers operating costs or reduces risk. If there is no clear answer, the project may have little commercial value regardless of how sophisticated the underlying model appears.
Adoption is closely related to the same issue. Employees are much more likely to use AI when the application solves something they already find frustrating. Forcing technology into a workflow without demonstrating a practical benefit creates resistance, while building around actual user problems makes adoption considerably easier.
The presentation also emphasised that successful AI deployment requires cooperation across an organisation. A specialist AI group cannot independently transform a bank. Technology teams, operations, risk departments, business leaders and senior executives all have to participate because each controls a different part of the process required to put applications into production. Executive alignment is particularly important because AI initiatives that remain isolated within individual teams may produce interesting prototypes but struggle to obtain the funding, permissions and organisational changes required for broader adoption.
External vendors form another part of the ecosystem. Banks increasingly rely on technology providers not only for products but also for information about emerging frameworks and capabilities. The challenge is balancing access to innovation with the need to maintain control over sensitive data, costs and technology architecture.
Governance therefore remains fundamental to the entire strategy. Not every AI system presents the same level of risk. An internal tool helping an employee summarise a document is different from a model influencing lending or interacting directly with customers. Banks consequently need governance structures that reflect the potential consequences of each application rather than treating every AI tool identically.
The handling of customer information is especially sensitive. During questions following the presentation, the KeyBank speaker said customer data used in AI development is tokenised and accessed through controlled internal data environments rather than exposing personally identifiable information directly inside development workflows. This was presented as part of the bank’s internal approach rather than a general industry standard.
This balance between experimentation and control will become increasingly important as AI spreads further through financial institutions. Banks hold some of the most sensitive data in the economy and operate within regulatory systems built around accountability, auditability and customer protection.
The broader lesson from KeyBank’s experience is therefore less about any single AI application than about the emerging operating model surrounding them. AI value is beginning to appear in practical areas such as understanding customer behaviour, finding documents, accelerating mortgage processing, improving internal search, building data pipelines and giving employees better productivity tools. None of these applications individually represents the futuristic autonomous bank frequently imagined in discussions about artificial intelligence.
Together, however, they could significantly change how a consumer bank operates. The most successful institutions may ultimately be those that resist the temptation to pursue AI everywhere and instead identify the relatively small number of problems where better intelligence or automation materially changes the customer experience or the economics of the business. For banking, the point at which AI becomes strategically important may therefore not be when customers begin talking to sophisticated digital financial advisers. It may arrive much earlier, when thousands of small decisions, searches, document reviews and operational tasks across the organisation quietly become faster and more intelligent.
Source: CIJ.World Research & Analysis Team
