AI in banking workflows
Banking teams can use GenAI for defined research and preparation tasks. The information available for a brief and the authority required for a banking decision need to be considered separately.
Does a company brief justify a credit recommendation?
AI in banking can support relationship-manager preparation, internal policy search, document summaries, customer-service drafting, operational support and training simulations. Teams need to define the source information, authorised actions and review requirements for each use case. Public information, internal knowledge and customer data have different access requirements, while regulated decisions depend on the bank’s approval and control processes. A trial should identify the accountable business owner, conditions for escalation and measures of the complete result. Manager practice includes checking source accuracy, missing context and exceptions before deciding whether the use case should continue or expand.
The first exercise separates public information, internal knowledge, customer data and decision authority. Banking teams then identify where AI can assist preparation, search or training within the access and approval requirements of the selected workflow.
Work on a decision you recognise.
A relationship manager requests a pre-meeting brief from public company filings. The AI draft also recommends increasing a credit limit, although no internal credit assessment was supplied. Which part of the output can support the meeting?
Review the brief's scope
The public-source summary can support meeting preparation once its statements are checked. The credit recommendation exceeds the evidence and authority of this task. Remove it from the brief and record the questions requiring the bank's assessment process.
Define the requested output as a company summary and a list of questions for further examination. A research-support workflow should be assessed against that purpose.
- Starting material
- Identified public filings, the meeting purpose and the approved scope of the briefing task.
- The task
- Summarise the business, cite the source and reporting period, and list questions the public material cannot answer.
- Review the result
- Verify company identity, reporting period and cited figures. Check that the draft does not make a decision or commitment outside the authorised task.
What participants work on.
- Leadership alignment on banking AI opportunities and boundaries
- Workflow inventory across customer, branch, risk, and operations teams
- Use-case prioritization by value, data sensitivity, and control
- Manager training on verification and accountable AI use
- Governance routines for privacy, compliance, and escalation
- Pilot plan for controlled banking workflows
Choose a defined starting workflow
Select work such as internal search, meeting preparation or document summaries where the source information and review requirements can be specified.
Name the business owner and agree how the result will be assessed before the trial begins.
Prepare managers to review the trial
Managers need to know which tasks are authorised, how outputs will be checked and who handles uncertain cases. Document these requirements in the workflow.
Review participant practice and the work produced before expanding the use case.
Match access and authority to the task
Examples include summaries of public regulatory material, internal communication drafts, training simulations and searches of approved policy information. Identify what the system may access for each task.
Customer data and consequential decisions may require additional approval and controls. Assess those requirements separately when selecting the use case.
Connect the problem to an action.
| Area | Question to examine | Programme response |
|---|---|---|
| Too many banking use cases | The team needs to prioritise use cases before beginning a trial. | Rank opportunities by repetition, business value, reviewability, data sensitivity, and ownership. |
| Regulated decisions need caution | The proposed use case has unclear decision authority or review requirements. | Keep humans in the review loop: define where AI researches, where it detects exceptions, where humans decide, and how outputs are verified. |
| Managers need practical fluency | Managers cannot govern AI workflows they cannot inspect. | Train managers on use-case design, review habits, escalation, and safe delegation to AI. |
Questions about the programme.
Where can banks start using GenAI safely?
Good first candidates are workflows with clear data boundaries and human review: internal knowledge search, meeting preparation, training simulations, document summaries, customer-service drafting, and operational support.
What makes AI adoption in banking different?
Banking has stronger trust, privacy, compliance, and accountability constraints. Adoption must define data boundaries, verification routines, escalation paths, and human decision ownership before scaling.
Who should attend an AI in banking workshop?
Business leaders, branch or regional managers, operations teams, L&D leaders, risk and compliance stakeholders, and teams responsible for early AI pilots.