Key takeaways
- AI fluency is entering selection criteria in traditional banking roles.
- Tool familiarity is weaker evidence than a realistic work-sample assessment.
- Hiring and L&D need shared competency language and scoring rules.
- Diagnostics can tailor onboarding instead of repeating generic AI foundations.
- Regulated roles must assess verification, data boundaries and escalation.
The Banking Interview Now Tests AI Judgment
On 6 September 2026, the Financial Times reported that UBS will require graduates and interns joining Global Banking and Markets in 2027 to show how they use AI to improve outcomes and efficiency. UBS job listings already ask applicants to evidence the experimental and responsible use of AI in this way. This is an emerging market signal, not an industry-wide standard. Still, it changes the operating question for HR and L&D: if AI fluency is tested at the gate, what exactly should the organisation expect on day one, after onboarding and six months into the role?
Tool Familiarity Is Not Competence
“Uses ChatGPT” is not a useful hiring signal. It says little about whether someone can choose an appropriate task, work within data boundaries, test an output, recognise an unsupported claim or stop and escalate. A candidate may know one interface well and still make poor decisions with AI. Another may move confidently between approved tools because they understand the work, the controls and the limits.
That distinction matters in finance. A polished summary, client note or market brief can look credible while containing a false statement, stale data or an unsupported inference. The standard must therefore test a work outcome and the judgment that surrounds it, not prompt-writing theatre.
AI Fluency Shows Up in Observable Work
A practical AI competency framework should describe actions that a manager can see and assess. The core test is simple: can the person improve a real work outcome with an approved AI tool, verify the result and retain accountable human judgment?
- Select a task where AI can create a clear improvement in speed, quality or coverage.
- Use an approved tool and apply the correct data boundary.
- Give the system enough context, constraints and success criteria.
- Check facts, calculations, sources, assumptions and omissions before use.
- Escalate uncertain, sensitive or high-impact cases and record the human decision.
This model also fits the compliance context. Article 4 of the EU AI Act requires providers and deployers to support AI literacy in a way that reflects people’s experience, training and the context of use. It does not prescribe a universal individual pass mark. That makes a role-based rubric more useful than one generic AI course or a single completion metric.
Work Samples Beat Abstract AI Questions
Build the AI fluency assessment around a short, sanitised scenario from the target role. An analyst candidate might need to turn a mock earnings pack into a draft briefing. An operations candidate might triage simulated exceptions. A compliance candidate might review a fictional case summary and identify where AI output needs further evidence.
Score the process as well as the answer. Did the candidate define the task well? Did they avoid restricted information? Did they validate key claims? Did they identify limitations? Could they explain the final human decision? This makes AI skills hiring more consistent across interviewers and reduces the risk of selecting for confidence with a fashionable tool.
Good to know
What should an AI fluency assessment measure?
It should measure task selection, appropriate tool and data use, output verification, awareness of limits and accountable human judgment.
Should every finance role use the same AI assessment?
Use one core rubric, then change the scenarios, risk thresholds and evidence requirements for each role family.
When should AI capability be reassessed?
Reassess after material tool or policy changes, role moves, control incidents and at regular capability review points.
One Rubric Should Survive the Offer Letter
Recruitment should not create a separate definition of competence that disappears in onboarding. Reuse the same rubric in an AI onboarding assessment during the first weeks. The diagnostic shows which behaviours are already reliable and where each joiner needs support. A new markets analyst may need stronger verification practice. A client-facing employee may need tighter data-handling habits. A manager may need better escalation judgment.
This is where banking AI training becomes more targeted. Instead of putting every employee through the same foundation module, route people into short role-specific scenarios, practice loops and manager-led reviews. App-Learning can connect these moments across the lifecycle: pre-hire scenarios, onboarding diagnostics, adaptive learning paths and reassessments using the same evidence model.
Build one AI standard that holds from interview to role readiness.
AssessCapability Must Be Rechecked as Work Changes
AI fluency will not remain stable. Approved tools change. Internal policies change. New workflows create new failure modes. Reassess when a role changes, when a new tool launches, after a control incident or on a defined review cycle. Keep the evidence lightweight but useful: scenario scores, recurring error patterns, escalation quality and manager observations.
The goal is not to turn every employee into an AI specialist. It is to establish a consistent threshold for safe, useful and accountable AI-supported work. Employers that align selection, onboarding and development around that threshold will gain a clearer view of capability than firms that count course completions or ask whether people have tried the latest chatbot.
