Written by: Jakub Pietroszek, Partnership Manager, Digital Colliers
UBS now requires AI skills when hiring junior investment bankers. It is one of the first big banks to make AI literacy a formal condition of the offer. That decision does not just change the candidate profile. It creates a contract the bank must now deliver on. If you hire on AI fluency, your onboarding must teach a specific, auditable curriculum. Most mid-market banks have avoided that work. They have let teams experiment with tools in the shadows. The UBS signal forces three questions into the open.
Which AI tools are actually sanctioned
The typical mid-market bank runs fragmented tool sprawl. Treasury uses one vendor's assistant. Credit risk uses another. Relationship managers installed ChatGPT Plus on personal devices. Compliance has no inventory and no logging. You cannot onboard a junior on AI fluency if you have not decided which tools are approved. The first question is which assistants belong on the sanctioned list. That decision cascades. The EU AI Act sets transparency obligations from August 2026 and high-risk obligations from December 2027. Fines reach €15M or 3% of global turnover for high-risk violations. DORA has been in force since 17 January 2025. Both frameworks demand you know which AI tools touch client data, where they run, and what fallback exists if they fail. If your sanctioned list does not exist, your onboarding cannot teach compliance.
What the supervised-use policy looks like per workflow
Sanctioning a tool is not the same as defining how it can be used. The second question is what the supervised-use policy looks like for each workflow. Can a junior analyst use an AI assistant to draft the first pass of a credit memo, or must a VP review every output before it moves forward. Can relationship managers summarise client call notes with AI, or does that summary count as a record that must be stored under your data retention policy. Most banks have not written those rules. They have let seniors make judgment calls on the fly. That works until you hire on AI fluency and promise to train juniors on it. Without a supervised-use policy per workflow, onboarding becomes a liability. The junior uses the tool the way they think makes sense. Six months later, an audit finds they ran a high-risk process unsupervised. The bank cannot point to a policy that was taught and broken. It can only point to a hiring criterion it never operationalised.
How usage is logged per matter or client
The third question is logging. If a junior uses an AI assistant to draft part of a pitch book, that usage should be logged against the client matter. If they use it to pull market data for a deal memo, that should be logged too. Logging serves two purposes. It gives compliance a trail if the regulator asks what AI touched which client file. It also gives you training data. You can see which workflows juniors use AI for most often, which tools they default to, and where they hit friction. That feedback loop lets you refine the onboarding curriculum. Right now, most mid-market banks log nothing. The tools run on personal devices or through departmental subscriptions with no central visibility. When you hire on AI fluency, that silence becomes a risk. 95% of enterprise GenAI pilots deliver zero measurable P&L impact. 88% of AI proofs-of-concept never reach production. The pattern behind those failure rates is usually the same. The organisation ran experiments without deciding what success looked like, how usage would be governed, or how learning would be captured. Banks that hire on AI fluency without logging are setting up the same failure mode.
Turning sprawl into a curriculum
The bridge from fragmented tool sprawl to a curriculum juniors can be tested against is not complicated. It requires three documents. The first is a sanctioned tools list with a one-paragraph description of what each tool does and what it must not be used for. The second is a supervised-use policy matrix that maps workflows to approval requirements. The third is a logging protocol that specifies which usage goes into which system and who reviews the logs. Once those three documents exist, onboarding can teach against them. Juniors learn which tools they are expected to use, which workflows require supervision, and how their usage will be tracked. The hiring criterion becomes testable. The bank can onboard with confidence because it has defined what AI fluency means in its own environment. The alternative is to hire on a capability you cannot train, cannot supervise, and cannot audit. That is the left-behind risk. UBS made the move. Mid-market banks that follow without doing the policy work will hire talent they cannot onboard and create exposure they cannot measure.