How Does AI Literacy Become Part of Culture?

Many financial services firms have already run AI awareness sessions or e-learning modules.

Yet a few weeks later, staff behaviour around AI tools often reverts to old habits — exceptions get escalated unnecessarily, AI outputs get ignored, or they get trusted without question.

Regulators and the government's AI Skills Compact increasingly expect organisations to demonstrate sustained AI capability, not simply a completed training record.

That means firms need practical ways to keep AI literacy alive in daily decision-making, particularly in underwriting, trading, operations and customer-facing roles where AI-assisted tools are already part of the job.

This article explains what separates a training event from a genuinely embedded capability.

Why training alone does not create lasting capability

A training session changes what people know at a single point in time.

Culture change requires that knowledge to shape behaviour repeatedly, under normal working pressure, long after the session has ended.

This distinction matters because most AI literacy investment to date has focused on delivery — completion rates, attendance figures, course content — rather than what happens afterwards.

Without reinforcement, even well-designed training decays. Staff forget detail, revert to familiar habits, or apply new knowledge inconsistently depending on who is supervising them that day.

Regulatory expectations, including those associated with the AI Compact, are shifting accordingly. Firms are increasingly expected to show ongoing, demonstrable capability across their workforce, not a single completed module logged in a training system.

Traditional approaches to embedding capability

Organisations have long understood that new skills only stick when they are reinforced through everyday structures. The same mechanisms that have embedded other capabilities — risk awareness, conduct standards, product knowledge — apply equally to AI literacy.

Mentoring and peer support. Pairing less experienced staff with colleagues who model good judgement around AI tools helps translate abstract training content into concrete, situational behaviour.

Communities of practice. Regular forums where staff discuss real examples of AI use — what worked, what didn't, where oversight mattered — keep the subject active in people's minds rather than confined to a one-off course.

Leadership modelling. When senior staff visibly use, question and discuss AI tools in normal working discussions, it signals that scrutiny and literacy are expected behaviours, not optional extras.

Incentives and recognition. Linking good AI-related judgement — appropriate escalation, sound use of AI-assisted analysis — to performance conversations reinforces that this is a valued skill, not a compliance checkbox.

None of these mechanisms are new. What is new is applying them specifically to AI-related judgement, at a moment when AI tools are becoming embedded across underwriting, trading and operational workflows.

Where AI helps

AI-enabled tools can support reinforcement at a scale that manual mentoring alone cannot match.

Contextual nudges can prompt staff to reconsider an AI-assisted recommendation at the point of decision, rather than relying on staff to remember training content from months earlier.

Scenario-based practice tools can generate realistic, varied situations for staff to work through repeatedly, keeping judgement sharp between formal training events.

Some organisations use AI systems to surface patterns in how staff interact with AI outputs — for example, flagging where escalation rates are unusually high or low — giving managers useful signals about where reinforcement is needed.

These tools genuinely extend what is operationally possible. They allow reinforcement to happen continuously and at individual level, rather than only through scheduled sessions.

However, they support embedding — they do not replace it. Human mentoring, peer discussion and leadership behaviour remain the primary drivers of lasting cultural change. An AI-generated reminder cannot substitute for a supervisor talking through a difficult exception with a junior colleague.

Operational considerations

Embedding AI literacy requires sustained attention, and several practical risks need managing.

Leadership sponsorship must be visible and ongoing. If senior staff treat AI literacy as an HR-owned initiative rather than something they personally model, embedding rarely takes hold.

Reinforcement mechanisms need to fit naturally into existing workflows. If nudges, reminders or review sessions feel like an additional compliance burden layered on top of daily work, staff will disengage from them.

Overuse of automated nudges and reminders can create fatigue. Reinforcement that arrives too frequently, or without clear relevance to the task at hand, loses its effect quickly. Human reinforcement — mentoring conversations, peer review, team discussion — remains essential precisely because it can be judged and adapted in the moment, in a way automated prompts cannot.

Finally, embedding is not a project with a completion date. It is an ongoing practice that needs to be revisited as tools, regulations and staff turnover change the operational landscape.

Example

A London-based commodity trading desk introduces an AI-assisted settlement exception tool. Initial training familiarises staff with the tool, but three months later, junior operations staff still escalate every exception to a supervisor rather than applying their trained judgement.

The operations manager introduces weekly peer review sessions where staff discuss recent AI-flagged exceptions together, and senior traders begin referencing AI outputs openly in daily stand-ups, normalising informed scrutiny of the tool's suggestions.

Within two months, junior staff independently resolve a higher proportion of flagged exceptions, escalating only genuine edge cases. The peer review habit becomes a standing weekly practice, embedding AI literacy into the team's routine rather than leaving it as a memory of a training session.

FAQs

• Is a single AI literacy training course enough to meet regulatory expectations?

  Unlikely on its own. Regulators and the government's AI Compact increasingly expect firms to demonstrate ongoing, sustained capability across the workforce, not a single completed training record. A course can build initial awareness, but firms need reinforcement mechanisms to show that literacy is maintained over time.

• How long does it take for AI literacy to become embedded in culture?

  There is no fixed end date. Embedding is a continuous process, typically measured in months and ongoing reinforcement cycles rather than a single milestone. Progress depends heavily on consistent leadership behaviour and peer reinforcement rather than the passage of time alone.

• Can AI tools themselves help embed AI literacy?

  Yes, to a point. AI-enabled nudges, scenario practice and contextual prompts can reinforce learning at scale and keep judgement active between formal sessions. However, human mentoring, peer discussion and visible leadership behaviour remain the primary drivers of lasting cultural change — AI tools support that process rather than replace it.