How Can General AI Literacy Be Delivered?

Most financial services firms have already accepted, in principle, that general AI literacy is now required across the workforce.

Regulatory attention and the government's AI Skills Compact have removed the option of treating this as optional or aspirational.

The harder question is rarely whether to build AI literacy - it is how to actually deliver it in a way that produces real understanding rather than a completed checklist.

Many organisations default to their existing compliance training infrastructure, assuming AI literacy can be delivered the same way as financial crime or conduct risk training.

That assumption usually does not hold, because AI literacy needs differ sharply by role and change quickly as the technology evolves.

This article sets out the practical delivery formats, sequencing and blend of methods that make general AI literacy programmes work in practice.

Why delivery deserves as much thought as content

Firms are under growing pressure to demonstrate workforce-wide AI literacy, not just capability within specialist data or technology teams.

Rachel Reeves' AI "Skills Compact" and related regulatory expectations focus on outcomes: staff who understand what AI systems can and cannot do, where risks arise, and when to escalate rather than defer to a system's output.

That kind of understanding is difficult to produce through a single generic module.

Underwriters, traders, operations staff, customer service teams and risk and compliance functions all interact with AI-driven tools differently, and each needs a different depth of practical exposure to reach genuine literacy rather than superficial awareness.

Delivery design has to account for that variation from the outset, rather than treating AI literacy as a single item to bolt onto existing mandatory training.

How organisations traditionally deliver mandatory training

Most financial services firms already have a well-established model for mandatory workforce training.

This typically includes:

• LMS-based e-learning modules completed individually.
• Annual or biannual refresher training.
• Classroom or virtual instructor-led sessions for higher-risk topics.
• Sign-off and completion tracking for audit purposes.

This model works reasonably well for stable, rules-based topics such as financial crime prevention, where the content changes infrequently and the primary goal is consistent awareness of fixed obligations.

Applied to AI literacy, the same model runs into three specific limitations.

First, AI capabilities and associated risks evolve far faster than most compliance topics, so static annual modules quickly become outdated.

Second, passive e-learning is poorly suited to building the practical, hands-on familiarity that genuine AI literacy requires - reading about how a model can be wrong is not the same as encountering it.

Third, a single generic module cannot reflect the very different exposure levels of, for example, a trader using an AI-assisted pricing tool and a customer service agent using an AI chat assistant.

Where AI genuinely improves delivery

AI-enabled tools can meaningfully improve how literacy training is delivered, without replacing the judgement-based elements that still require human-led sessions.

Practical uses include:

• Adaptive learning pathways that adjust content depth based on a learner's role and demonstrated prior knowledge.
• AI-generated scenario practice, allowing staff to work through realistic examples of AI outputs, including plausible-looking errors.
• Scalable simulation of real AI-related decisions, letting large teams practise interpreting and challenging AI outputs rather than only reading about the theory.
• Automated identification of knowledge gaps, helping L&D teams target refresher content more precisely.

These tools help scale delivery across a large, dispersed workforce and personalise content that would otherwise be generic.

However, topics involving judgement, escalation and regulatory obligation - particularly for risk, compliance and oversight roles - still benefit from human-led sessions where questions can be explored and organisational context applied. AI can prepare and support these sessions; it should not be the only channel through which they happen.

Sequencing, cadence and organisational reality

Delivery works best when it follows a clear sequence rather than attempting to cover everything at once.

A practical order is:

1. A short, mandatory foundational module for the entire workforce, establishing shared baseline concepts and vocabulary.
2. Role-specific practical sessions for teams most exposed to AI-driven tools and decisions, using realistic scenarios relevant to their work.
3. Ongoing reinforcement through periodic updates, refresher content and hands-on practice as tools and risks evolve.

One-off training events, however well designed, do not build lasting literacy. Cadence matters as much as initial rollout, because the technology and its risks will look different in twelve months.

Delivery choices should also reflect where the organisation is investing in AI capability. Teams closest to new AI-assisted tools - such as those handling settlement, pricing or customer decisions - warrant earlier and deeper practical training than teams with less direct exposure.

Example

A London-based commodity trading firm rolls out general AI literacy ahead of introducing an AI-assisted settlement reconciliation tool.

Rather than a single e-learning module, the firm delivers a short foundational course to all staff, followed by hands-on scenario-based sessions for settlement and operations teams using anonymised real trade data, with risk and compliance staff receiving additional sessions on oversight obligations.

Staff directly affected by the new tool gain practical familiarity before go-live, while the wider workforce achieves baseline literacy, reducing resistance and misuse risk once the tool is deployed.

FAQs

• Is e-learning enough to deliver AI literacy?

  E-learning can cover foundational awareness effectively, but it is rarely sufficient on its own. AI literacy depends on practical exposure and role-specific depth that passive modules struggle to provide, so e-learning works best as one part of a blended approach rather than the whole programme.

• How often should AI literacy training be refreshed?

  AI capabilities and associated risks change quickly, so a single onboarding session is not enough. Organisations should plan for periodic refresher content and updates that reflect new tools, use cases and risks as they emerge, rather than treating literacy as a one-time requirement.

• Should delivery be the same for every role?

  No. A shared foundational layer is appropriate for the whole workforce, but delivery depth and focus should vary according to how much a role is exposed to AI-driven tools and decisions. Traders, underwriters, operations and customer service teams typically need different levels of practical training beyond the foundational baseline.