Do We Need AI Literacy Across the Whole Workforce?

This question comes up in almost every planning conversation about AI literacy.

A firm decides it needs to "do something" about AI capability, and the first instinct is to ask which team should receive the training. Usually, the answer defaults to whoever is closest to the technology: IT, data, or an innovation function.

That instinct made sense for earlier waves of technical change. It does not hold up well against how AI-enabled tools are actually being used across financial services today.

Underwriters, claims handlers, compliance monitors, customer service teams and operations staff are all now likely to encounter AI-generated outputs, whether or not their role has ever been labelled "technical".

The real question is not whether AI literacy should be firm-wide. It is how much depth each part of the workforce genuinely needs.

Why This Question Comes Up

AI tools rarely arrive in an organisation through a single, planned rollout.

More often, they arrive piecemeal: a triage tool embedded in underwriting software, a summarisation feature added to a case management system, a chatbot bolted onto a customer service platform. Each of these can be introduced by a vendor update rather than a deliberate internal decision.

The result is that exposure to AI spreads across the organisation faster than anyone formally decides it should. Leaders are then left asking a scoping question after the fact: does everyone need to understand this, or just the people who selected and configured the tool?

This uncertainty is understandable. Historically, technical capability sat with technical teams, and there was little reason for a claims handler or a customer service adviser to understand how the underlying system worked. AI unsettles that assumption because its outputs are increasingly visible to, and relied upon by, people well outside the team that implemented it.

The Traditional Approach — Training the Specialists

For most of the last two decades, technology training in financial services followed a simple logic: train the people who build or configure the system, and let everyone else use it as a black box.

This approach worked reasonably well for earlier systems. A policy administration platform or a trading reconciliation tool behaved predictably. Once configured correctly by a technical team, non-technical staff could operate it safely without understanding its internal logic. Errors were largely a function of user mistakes or system bugs, both of which specialist teams could diagnose.

Under this model, training budgets and time were concentrated on IT, data and change functions. Front-line staff received operational instructions on how to use a system, not an explanation of how it reached its outputs. This was a reasonable, risk-proportionate choice at the time, because the risk genuinely was concentrated in a small number of technical roles.

Why AI Changes the Calculation

AI-enabled tools behave differently from the systems this traditional model was built around, and that difference matters for scoping decisions.

AI outputs are probabilistic rather than fixed. A model may flag a claim as high risk, suggest a premium adjustment, or draft a response to a customer query, and it may do so with a confidence level rather than a hard rule. Understanding that a recommendation is a probability, not a certainty, is now relevant to anyone acting on that recommendation, not just the team that built the model.

This has widened who is genuinely "exposed" to AI risk. A customer service adviser explaining a decision to a policyholder needs enough understanding to do so accurately and honestly. A compliance monitoring team auditing outcomes needs enough understanding to know what a reasonable check actually looks like. Neither of these roles would have needed deep technical training under the old model, but both now sit close enough to AI-influenced decisions that some literacy is required.

This shift is reinforced by policy direction. The UK Government's AI Skills Compact signals an expectation that organisations build AI capability across their workforce, not just within specialist pockets. It treats broad workforce capability as a marker of a responsibly run organisation, which makes a technical-only training programme increasingly difficult to defend, both operationally and reputationally.

None of this means every role needs the same training. It means the population that needs some form of literacy is far larger than the population that needs deep technical skill.

Practical Considerations for Scoping Your Programme

A workable starting point is to think in terms of three broad depths of literacy, matched to how a role relates to AI-enabled processes.

General awareness suits the majority of staff: enough understanding to know that a tool is AI-enabled, what its general limitations are, and when to escalate rather than accept an output unquestioningly. This tier is appropriate for most customer-facing and support roles.

Operational literacy suits staff who use AI-enabled tools directly to inform decisions, such as underwriters working with a triage system or trading support staff relying on an AI-assisted pricing feed. This tier needs a working understanding of how the tool generates its outputs and where its blind spots typically lie.

Oversight literacy suits staff responsible for supervising, auditing or challenging AI-influenced decisions made by others, including compliance monitoring and risk functions. This tier needs the deepest understanding, because these roles must be able to identify when an output should be questioned, not just how to use the tool themselves.

Two mistakes are common when scoping a programme against these tiers. The first is under-scoping customer-facing and support teams, on the assumption that if they don't configure the tool, they don't need to understand it. The second is over-engineering training for low-exposure roles, applying the same depth of content to everyone regardless of how closely their work touches AI-influenced decisions. Both waste effort and leave genuine gaps unaddressed.

A practical next step is to map which functions currently interact with any AI-enabled tool, directly or indirectly, and assign each an initial tier before designing content. This produces a defensible starting scope rather than a guess.

Example

A London-based specialty insurer introduces an AI-assisted underwriting triage tool. Initially, only the underwriting team receives training, on the assumption that they are the only function affected.

Six months later, it becomes clear that claims handlers, compliance monitoring staff and customer service teams are all interacting with outputs from the same tool. Claims handlers review flagged cases, customer service staff field questions from policyholders about how decisions were reached, and compliance staff audit the tool's outcomes for fairness and consistency. None of these teams received any formal training on how the tool works or where its limitations lie.

The firm redesigns its literacy programme into three tiers: general awareness training for all customer-facing and support staff, operational literacy for the underwriters directly using the tool, and oversight-level literacy for the compliance staff responsible for auditing its outputs. This closes a gap that had left customer-facing staff unable to explain decisions accurately and compliance staff under-equipped to challenge them.

FAQs

• Does every single employee need the same AI training?

  No. The depth of training should vary according to how closely a role interacts with AI-enabled tools and decisions. A baseline level of general awareness is appropriate for nearly everyone, but staff who use AI tools directly, or who oversee and audit AI-influenced decisions, need additional depth suited to that responsibility.

• What happens if we only train our technology team?

  Front-line, compliance and customer-facing staff are left without any understanding of tools they are actually relying on or being asked to explain. This creates operational blind spots, such as customer service teams unable to explain decisions accurately, or compliance staff without the grounding needed to challenge questionable outputs.

• How does the AI Skills Compact affect our scoping decisions?

  The AI Skills Compact signals a policy direction towards broad workforce AI capability rather than capability confined to specialist teams. It makes narrow, technical-only training programmes increasingly out of step with expectations, though the detailed regulatory implications are covered in a dedicated article.