Your Governance Framework Assumes a Workforce You Haven't Measured

AI competence is not a training attendance figure — it is the measurable capability without which every oversight clause in your framework is a promise you cannot keep.

Every AI governance framework rests on a single unstated assumption: that the people named in it can actually do what it asks. Most firms have never tested that assumption. They have counted who attended the training and called it competence.

This series has argued from the start that AI in a regulated business is a leadership and operating-model decision before it is a technology one. The through-line has been consistent: governance frameworks are worthless if the workforce behind them lacks the competence to exercise the oversight the paperwork claims exists. This piece finally makes that argument measurable.

The JFSC's July 2026 guidance contains one clause that quietly underwrites all the others. Among the proportionate measures for material use cases, it requires firms to ensure they have sufficient people with the required skills and capabilities to apply human oversight and deliver accountability. Every other requirement — human review, challenge of outputs, spotting fabrication, identifying anomalies — depends on that clause being true. This piece is about how a firm would know whether it is.

It is worth stating plainly why this is the piece the series has been building toward. Every earlier part — infrastructure, data, training, audit, accuracy, agents — assumed a workforce capable of operating the controls it described. That assumption has never been tested in most firms, and the guidance’s people clause is the first regulatory instruction that makes testing it non-optional for material use cases.

The oversight clause is a competence clause in disguise

Read the guidance's requirements as a list of tasks and a pattern emerges: almost every safeguard is a human capability, not a system feature. A person must challenge model outputs. A person must identify anomalies in AML cases. A person must spot fabricated content in a regulatory report. A person must remain responsible for high-impact or unusual cases. The framework does not run on models — it runs on the judgement of the people watching them.

Human oversight is not a control a firm can install. It is a competence a firm must possess. A framework that requires a named human to validate a Tier 3 output has not created a control until it can show that the named human can actually tell a good output from a plausible bad one. Until then, the clause is aspirational — and an aspirational control is a particularly difficult kind to catch, because it looks exactly like a real one on paper.

Why attendance is the wrong measure

The instinct, when a regulator asks about AI competence, is to produce the training log. This is the measure I trust least, because it records exposure, not capability. A person can attend every session, pass every quiz, and still clear a fabricated figure because they were never actually tested on the one skill that matters: catching the plausible error under time pressure in a live case.

The Jersey AI Playbook names skills development as the most consistently cited priority across every major stakeholder group — Digital Jersey, Jersey Finance, the JFSC, the IoD and the Government of Jersey — and warns that the cost of delayed investment in skills exceeds the cost of the investment itself. A market that agrees skills are the priority and then measures them by attendance has agreed on the goal and adopted the wrong yardstick.

A competence scorecard, tiered to materiality

Competence is not one thing, and it should not be measured as one thing. The capability a firm needs is proportionate to the materiality of the use case — the same risk-based logic the JFSC applies to governance depth applies to the skills required to exercise it. Table 1 is a self-assessment instrument: for each material AI use case, it asks not whether people were trained, but whether they can demonstrably do the oversight the framework assigns them.

Table 1 — AI competence scorecard: for each material use case, score the demonstrated capability, not the training attended.

The scoring rule I recommend mirrors the one this series has used before for foundations: if the capability behind any material use case sits in the left-hand column, the firm is not ready to rely on human oversight for that use case — regardless of what the training log says. The scorecard's value is that it makes an uncomfortable gap visible before an incident does — and gives the firm a defensible, evidence-based answer when the JFSC asks how it assures itself that human oversight is real rather than nominal.

The manager layer is where competence is really tested

When firms do invest in AI competence, the investment almost always lands on two groups: the staff who use the tools and the senior people named in the policy. The layer in between — the managers who supervise the work — is consistently the most under-invested, and it is the layer where oversight actually happens day to day. A manager who cannot recognise when a team member is over-trusting an AI output is a gap no amount of front-line training closes.

Manager competence is a different skill from user competence. It is the ability to spot the pattern across cases: the reviewer whose override rate has quietly fallen to zero, the queue that is clearing suspiciously fast, the KPI that has drifted because the team is optimising for the metric rather than the outcome. The JFSC's requirement that a competent person remains responsible for high-impact or unusual cases is a manager-layer requirement as much as a senior one — and a competence scorecard that measures only the front line and the board will pass a firm whose actual point of oversight failure sits untested in the middle.

Competence as Jersey's differentiator, not its liability

There is a version of this argument that sounds like a warning, and a better version that is an opportunity. Jersey's compact scale means the island can build genuine AI oversight competence faster than larger, more fragmented markets — through shared standards, coordinated professional development, and the kind of cross-firm learning the Playbook frames as the island advantage. The bodies that would carry that, from the island's professional-development institutions to its regulator, are already aligned on skills as the priority.

The Playbook describes responsible AI adoption as a source of ethical differentiation — competitive advantage earned in a jurisdiction built on trust and regulatory excellence. Demonstrable workforce competence is where that differentiation becomes real. A firm that can prove its people can exercise the oversight its framework promises is not just compliant. It is offering the one assurance a client in a regulated jurisdiction actually values: that the humans, not just the models, are up to the job.

The uncomfortable truth is that competence is the cheapest of the AI investments a firm can make and the one most often deferred, because it does not come with a vendor demo or a dashboard. Yet it is the investment every other one depends on. A firm can buy the best model, write the best framework, and appoint the right accountable executive — and still fail its first real test if the person the framework names cannot tell a good output from a plausible bad one. The scorecard exists to find that out on the firm’s own terms, before a regulator or an incident finds it out on theirs.

For each material AI use case in your firm, could you demonstrate — not assert — that the named humans can catch the error the framework trusts them to catch, or does your evidence stop at the attendance list?

Next in the series: the last mile of trust — building explainability you could put in front of a regulator, not just a diagram you put in a policy.

Sources: JFSC, Guidance on the use of AI in Jersey's financial services sector (July 2026) — Principle 1 (governance and accountability, the sufficient-skills clause), and the oversight requirements across Principles 3 and 5; Government of Jersey et al., Jersey AI Playbook (2026) on skills as cross-stakeholder priority and ethical differentiation.

Professional disclaimer: The views expressed are my own and are shared for general informational purposes. This content is intentionally non-confidential and does not include proprietary, client-specific, or employer-sensitive information. It is not legal, regulatory, investment, or professional advice; readers should seek appropriate advice for their circumstances.

Financial services leader (20+ years) | Lean Six Sigma Green Belt | London School of Economics (LSE) AI Leadership Accelerator

A version of this article was first published on LinkedIn as part of the Sequence body of work. This is the canonical version.

The views expressed are the author's own and shared for general information. This content is intentionally non-confidential and does not include proprietary, client-specific or employer-sensitive information. It is not legal, regulatory, investment or professional advice. © 2026 Paul Soares. Sequence is the trading name of Sequence Advisory Limited (in formation) · hello@sequenceadvisory.com