AI training, courses & certification

AI Competency Matrix: How to Define What "AI-Ready" Means for Every Role in Your Organisation

· By AIHQ Team

Why "AI-ready" breaks down the moment you name a role

Ask five department heads in the same company what "AI-ready" means and you will usually get five answers: one describes prompt tips, one describes a licence rollout, one describes a policy document, one describes dashboards, and one says the team is "already using ChatGPT, so we're fine." HR then has to defend a training budget against none of those definitions.

The gap is structural. Most organisations have a job architecture, a skills dictionary for technical roles, and a leadership capability framework. Few have any written statement of what good AI behaviour looks like for a finance analyst versus a legal counsel versus a customer service team lead. Without that statement, AI readiness becomes a feeling, and training gets purchased as a block instead of as a response to a diagnosed gap.

This guide gives HR and L&D teams the artefact that fixes that: a role-level AI competency matrix with four proficiency levels, observable behaviours for six job families, a scoring rubric you can run as a skills audit, and a method for converting the resulting gaps into a sequenced course roadmap. It works whether you are 200 people or 9,000, and it is deliberately conservative about what counts as evidence.

The four levels, and why the labels matter less than the behaviours

Most AI competency frameworks fail because the level names are aspirational nouns — "AI champion", "AI evangelist", "AI ninja". Nothing about those labels tells a line manager what to look for on a Tuesday afternoon. Use four levels with plain names and anchor every level to something a manager can observe in a normal week of work.

Level 1 — Aware. The person understands what large language models do and do not do well, knows which tools are approved, understands the organisation's AI usage rules and the data classification boundaries under Malaysia's PDPA, and can explain why some inputs must never enter a public tool. They may not use AI daily.

Level 2 — Applying. The person uses approved AI tools on recurring tasks in their own role, reviews output before it leaves their desk, and can describe two or three workflows where AI saves them time. Output quality is checked, not trusted.

Level 3 — Building. The person designs repeatable workflows and reusable prompt structures for their team, can evaluate whether a tool is the right answer or whether the workflow needs automation, and coaches colleagues. They can also articulate limitations clearly enough to escalate.

Level 4 — Governing. The person sets standards: usage policy, approval pathways, risk assessment for new use cases, review of third-party tool terms, and decisions on when a workflow needs a custom AI workflow rather than an off-the-shelf assistant.

Two design rules matter more than the wording. First, Level 4 is not automatically the most senior grade — a compliance officer may sit at Level 4 on governance while being Level 2 on day-to-day use. Second, levels are role-relative, not a corporate ladder. A customer service agent can be Level 3 in their own domain while a director sits at Level 2.

The step-by-step build, in the order that avoids rework

Step 1: Start from your job architecture, not a course catalogue

Pull your existing role families and job titles. Do not start by listing courses — that is how matrices become vendor brochures. Group roles into six to ten families: finance, HR, operations, customer service, IT, legal and compliance, marketing and sales, procurement, admin and support, and leadership.

Step 2: Write the outcome, not the syllabus

For each family, state in one sentence what AI should change about the work. For finance it might be "faster reconciliation commentary and variance analysis, with every figure traced back to source." For customer service it might be "faster first drafts of responses with correct escalation on regulated queries." Write these yourself with department heads in the room; do not delegate this to a trainer.

Step 3: Assign a target level per family and per seniority band

Not everyone needs Level 3. A defensible default: all staff reach Level 1; most role families target Level 2; selected specialists and team leads target Level 3; risk, compliance, legal and a small IT group target Level 4 in governance. Then note the exception roles explicitly — someone handling personal data under PDPA obligations, for example, may need a stronger Level 1 data-handling standard than the general population.

Step 4: Write three observable behaviours per cell

This is the work. Each cell in the matrix needs behaviours a manager can verify without a test. "Good at prompting" is not observable. "When asked, can show three saved prompt templates they reuse for monthly reporting, and explains why each includes a source-reference instruction" is observable.

Step 5: Attach evidence sources

Decide in advance what counts as evidence: a manager observation form, a submitted work sample, a short practical assessment, or a workflow log. Evidence should be lightweight and role-relevant. Avoid exams that test definitions; they measure recall, not behaviour.

Step 6: Run the audit and hold the scoring steady

Score each person once against their target level, not against the highest level. Use a consistent rubric across departments — see below — and have one central calibration session so that two managers scoring similar work give similar scores.

Step 7: Convert gaps into a sequenced roadmap

Do not commission a single organisation-wide programme. Group gaps by family and level, then sequence: Level 1 foundation first for everyone, Level 2 applied workshops per family, Level 3 for identified leads and champions, Level 4 governance for risk, legal and IT. Sequence by dependency, not by enthusiasm.

Observable behaviours by job family

Use these as drafting prompts, then adapt to your own systems and role titles.

Finance. Level 2: produces a first-draft variance commentary that cites the source ledger lines; never pastes client-identifiable figures into unapproved tools; flags any AI-generated number it cannot trace. Level 3: builds a reusable monthly reporting pack workflow and validates it against one closed month before rolling it out.

HR. Level 2: drafts role descriptions, interview question sets and internal comms drafts with AI, then applies a bias and accuracy check before circulation. Level 3: designs a screening-support workflow with a human decision at every shortlisting gate, and documents where candidate personal data is and is not permitted to go.

Operations. Level 2: uses AI to summarise incident logs and draft shift handovers, keeping the human sign-off step intact. Level 3: maps a repetitive administrative process end-to-end and identifies which steps are automatable, which need AI implementation support, and which must stay manual for control reasons.

Customer service. Level 2: drafts replies faster while correctly escalating regulated, contractual or complaint-handling queries to a human owner. Level 3: maintains the approved response library and tests customer enquiry chatbot style escalation behaviour before any change goes live.

IT. Level 2: uses coding assistants for scaffolding, refactoring suggestions and test drafts while retaining review responsibility for every merged line. Level 3: evaluates tool and agent options against security, data residency and exit terms — the same discipline described in this ai tool pricing comparison and in the questions covered in internal chatbot governance.

Legal and compliance. Level 3: reviews AI vendor terms, assesses whether a proposed use case touches personal data or regulated advice, and writes the practical boundary into the usage policy rather than leaving it abstract. Level 4: owns the approval pathway for new tools and agents.

Leadership. Level 2: can explain the organisation's AI priorities and one measured outcome per function. Level 4: sets decision rights, funds capability before tooling, and requires a review gate before scale-up.

A scoring rubric that survives a calibration session

Score each behaviour from 0 to 3, then average within the cell.

  • 0 — Not present. No evidence of the behaviour, or the person has not been observed attempting it.
  • 1 — Emerging. Attempts the behaviour with prompting; output usually needs substantial correction.
  • 2 — Consistent. Performs the behaviour unaided in normal work; output needs routine review only.
  • 3 — Exemplary. Performs it reliably and improves the practice for others; can explain the reasoning behind the standard.

Convert to readiness bands: below 1.0 is a foundation gap, 1.0–1.9 applied gap, 2.0–2.4 at target, above 2.4 ready to coach others. Two calibration habits prevent trouble: score only against the role's target level, and require one work sample rather than self-report. Self-assessment is useful for sentiment, not for audit evidence.

A note on scope. This deliberate, written approach to levels and evidence is where a matrix differs from a prompt technique chart. If your organisation is already beyond definitions and needs to see what specific prompting proficiency looks like within a family, the prompt engineering skills matrix goes deeper at that layer.

Turning the matrix into a training plan people actually attend

Once scores are in, four moves do most of the work.

  1. Fix the foundation first. If more than 30% of a family scores below 1.0 in data handling and tool literacy, no advanced workshop will stick. Run a short foundation programme, then re-score a sample after 60 days.
  2. Build family-specific applied sessions. Level 2 gains come from practising the actual workflow with the actual documents, not from generic prompting tips. This is where most budget is wasted on generic content.
  3. Create a Level 3 cohort deliberately. Identify 10–20 leads, give them a design brief, and make their output the reusable templates and coaching material for everyone else.
  4. Sequence governance before scale. Level 4 work — policy, approval pathways, review gates — should land before usage spreads, not after. A governance session is cheaper than a cleanup.

On funding: AIHQ is a registered HRD Corp training provider, and programmes can be structured to be HRDC claimable, subject to client eligibility, grant approval and HRD Corp submission requirements. Budget planning for the capability layer should still account for the wider year-one cost picture covered in the enterprise ai roadmap cost model, and claim submissions are worth preparing against the common failure modes in this analysis of hrdc claim rejection.

Making the matrix stick after the first audit

Three maintenance rules keep the artefact alive. Review the target levels every six months, because tool capability moves and last year's Level 3 behaviour may now be baseline. Keep the behaviour statements short enough that a busy line manager will read them. And report at the family level to leadership, not as an individual ranking — matrices used as performance scores invite gaming and quietly stop being useful.

One measurable reference point: when AIHQ supported Media Prima through a 12-month capability journey spanning awareness, fundamentals, intermediate LLM skill-building and advanced application workshops, programme feedback showed 98% of participants satisfied, 90% reporting increased practical knowledge and skills, and 92% finding the training relevant and applicable to their work. Those figures describe that programme, not a general guarantee — but they illustrate what happens when capability is sequenced over time rather than delivered as a single event.

Start with one family, one target level and three observable behaviours. A rough matrix in use beats an elaborate framework still in draft.

You can also read the AIHQ profile for how leadership alignment, role-based training and implementation support fit together across sectors.

← Back to all articles