AI Talent and Capability Building

An AI capability programme succeeds when people can perform the work their service requires. Course attendance may support that goal, but it does not show whether a supervisor can recognise an unsafe recommendation, a knowledge owner can maintain an approved source, or a manager can make a sensible decision from evaluation evidence.

This guide is for Saudi HR and learning leaders, business sponsors and technology managers building a practical capability plan. Begin with the roles needed by the next services you intend to operate. Then decide which capabilities to develop, recruit or obtain through a partner, and how people will demonstrate competence.

Translate the service plan into human tasks

Take an actual workflow and identify what people must do before, during and after AI assistance. Include approving information access, reviewing uncertain outputs, maintaining sources, investigating incidents and deciding whether the service still creates value. These tasks are easy to overlook when a skills plan starts with a catalogue of model-development courses.

SDAIA's National Occupational Standards for Data and AI organise roles around tasks, skills, knowledge and abilities. Use that task-based perspective when drafting role requirements. The role matrix below is an original planning aid; it does not claim to reproduce or certify an official occupational standard.

Connect the tasks to your AI operating model. If nobody owns source quality or release acceptance, training alone cannot repair the missing responsibility. Assign the work, estimate its demand and confirm that the person has time to perform it.

Create a role-specific capability matrix

Example capability matrix for a business AI service

Role	Capability to develop	Demonstration of competence
Executive sponsor	Interpret value, uncertainty and delegated risk decisions	Explain a proceed, narrow or stop decision from an evidence brief
Process owner	Define the changed workflow and acceptance conditions	Map an exception from detection to resolution without an ownership gap
Knowledge owner	Maintain approved sources and resolve conflicting versions	Correct an outdated source and verify the affected answers
User or reviewer	Check outputs and recognise when to escalate	Handle a realistic mixed set of sound, uncertain and incorrect outputs
Technical operator	Investigate failures and restore the agreed service boundary	Complete a supervised incident exercise and explain the evidence
Risk or procurement specialist	Challenge claims and identify missing evidence	Resolve a supplier statement into a scoped question and decision condition

A person does not need the same depth in every row. Executives need to understand consequences and evidence, while operators need deeper diagnostic practice. Give everyone a shared understanding of permitted information, responsibility and escalation, then specialise around their actual work.

Record the current evidence of capability and the required level. A simple progression is observing, performing with support, performing independently within an approved scope, and coaching others. These are proposed internal levels. Define the observable behaviour behind each, rather than awarding a level from self-confidence alone.

Assess the gap with a work sample

Use a short exercise based on authorised representative material. For a knowledge owner, provide two conflicting policy versions and an answer that cites the older one. Ask them to identify the owner, resolve the approved source, request the necessary change and verify the result. Their response reveals whether the gap is knowledge, authority, tools or time.

For reviewers, include outputs that are correct, subtly wrong and legitimately uncertain. Observe whether they check the important evidence and use escalation appropriately. Do not reward rejection of everything; a useful reviewer distinguishes acceptable work from cases needing intervention.

Assess Arabic and English tasks where both are part of the job. A translated slide deck does not demonstrate competence with bilingual documents, local terminology or mixed-language customer requests. Use the language, document structure and decision context that employees will actually encounter, with clear reference answers for assessors.

Choose build, recruit or partner by the duration of the need

Build capability internally where the task is recurring, closely tied to business knowledge or essential to accountable ownership. Recruit when the need is sustained and the organisation can provide meaningful work and professional support. Use a partner for a bounded specialist need or to accelerate learning, with explicit transfer and handover requirements.

The consulting versus in-house comparison develops the sourcing decision. Avoid treating all expertise as permanent headcount or all external help as a substitute for internal ownership. The organisation still needs people who can judge the supplier's work and operate the resulting service.

For a shared AI centre of excellence, choose capabilities that several services need and can reuse. Do not fill a central team with specialists whose work has no funded demand while local knowledge maintenance remains unowned.

Design learning as practice with feedback

A useful programme combines short explanations, realistic exercises, supervised work and an observed assessment. Schedule practice close to the service's launch so the learning has an immediate application. Give learners access to an approved environment and examples that can be used without exposing confidential information unnecessarily.

For an illustrative eight-week plan, the first two weeks establish role expectations and baseline work samples. The next two focus on guided exercises. Weeks five and six use supervised service tasks. The final two review evidence, reassess gaps and agree the limits of independent work. Adapt the duration to prior skills, consequence and available practice; eight weeks is an example, not a promise.

Let failed exercises produce a specific improvement plan. A reviewer who misses unsupported monetary claims needs targeted practice on evidence checking, not another general introduction to AI. A manager who cannot interpret a test result may need support with measurement and uncertainty before approving expansion.

Calculate capacity as well as competence

Consider a fictional service handling 600 cases per week. If 20% need human review and each takes four minutes, the expected review work is 600 × 20% × 4 = 480 minutes, or eight hours per week. These are illustrative planning assumptions. They exclude source maintenance, meetings, incidents and variation in complex cases.

If the review share rises to 30% and average review time to five minutes, the same volume requires 900 minutes, or fifteen hours. This sensitivity matters when assigning a role to someone who already has a full workload. A training programme does not create the additional hours.

Measure actual demand during the proof of concept or controlled pilot where appropriate, and update the staffing assumption before production handover. Include cover for leave and peaks. The service should not depend on one newly trained person being permanently available.

Make knowledge transfer an acceptance condition

Ask a delivery partner to define what the internal team must be able to do at handover. Useful outputs include an operating guide, representative failure examples, a source-update exercise and a supervised recovery rehearsal. Have internal staff perform the task while the partner observes, rather than accepting a presentation as proof of transfer.

Keep reusable learning materials with the organisation and identify who maintains them after changes. If a supplier performs a specialist activity that will remain external, state that clearly and maintain enough internal understanding to commission and evaluate it.

Coordinate role changes and employee concerns through AI change management. Capability building addresses whether people can perform the work; adoption also depends on whether the work is clear, supported and sensibly rewarded.

Review capability as the service evolves

Use a small evidence record per critical role: authorised scope, completed demonstrations, remaining support needs and reassessment trigger. New action authority, materially different information or a changed operating process may require new practice. Avoid permanent competence badges that outlive the task they originally demonstrated.

Open for AI's agentic AI consulting can help review readiness, map responsibilities and sequence the next steps. Bring a role list, one real workflow and current training material to discuss a task-based AI capability plan. Agree any training or transfer deliverables explicitly within the engagement.