AI Training and Awareness Programme

An AI training and awareness programme should help people make the decisions their role actually requires. An employee may need to recognise an inappropriate upload, a reviewer may need to challenge an unsupported answer, and a buyer may need to identify a missing supplier condition. These are different skills, even when all three people use the same AI product.

For Saudi organisations, the programme should connect learning to approved tools, Arabic and English work, information handling, review authority and reporting routes. Attendance is useful administration, but the stronger evidence is whether a participant can complete a realistic task, recognise uncertainty and take the correct next step.

This guide provides an original programme design for learning teams, governance leads and department managers. It includes role pathways, a worked publishing-company exercise and a method for evaluating learning without inventing productivity gains. The programme supports AI governance setup by making the operating rules usable by the people expected to follow them.

Define the decisions people must learn

Start with a task inventory rather than a course catalogue. Ask each role what it will do with AI, what information it encounters and which mistakes could matter. Identify the decision points: choosing an approved environment, preparing inputs, checking outputs, approving an action, handling an exception or reporting an incident.

Write learning outcomes as observable behaviour. “Understands AI risk” is difficult to assess. “Finds the approved tool entry, identifies an excluded attachment and uses the correct review route” can be observed. The outcome should match the participant's responsibility rather than require specialist legal or technical judgements outside their role.

The UK's AI foundation skills for work benchmark includes technical, non-technical, responsible and ethical skills for simple workplace tools. It is a useful international reference, not a Saudi training mandate. A local programme still needs task-specific outcomes, organisational rules and the relevant working languages.

Build a common foundation and separate role pathways

Suggested training pathways linked to work

Audience	Learning focus	Evidence of application
All participating staff	Approved environments, information boundaries, output limits and reporting	Choose a safe next step in a realistic uncertain case
Contributors and operators	Prepare inputs, use the tool and preserve source meaning	Produce a usable draft with appropriate checks and records
Reviewers and managers	Challenge evidence, reject unsupported claims and respect authority	Correct or escalate a consequential issue
Procurement and supplier owners	Recognise capability changes and unresolved supplier conditions	Route a purchase or renewal to the right decision
Administrators and developers	Implement approved permissions, changes and operating controls	Demonstrate the configured boundary and response procedure
Executive sponsors	Interpret conditions, ownership and residual concerns	Make a scoped decision without treating a dashboard as complete assurance

People may need more than one pathway. A manager who uses AI personally should still complete the relevant contributor exercises. A technical specialist who procures a service needs the purchasing route as well as system skills. Avoid treating seniority as evidence that foundational mistakes cannot occur.

Keep the foundation short enough to leave time for practice. Explain the concepts needed for the decisions, then use the organisation's actual policies and request routes. The employee generative AI policy and acceptable-use matrix should be available during exercises just as they will be during work.

Design exercises around realistic ambiguity

Use synthetic or appropriately prepared materials that resemble the work without exposing unnecessary real information. Include a source document, the AI output and the relevant policy entry. Ask the participant to decide what can proceed, what needs correction and what requires another owner.

Make some outputs mostly correct. An obvious nonsense answer tests vigilance, but a subtle omitted condition tests whether the reviewer understands the task. Include an uncertain input-permission case where the correct response is to ask for help. Good escalation is a skill, not a failure to complete the exercise.

Provide an assessor guide explaining the material issue and acceptable responses. More than one wording may be valid. Score the decision and reasoning against the task, not whether the participant repeats a trainer's preferred phrase. Record unresolved ambiguity in the exercise itself so the programme improves rather than marking learners unfairly.

Worked example: three publishing-team simulations

Consider a fictional publishing organisation training 24 staff: eight contributors, eight reviewers and eight procurement or supplier owners. It uses three separate simulations. The counts and outcomes below are illustrative; they are not measured Open for AI client results or an estimate of training effectiveness.

Contributors receive a public research extract and an AI-generated article paragraph that turns a tentative finding into a definite claim. Reviewers receive a bilingual summary that omits a material exclusion in Arabic. Procurement staff receive a renewal notice that adds an AI rewriting feature with an unresolved data-use condition. Each group must use its own role's procedure.

Illustrative first-attempt observations

Group	Independent correct response	Needs targeted coaching
Eight contributors	Six preserve the source qualification and record the correction	Two need help distinguishing evidence from confident wording
Eight reviewers	Five identify the bilingual exclusion mismatch and route correction	Three need practice checking meaning across versions
Eight supplier owners	Four recognise the renewal change and hold feature activation for review	Four need practice separating renewal from new capability permission

The first exercise therefore records 15 independent demonstrations and nine people needing coaching. It does not establish that the organisation is “AI ready” or that every participant will perform correctly under pressure. It identifies specific learning and workflow needs.

After targeted coaching, the nine participants receive different but equivalent cases. Six demonstrate the required response independently and three remain under supported practice. The programme can report 21 people with the defined demonstration and three with an open learning action. It should not count coached answers as independent success or treat a course result as permission to use information outside the approved scope.

Separate knowledge, skill and operating permission

A knowledge check can establish whether someone recognises a concept or rule. A practical exercise tests whether they can apply it. Operating permission depends on the role, approved use and organisational authority. Keep these records connected but distinct.

For example, a reviewer who completes training on identifying unsupported claims does not automatically gain authority to approve a technical assurance in a client document. The organisation must still assign a qualified decision owner. Similarly, an employee who passes a tool exercise may not be authorised to connect a confidential repository.

Use training results to support a sensible rollout. People who need more practice may use a supervised workflow or a narrower task while they learn. Define the support and reassessment rather than attaching a permanent failure label. The programme should help managers make workable assignments, not create a decorative certificate that obscures actual capability.

Make Arabic and English exercises equivalent in consequence

Develop the Arabic materials for the work context, not as a shortened translation of an English presentation. Preserve the same source facts, uncertainty, responsibilities and decision consequences. Check terminology with people who understand both the language and the subject.

Include mixed-language interfaces and documents where they reflect daily work. A reviewer may need to understand an English product setting while assessing an Arabic customer-facing output. Provide the exact interface label when useful, then explain the action in clear Arabic. Avoid unnecessary transliteration that makes the exercise harder for reasons unrelated to the skill being tested.

Review assessment fairness across language versions. If one version makes the critical clue obvious while another hides it in awkward wording, the results are not comparable. Pilot both versions and revise the materials before using the outcome to inform access or work assignments.

Give managers a specific training responsibility

Managers should understand the approved task boundaries and the time needed for review. Training employees to verify outputs while measuring them only on speed creates conflicting incentives. Ask managers to plan the review capacity and explain how work proceeds when a claim cannot be verified.

Teach managers to distinguish a staff skill gap from a system or policy gap. If participants cannot locate the approved tool list, fix the list's accessibility. If the escalation owner is unknown, assign one. If the output cannot be checked because the source is unavailable, redesign the workflow rather than instructing staff to be more careful.

Use a manager exercise involving a deadline and an unresolved condition. The expected response should demonstrate legitimate prioritisation and escalation, not pressure to bypass the policy. The approval-route guide should make the manager's authority and limits clear.

Build awareness around changes and real questions

Awareness communications should tell people what changed and what action to take. A new approved tool, a prohibited feature or an updated reporting route deserves a short targeted message. Repeating generic warnings about AI every month can make the information easy to ignore.

Use anonymised or synthetic versions of recurring questions to explain decisions. For example, show why a draft can be prepared from a public source while a confidential attachment needs review. Protect the people involved in original incidents and avoid turning awareness into public blame.

Connect communications to change triggers and the shadow AI response. When employees repeatedly seek an unavailable capability, the organisation should assess the need and explain the outcome. Awareness is a two-way source of operating information, not only a broadcast channel.

Maintain a proportionate evidence record

Record the learner, role, programme version, relevant exercise, assessment outcome and follow-up action. Keep the evidence needed to explain the decision without retaining unnecessary prompt content or personal details. Define who may access training results and how long they are kept under the organisation's applicable arrangements.

Distinguish attendance, attempted exercise, coached completion and independent demonstration. A dashboard that combines them into one completion percentage can mislead sponsors. The publishing example makes the distinction visible: 24 participants attended, while the observed skill state changed through assessment and coaching.

Report patterns rather than using a single score as a complete view of competence. If several reviewers miss the same exclusion, the issue may be the exercise wording, policy clarity or a real skill gap. Investigate before changing access or drawing conclusions about individual performance.

Evaluate transfer into work without inventing causation

After training, review a proportionate sample of relevant work through the normal quality process. Look for the behaviours taught: source checks, correct escalation, preserved conditions and appropriate tool use. Compare findings with the operating context, including changes to tools, workload and review procedures.

A reduction in errors after training does not by itself prove training caused the entire change. Conversely, more incident reports may reflect improved recognition and willingness to report. Explain the limitations of the evidence rather than publishing a precise return-on-training claim that the design cannot support.

Use the findings to improve the programme and the workflow. Retire exercises that no longer match the tools, add cases from new approved capabilities and refresh role pathways when responsibilities change. The governance KPI guide can connect these observations to useful programme reporting.

Calibrate assessors before judging participants

Write the expected decision and acceptable alternatives for each exercise. In the publishing example, a contributor might either remove an unsupported claim or flag it for an authorised reviewer; both can be appropriate if the task requires a draft rather than final publication. The marking guide should explain the boundary so assessors do not reward one preferred wording instead of sound judgement.

Ask two assessors to review a small controlled set of example responses independently. Discuss disagreements about what counts as independent performance, a useful clarification or material prompting. Refine the rubric before applying it to participants. This is a calibration exercise, not a statistical validation claim about the assessment method.

Record critical errors separately from ordinary improvements. Missing a consequential disclosure or treating an unsupported claim as verified may require a stop and coaching. A less clear but accurate sentence may need editing without indicating a failed governance decision. Define those distinctions for the actual role and task rather than imposing one universal pass percentage.

Keep the exercise achievable with the resources employees will have at work. If a reviewer is expected to consult an approved source or ask a specialist, permit that action in the exercise. Independent performance means making the decision without the assessor supplying the answer; it does not necessarily mean memorising every rule without reference material.

Turn an unsuccessful exercise into a specific learning task

For each participant needing coaching, identify the decision that failed and the cue they missed. A contributor may have trusted a plausible citation; a reviewer may have overlooked a stronger Arabic commitment; a supplier owner may have accepted an answer outside the purchased feature. Those errors need different feedback and a new case that tests the same judgement without repeating the answer.

In the illustrative results, the nine people needing coaching receive targeted support. The later equivalent exercises show six of those nine performing independently, leaving three open. Preserve that history: twenty-one demonstrated the required decisions by that point, while three still needed further work. Do not describe the programme as complete for all twenty-four merely because everyone attended.

Agree what the unresolved status means at work. It might require supervised practice, a narrower assignment or another assessment before a person receives a particular review role. The line manager should decide the operating arrangement through the established process. Training staff should not silently grant or remove business authority by changing a learning-system status.

Check transfer with a small, fair work review

After the programme, choose a defined sample of appropriate work and ask whether the taught decision appears in practice. For example, can a reviewer identify the source of a published claim and show how an unresolved point was handled? Use material the reviewer is authorised to inspect and avoid collecting unrelated employee communications.

Compare the work with the expected process, not with an idealised answer that employees were never taught. Record contextual barriers such as an unavailable approved tool, a source that cannot be found or a manager demanding an unreviewed release. Those findings may require process or management changes rather than another training module.

Do not infer that training caused every improvement or decline. Work mix, staffing, tool changes and supervision can affect outcomes. Report observed practice, the scope of the review and plausible explanations that need further investigation. This gives leaders a useful learning signal without inventing an effectiveness statistic.

Maintain the programme as a versioned operating resource

Give each exercise an owner, target role, source policy version and review trigger. When an approved workflow changes, identify which exercises and job aids need revision. Retire obsolete examples so staff are not assessed against an earlier rule while being expected to follow a new one.

Provide a short route for participants to challenge an answer or report an ambiguous instruction. A well-founded challenge may reveal that the policy, interface and training disagree. Resolve the conflict with the relevant process owner and explain the updated interpretation. Treat the question as evidence about the operating system, not automatically as resistance to learning.

For commissioned training, require delivery of editable exercises, assessor guidance, Arabic and English materials, correction history and the agreed rights to maintain them. Name the internal owner who can adapt the programme after the supplier leaves. The lasting deliverable is a repeatable way to build and verify judgement as the organisation's AI uses change.

Commission training as an operating deliverable

A training brief should specify audiences, decisions, languages, materials, practical assessments, assessor guidance and post-training support. Ask a provider to demonstrate how the programme uses the organisation's actual policies and escalation routes. A generic prompt-writing session may be useful, but it is not the full governance programme.

Require handover of maintainable materials and clear rights to update them where agreed. Assign an internal owner for version changes and new-starter delivery. Plan a review when tools, policy or roles materially change rather than relying only on a fixed annual refresher.

Bring role descriptions, approved-use rules and recent questions to a governance setup discussion about AI training. A well-designed programme leaves the organisation with observable skills, practical support routes and a clear record of what people are ready to do within their authorised roles.