Roles, skills and upskilling

The skills gap on Andre's scorecard was not really one gap. The board needed enough understanding to ask good questions about AI proposals. Site managers needed to know how to judge whether a tool was working in their clinic. Reception staff needed to know what the no-show flag meant and what to do with it. Emergency technicians needed to know how to check an AI-drafted call note. And a handful of people, the champions and initiative owners, needed much more. "One course for everybody will bore half of them and lose the other half," Noor said. The third Thursday session was about matching skills to roles.

AI literacy for everyone

The exam guide asks you to select workforce development approaches that build AI literacy across the whole enterprise, and it names examples: proofs of concept, hackathons, training programs and responsible AI training. AI literacy is the basic understanding everyone in an AI-using organization needs: what AI can and cannot do, why it sometimes gives wrong or invented answers, how to use approved tools safely, what data must never be entered into them, and how to raise a concern. It is not technical training. It is closer to the safety briefing every new employee gets.

Literacy has a practical purpose beyond awareness. The most common AI failures in everyday work come from people trusting outputs they should check, entering data they should protect, or using tools nobody approved. A short literacy program aimed at those three behaviours reduces real risk. It also reduces fear, because people tend to fear most what they do not understand.

Role-based learning paths

Beyond literacy, each group needs a path matched to what it will actually do. Andre and Noor sorted Cascade Ridge's 420 staff into five groups and gave each a different mix of learning.

Lorraine pointed out that 1,050 of the 1,707 hours came from the all-staff modules. Could they be shorter? Noor's view was that 2.5 hours across a year was already lean for something that would change how everyone worked, and that the responsible AI hour in particular was not optional: it covered the data rules and the duty to report problems. They agreed to deliver both modules in short pieces, of no more than twenty minutes each, during existing team meetings, so that clinics would not have to close appointment slots.

Learning by doing

Courses build understanding; doing builds confidence. The exam guide's examples of proofs of concept and hackathons are ways of letting people learn by building or trying something real in a safe setting. A proof of concept (POC) is a small, quick test of whether an idea can work at all, usually with sample data and no expectation that it will go into production. A hackathon is a short event, often a day or two, where mixed teams build rough prototypes of ideas. Both are useful mainly for learning and for generating ideas; neither is a substitute for a properly designed pilot.

Cascade Ridge planned a half-day "AI lab" at each regional meeting in 2027. Mixed groups of reception staff, technicians and managers would use an approved generative AI tool on invented examples: drafting an aftercare sheet from bullet points and then finding its mistakes, summarizing a fake call transcript and comparing it with a technician's note, or asking the assistant questions it should refuse. The aim was to give people hands-on experience of what the tools do well and badly, in a setting where mistakes cost nothing. Ideas that came out of the labs would go through the normal intake form.

Responsible AI training

Responsible AI training is named separately in the exam guide because it is easy to leave out. It covers the organization's AI policy, the data rules, the reasons for human review, how bias and errors arise, and what to do when something goes wrong. For most staff it is part of the one-hour module. For initiative owners, champions and members of the steering group it goes deeper: how to read a model's documentation, how to set and check guardrails, how to handle a complaint, and how to decide whether a new use needs a full risk review. Lessons 6 and 7 cover the content they need.

From doing the task to overseeing it

The exam guide asks you to identify where human roles can shift from manual work to oversight of AI and collaboration with it, balancing human strengths such as critical thinking, empathy and creativity against what AI does well. This is where the transformation becomes personal for staff. A good way to think about it is task by task, not job by job. Most jobs contain some tasks AI can take over, some it can help with and some it should not touch.

Two things matter in this kind of redesign. First, oversight is real work and needs time, skill and authority. If technicians are expected to check AI drafts but are given no time to do it, they will approve them without reading, and the human review becomes a formality. The exam guide's interest in human oversight, covered in lesson 6, depends on oversight being designed into the job. Second, freed time has to go somewhere deliberate. If the six minutes saved per call note simply disappear into a busy shift, the organization has gained a little slack but not the value in the business case, which counted reduced overtime. Andre planned to track overtime hours and time spent with patients as measures for the emergency hospital.

Hiring, partnering and growing skills

Some skills are better hired or bought than grown. Cascade Ridge's plan for 2027 combined all three. It would grow AI literacy and oversight skills in its own staff, because those must sit with the people doing the work. It would hire one data analyst, because data foundations and dashboards needed a permanent person rather than a contractor. And it would rely on a partner for the client-support assistant, as decided in the strategy course, with an explicit plan for Kwame to work alongside the partner so that knowledge transferred rather than leaving with the contract.

Key-person risk ran through the whole plan. Noor was a contractor, and Kwame was the only developer. Andre added two protections: every initiative would keep its documentation and evaluation sets in a shared place, and the data analyst's job description would include learning to run the monitoring Noor set up, so that a single departure would slow the program rather than stop it.

Measuring whether learning worked

Training is easy to count and hard to evaluate. Completion rates show that people attended, not that anything changed. Andre chose measures at three levels: completion of the modules by each group; a short scenario check at the end of the literacy module, such as spotting which of three example prompts breaks the data rules; and behaviour in the work, such as the share of AI-drafted call notes that technicians edited before approving, the number of concerns reported, and the number of unapproved tools found in the six-monthly review. A rising number of reported concerns in the first months would be read as a good sign: it would show that people knew how and felt safe to report.

Skills for leaders

Leaders need a different kind of AI skill. They do not need to know how a model is trained, but they do need to ask the right questions of any proposal: what business objective it serves, what the baseline is, what could go wrong and for whom, who owns it after launch, what it costs over its whole life, and what result would stop it. The six-hour leaders' session at Cascade Ridge was built around those questions, using the business cases and risk reviews from the 2027 portfolio as practice material. Board members would review one real proposal in the session and say whether they would approve it, and why.

Site managers also needed the skill of reading a dashboard honestly. A manager who sees no-shows fall at her clinic in the first month of the pilot needs to know whether that is the reminders working, a seasonal pattern or chance, and to know that the comparison with non-pilot clinics is what answers the question. Without that skill, early numbers are either celebrated or dismissed too quickly.

Workforce questions on the exam

Exam scenarios on workforce development often describe an organization that has bought AI tools but sees little use of them, or one where a single expert team does all the AI work and everyone else waits. The best answers usually combine broad literacy with targeted, hands-on learning for the people who will use or oversee the tools, and they treat responsible AI training as part of the package. When a scenario asks how roles should change, look for answers that redesign tasks, keep humans in charge of judgement and empathy, and give oversight real time and authority. Answers that train everyone in model building, or that remove human review to save time, are usually distractors.