What AI skills does a team actually need?
eams need judgement more than prompt tricks: when to use AI, clear context, checking output, safe data handling and knowing when to escalate.
Judgement matters more than prompt tricks
Lists of clever prompts date quickly as tools change. What lasts is judgement: the ability to decide whether AI suits a task, steer it properly and catch it when it goes wrong. Build training around that, and the tool-specific tips become easy to pick up later.
The five core skills
Choosing when to use AI and when not to: Recognising tasks where AI saves time, such as a first draft of a policy summary, and tasks where it adds risk, such as calculating a client's final tax position.
Giving clear context: Explaining the audience, purpose, constraints and source material, so the output fits the job rather than sounding generic.
Reviewing output critically: Checking facts, figures, tone and omissions with the care you would give a new starter's work, because the tool can sound confident and still be wrong.
Handling data safely: Knowing what information can go into which tools, and what must never leave approved systems, such as client personal data or unpublished financials.
Knowing when to escalate: Recognising when an output, a request or a near miss needs a second opinion from a manager, IT or a compliance lead.
Your AI acceptable use policy should set the rules for data handling and escalation, so training and policy say the same thing. Data protection obligations vary between organisations, so take advice on your own situation when setting those rules.
Map skills to real tasks by role
The same five skills look different in each role. Start with the tasks people actually do each week, then decide which skills matter most.
Finance: Heavy emphasis on critical review and data handling, for example checking AI-drafted variance commentary against the underlying ledger.
Sales: Emphasis on context and tone, such as tailoring a proposal draft to a specific buyer's priorities without overstating what you can deliver.
Operations: Emphasis on choosing when to use AI, for instance documenting a process step by step while keeping supplier negotiations human.
Client services: Emphasis on review and escalation, because a polished but wrong reply to a complaint does more harm than a slow one.
A short skills map for each role, listing three or four target tasks, gives you a far better training brief than a generic course outline.
Signs the skills are missing
You can usually spot gaps before any formal assessment.
Output pasted without edits: Drafts arrive with generic phrasing, American spelling or details that do not match the client.
Avoidance of anything complex: People only use AI for trivial tasks because they do not trust themselves to steer it on harder ones.
Uncertainty about data: Staff ask whether they are allowed to use a tool, or worse, never ask at all.
Each of these points to a specific skill to work on, rather than a need for more general training.
Different starting points
Your team will not start from the same place. Some people need the basics of safe use, while others need help moving from casual use to dependable use on real work. Plan for that range from the outset, as covered in training a team with mixed AI skill levels.
More questions