AI skill map · Role-based AI competency assessment
AI Skill Map
What Is an AI Skill Map?
- Each role needs different AI skills
- Sort tasks into automated, AI-assisted, or human-judgment
- Assess with real work tasks, and keep results out of performance reviews
What a skill map really changes is the unit of training planning. Before, the benchmark was “how many AI training hours for how many people this year.” Once a skill map is in place, budgets are built on numbers tied to people and levels, such as “12 people short of the Level 3 target in Sales, 8 short of Level 2 in Production Management.” The very sentences written in the approval request change. When executives ask, “We did this last year, why again?”, laying out a gap table of required versus current level by job function shortens the debate.
A skill map is usually built on two axes. Job roles run down the rows, AI competency areas (prompt design, output verification, data preparation, tool integration, security judgment) run across the columns, and each cell records the required level on a scale of 1 to 4. Drawing a common baseline first makes the map far easier to run. For example, make Level 1 in security and verification mandatory for everyone, and spread only the role-specific skills out to Levels 3–4. For assessment, cover everyone with a 25–30 item self-assessment, then re-verify only about 20% of key people in each role with real deliverable tasks. That balances cost and accuracy. It is easy to confuse with a training needs analysis (TNA), but where TNA asks "what do people want to learn?", a skill map looks only at the gap between the level the job requires and the current level.
The point where this most often breaks down in practice is the moment assessment results flow into HR data. Once even a single precedent of results affecting evaluations exists, the next year’s self-assessment average inflates across the board, and an organization where no one can write a single line of prompt gets tallied on paper as “mostly Level 3.” State the no-linkage principle on the first screen of the assessment and report results only as distributions by job function and level, never by individual. The other pitfall is making the skill map overly elaborate. If a 100-cell table takes six months to finish, tools and features change in the meantime and half of it becomes useless. Run a one-page draft with a single cohort first and revise it every 6–12 months, which is more realistic.
John KimInstructor John Kim, author of “B2B Marketing Design” and CEO of Nanum Consulting
Junyoung NohConsumer and generational trend analyst & AI content marketing expertThis is competency modeling and job analysis applied to AI skills. As governments and international bodies published AI literacy frameworks, organization-level skill map design spread along with them.
Consider a hypothetical case. Company A, an auto parts manufacturer with 480 employees, ran a 4-hour introductory generative AI course for all staff last year. Satisfaction came in at 4.1 out of 5, but a usage survey three months later found only 18% were actually using it, and line managers were saying, "We just heard things that have nothing to do with our work." Before spending this year's KRW 60 million (≈ US$44,900) budget the same way, the training team decided to build a role-based skill map first.
- The training team lead picked four functions as the first target, in order of headcount: sales, quality, production management, and business support. For each function, they brought in two senior managers (section chief level), forming a working group of eight.
- Each member wrote down a week of work hour by hour, and the group broke it into 14–16 tasks per job function. A quality team section manager wrote, "I spend 6 hours a week just drafting replies to customer complaints."
- After marking which tasks AI can assist, sales came out at 9 of 15 and production management at 4 of 14, and the training team lead concluded that the two groups should not get the same hours.
- Before running the 28-question self-assessment, the HR manager put "this will not be reflected in evaluations" on the first line of the announcement email. The result: an 86% response rate and an average level of 1.7.
- Based on the results, they designed 2 hours of shared training for all employees (security and verification) plus 4 role-specific hands-on tracks, and set hours by role: 8 hours of proposal drafting for sales and 3 hours of manual search for production management.
- In the follow-up survey six months later, actual usage rose from 18% to 61%, and the average time to draft a sales proposal fell from 4 hours to 1 hour 40 minutes, leading three more departments to request their own sessions.