An HR manager at an SMB opens Excel, types 50 names and 25 skill columns, and three hours later ends up with nothing useful. A skills matrix is not a tracking spreadsheet. It is a decision tool — where to hire, where to train, and who to pair as mentor.
For a team under 100, that matrix is built in a day if you follow six steps. Most SMBs skip it because they conflate two levels: formal job titles and actual operational skills. This guide separates both.
What a skills matrix actually solves
A skills matrix is a grid: employees on the left, competencies on top, scores in each cell. The output is three decisions HR needs over any twelve-month cycle:
- Gaps before reorganisation. Which department lacks bench for a key skill.
- L&D budget priority. Which three competencies to fund before the rest.
- Mentoring pairs. Who can teach whom internally without an external vendor.
What it is not: an annual development review, a performance review, or a job description. It does not replace any of those. The matrix is a snapshot.
Six steps to build it
1. Narrow the scope
One matrix covers one department or one function. Mixing IT and sales into one grid is unreadable. For a 60-person team across three departments, build three matrices, not one grid with 70 columns.
2. Pick 10 to 15 skills
Over 20 columns triggers score fatigue — the rater rushes and the data turns into noise. Pick three buckets of 4 to 5 skills each:
- Technical (role-specific: SQL, CNC programming, TIG welding)
- Process (ISO 9001 awareness, GDPR basics, internal escalation)
- Behavioural (structured problem-solving, mentoring, feedback)
Keep each skill a one-line label. "SQL — write intermediate queries with JOINs" beats "Advanced SQL including optimisation".
3. Use a four-level scale
Five levels add nothing; four force a decision:
- 0 — unaware
- 1 — basics with supervision
- 2 — independent in standard cases
- 3 — can mentor others
"Advanced" and "expert" are banned labels. They are ambiguous.
4. Score twice
The line manager and the employee score independently. Compare the delta:
- Same scores = accept the result.
- Delta of 1 = short conversation, agree on one score.
- Delta of 2 or more = not a dispute. The manager and employee read the skill differently. Rewrite the label.
Skipping this step is the most common mistake. Self-scores alone inflate numbers; manager-only scores miss blind spots.
5. Visualise the gaps
Use conditional formatting in Excel: red (0), orange (1), yellow (2), green (3). In three seconds you see where one person is the sole green in a column of 0s and 1s. That is a single-point-of-failure skill.
Set two thresholds:
- Coverage threshold: how many people in a column must sit at 2 or above.
- Depth threshold: how many must sit at 3 (mentor level).
For a 20-person team: coverage 3, depth 1. For a 60-person team: coverage 6, depth 2.
6. Attach an action to each gap
Every gap gets an owner and a deadline. Four action types:
- Internal mentor. A level-3 employee brings two colleagues to level 2.
- Microlearning module. For process skills under 15 minutes.
- External training. For specialised certificates (CNC, welding permits).
- Hire. Gap with no internal candidate and the skill is non-optional.
Keep the deadline realistic: 3 months for microlearning, 6 for mentoring, 9 for an external certificate.
Example 1: 45-person digital agency
A European digital agency, 45 people, 60 % junior to mid-senior. HR builds the matrix for the Delivery department (project managers, strategists, UX): 18 people, 12 skills.
Skills chosen: Figma, HTML/CSS reading, SEO on-page, GA4 reporting, structured kickoff, mentoring, GDPR basics, internal process standard, timeline estimation, risk log, retrospectives, English B2+.
Gaps after scoring:
- GA4 reporting: 2 people at level 3, 4 at level 2, 12 below level 2. Coverage below threshold.
- Risk log: 0 at level 3, 2 at level 2. Nobody mentors.
- Figma: covered, in fact over-deep (8 at level 3) — potential for cross-staffing.
Actions taken:
- A 90-minute GA4 bootcamp led by an internal senior strategist, two sessions, 10 people per session.
- One-day external risk-management workshop for four project managers.
- Two Figma experts start mentoring newcomers in brand design and content.
Execution window: 11 weeks. Target matrix state: coverage 6, depth 2 on both bottlenecks.
Example 2: 80-person plastics plant
A plastics manufacturer, 80 operators plus 12 maintenance staff. HR builds the matrix for maintenance: 12 people, 14 skills.
Skills: hydraulics, pneumatics, Siemens S7 PLC, welding, machine safety inspection, internal fault log, SAP PM module, solo night shift, English A2+, mentoring juniors.
Gaps:
- Siemens S7 PLC: one expert at level 3, the other 11 below level 2. Dependency on one person.
- Solo night shift: 4 autonomous people, on the edge (target is 6).
- SAP PM module: coverage of 3 out of 12 is enough.
Actions:
- PLC: a four-week shadow programme with the main expert and three juniors; one external Siemens certificate for one of them.
- Night shift: two maintenance staff assigned to a mixed-shift rotation for two months.
- The yearly machine-safety refresher stays on the calendar as required by the local supervisory authority.
Important: the production matrix must be kept separate from pay reviews. Operators cooperate only when the matrix cannot be used against them at the negotiating table. Communicate the intent up front.
For the link to learning design, see our guide on learning objectives.
Review cadence
How often you refresh the matrix depends on how fast the team changes:
- Annually for stable production teams.
- Half-yearly for growing tech, sales, or agency teams.
- Quarterly for teams with turnover over 25 % a year or in transition.
A refresh takes 1 to 2 hours per manager once the structure exists. The first build takes 3 to 5 hours.
Common mistakes
- Too many skills. 30 columns means nobody scores carefully. Keep it under 15.
- Five levels instead of four. "3 — average" becomes the default for everyone. Four levels force a call.
- Self-score only. Inflates scores by 1 to 2 levels, mostly on technical skills.
- Matrix as a pay record. Cooperation vanishes immediately. Keep the purpose developmental.
- Updating only inside the annual review. Too late. The review should use the matrix, not build it.
Disconnection from the L&D budget is the fifth mistake. Once the matrix shows gaps, HR needs approved budget for training. For a framing of that, see L&D budget per employee.
What to do after three months
The first matrix is rarely perfect. After three months HR will see which skills were poorly defined and which levels are unusable. That is normal. The second iteration is much faster — people already understand the scale and managers know the calibration.
A skills matrix is not a one-off project; it is a steady pulse on the team. An LMS like Mentor stores the matrix as a living document, links scores to ready-made modules, and triggers follow-up actions without more spreadsheet admin. What that frees up is time to discuss the real gaps — not to maintain a grid of cells.