Skills matrix template: how to build one and test it against real evidence
A skills matrix template is a grid with people (or roles) down the side, skills across the top, and a rating in each cell. This guide gives you that template as copy-ready tables, a level scale with observable anchors, and a method for testing the ratings with short skill checks, so the matrix records what people can do and not only what they say.
It is written for L&D managers, training company operations leads and team leads who need to staff work, plan training or spot coverage risk for a team of roughly 5 to 50 people. Copy the tables below into Excel or Google Sheets; there is no file to download.
Key takeaways
- Build the matrix to support one decision, then list only the 8 to 12 skills that decision needs.
- Define each level by what a person can do without help, so two raters read the same cell the same way.
- Tag every rating with its evidence (self, manager, observed or tested). A self-rated level is a claim, not a record.
- Test the cells that a decision depends on, and change a level only when a second source agrees.
What is a skills matrix, and what decision should yours support?
A skills matrix is a grid that shows which people have which skills, and at what level. Build it to support one decision. That decision tells you which skills to list and how precise the ratings must be.
Three decisions suit a first matrix:
- Coverage: who can deliver or support this work today?
- Development: which gaps should training close first?
- Staffing: who is ready for a new assignment?
Pick one for the first version. A matrix built for all three tends to grow to 40 skills and then goes unmaintained.
What should a skills matrix template include?
A working skills matrix template has two sheets. The first defines each skill, its target level and the evidence you will accept. The second records each person’s level and the evidence behind it. Several templates reviewed for this post show only the second sheet, which is where rater disagreement starts.
Sheet 1 is the skill definitions. The rows below are illustrative examples for a training team; replace them with your own.
| Skill | Observable description | Target level | Accepted evidence |
|---|---|---|---|
| Facilitate a live session (F) | Runs a session to plan, adjusts pace, answers learner questions correctly | 3 | O |
| Write quiz items (Q) | Writes items that match a stated objective, each with one defensible answer | 3 | T or O |
| Read LMS reports (R) | Pulls a completion and score report and explains it to a client | 2 | T or M |
| Handle escalations (E) | Resolves a learner complaint within policy and records it | 3 | O or M |
| Coach a new trainer (C) | Gives specific feedback after observing a session | 2 | O or M |
Write each description as “can do this, to this standard.” A phrase like “good communicator” cannot be observed, so split it into the actions that show it. Several published guides also add an interest column, which helps when the decision is development.
How do you write skill levels that two raters read the same way?
Define each level by what the person can do without help, and what happens when the case is unusual. Published guides use anywhere from three to ten levels. Fewer levels are easier to rate consistently, so add one only when you can write a different observable anchor for it.
For comparison, the SFIA skills framework defines seven levels of responsibility, from “Follow” to “Set strategy, inspire, mobilise.” A team matrix rarely needs that many; five levels is enough here.
| Level | Label | What you would see |
|---|---|---|
| 0 | Not yet | Has not done the task, or cannot do it even with guidance |
| 1 | Follows | Does it with step-by-step guidance and checks along the way |
| 2 | Assists | Does routine cases alone; asks for review on unusual ones |
| 3 | Independent | Does routine and unusual cases alone, to the standard |
| 4 | Coaches | Does it independently, and others learn it from this person |
This scale is an editorial design, not a published standard. Its top level echoes the “can train others” wording that several guides use. Set targets per skill, because not everyone needs a 4.
How do you fill in the skills matrix template without relying on self-ratings?
Record a level and an evidence code in every cell. Self-rating is a fast first pass, but a 2006 JAMA systematic review of physician self-assessment found that 13 of 20 comparisons with observed measures showed little, no or an inverse relationship. That review covers physicians, not your team, and the authors call the evidence suboptimal in quality. It is still the reason to tag each rating with its source.
| Code | Evidence | Speed | Main weakness |
|---|---|---|---|
| S | Self-rated | Fast | Can be far from observed performance |
| M | Manager or peer judgment | Fast | Only as good as what the rater has actually seen |
| O | Observed against a checklist, or a work sample scored with a rubric | Slow | Needs assessor time and a trained second rater |
| T | Scored check, such as a quiz or practical test | Medium | Measures only what the check covers |
Treat cells with only an S as claims. Use them to decide which cells to check, not to make decisions.
Pro tip: Store the Date, Not Just the Level. A 3 rated two years ago and a 3 rated last month are different facts. Add a rated-on date and a rater name to each cell, and shade anything older than your review period so stale ratings are visible.
What does a filled-in skills matrix template look like?
Each cell holds a level and an evidence letter, so 3M means level 3 on a manager’s judgment. Under the grid, count how many people meet the target for each skill. In this illustrative five-person team, that one count shows a coverage problem that an average score would hide.
| Person | Facilitate (F) | Write items (Q) | LMS reports (R) | Escalations (E) | Coaching (C) |
|---|---|---|---|---|---|
| Target level | 3 | 3 | 2 | 3 | 2 |
| Asha | 4M | 3S | 2M | 3M | 3M |
| Ben | 3O | 2S | 3S | 2M | 1S |
| Chen | 3M | 4S | 1M | 3M | 2M |
| Dana | 2M | 2M | 2M | 4S | 0M |
| Eli | 1M | 1M | 4S | 1M | 0M |
| At or above target | 3 of 5 | 2 of 5 | 4 of 5 | 3 of 5 | 2 of 5 |
For a person, the gap is the current level minus the target; a negative number is a development need. For a skill, use a simple coverage rule: fewer than two people at target on a skill you must deliver is a coverage risk. That threshold is an editorial judgment, so set your own.
Item writing sits at 2 of 5, which looks acceptable. But both people at target, Asha and Chen, are rated only by themselves. That is the first cell to test.
How do you test a skills matrix against real performance?
Pick the cells where a decision depends on the rating and the evidence is weakest, check them with a short test matched to the skill type, and change a level only when a second source agrees. Testing every cell is slow and rarely needed.
Which cells should you test first?
Start with cells that meet any of these conditions. Item writing in the example grid meets the first and third.
- The rating is self-only and the skill is needed for delivery.
- Self and manager ratings differ by two or more levels.
- Only one or two people reach the target level.
Which kind of check fits which skill?
A quiz can show whether someone knows the rules, but not whether they can run a room.
| Skill type | Check | Example from the grid | Scored by |
|---|---|---|---|
| Knowledge of rules or facts | Multiple-choice quiz of 8 to 12 items | Item-writing rules (Q) | Auto-scored |
| Judgment in realistic cases | Scenario-based questions | Escalation decisions (E) | Auto-scored or rubric |
| Hands-on performance | Observation checklist with Met and Not met | Running a session (F) | Trained observer |
| Work product | Work sample scored with a rubric | Two items written to a brief (Q) | Two raters if the decision matters |
For hands-on skills, an observation checklist for practical training gives you criteria that two assessors can score alike. For work products, use an assessment rubric with written descriptors for each level.
How do you turn a quiz score into a level without over-reading it?
Map score bands to levels before anyone sits the quiz, treat the result as one piece of evidence, and act on a mismatch only when it is two levels or more. A short quiz is coarse, so a one-level difference can easily be noise.
The bands here are illustrative placeholders for a 10-item quiz: under 40% maps to level 1, 40 to 59% to level 2, 60 to 79% to level 3, and 80% or more to level 4. Set real bands with a defensible method; the guide to cut scores for professional training covers the options. These illustrative results use the item-writing quiz.
| Person | Self rating | Quiz score | Band level | Difference | Action |
|---|---|---|---|---|---|
| Asha | 3S | 72% | 3 | 0 | Record as 3T |
| Ben | 2S | 68% | 3 | +1 | Keep 2, log the result, recheck later |
| Chen | 4S | 48% | 2 | -2 | Review a scored work sample before changing |
Suppose a work sample for Chen, scored by two raters with a rubric, also lands at level 2. Chen’s cell becomes 2O, and item-writing coverage falls from 2 of 5 at target to 1 of 5: Asha alone. That is a training or hiring decision a color-coded cell would not have shown.
Pro tip: Keep One Fixed Form per Skill. Give everyone the same 8 to 12 items for a skill, and keep those items out of course slides and handouts. Then a score means the same thing for every person and every round.
How do you keep ratings consistent between raters?
Have two raters score the same few cells independently, compare the results, and discuss every difference before rating the rest. Report agreement carefully: simple percent agreement can overstate it, because it ignores the chance that raters guessed. McHugh’s 2012 article on the kappa statistic explains why, and the observation checklist guide above includes a worked kappa example.
How often should you update the matrix, and who should see it?
Re-rate a person when something changes, such as finishing a course, changing roles or adopting a new tool, and review the whole grid once or twice a year. Those intervals are editorial suggestions. Ratings are information about named people, so decide who sees them and tell staff how they will be used first.
If matrix results will feed hiring, promotion or pay, treat the checks as tests. The Standards for Educational and Psychological Testing from AERA, APA and NCME is the reference for judging whether a test supports a decision. Employment and data-protection rules vary, so check with HR or counsel; this is general information, not legal advice.
What should you do next?
Pick one team and one decision. Copy Sheet 1 into a spreadsheet, write 8 to 12 skills in the “can do this, to this standard” form, and set targets. Rate everyone with evidence codes, then test the self-rated cells.
A spreadsheet is enough for one team. If the checks need a place to be authored and scored, SimpliAssess’s features page lists item banking, AI-assisted authoring, tag management and analytics and reporting. SimpliAssess’s published pages do not describe a skills matrix or competency mapping, so the grid itself stays in your spreadsheet.
If you would rather track skills inside a training platform, SimpliTrain, our separate training management system with a built-in LMS, lists Skill Tracking among its features. Its skill tracking page describes a tag-based skill framework that connects training activity to competency data and identifies skill coverage gaps. It does not describe proficiency levels or a matrix grid, so ask about those when you request a SimpliTrain demo.
Frequently asked questions
What is the difference between a skills matrix and a competency matrix?
Guides disagree. Several, including eLearning Industry’s, use skills matrix for a grid of specific abilities used in staffing and training, and competency matrix for a broader, more strategic view. Many other pages use the terms interchangeably. For a team grid, the label matters less than defining each level in observable terms.
How many proficiency levels should a skills matrix have?
Published guides use anywhere from three to ten levels, and SFIA defines seven levels of responsibility for its skills. For a team matrix, three to five is enough. Choose the number by asking whether you can write a different observable anchor for each level. If you cannot, remove the level.
How many skills should a skills matrix include?
Start with 8 to 12 for one team and one decision. That is an editorial rule of thumb; Workhuman’s guide suggests 10 to 15 per role or team. A longer list is harder to rate honestly and to keep current. Add a skill only when someone will act on a gap in it.
Can I build a skills matrix in Excel or Google Sheets?
Yes. Two sheets are enough: one for skill definitions and targets, one for people and ratings. Copy the tables in this post into a spreadsheet, add data validation so cells accept only the levels you defined, and use a counting formula for people at target. Move to software only when upkeep or access control becomes the problem.
How do you test whether employee self-ratings are accurate?
Compare them with a second source on a sample of cells: a scored quiz for knowledge skills, an observation checklist for hands-on skills, or a rubric-scored work sample for outputs. Treat a gap of two or more levels as a reason to look again, not as proof, and change the level only when another source agrees.
How often should you update a skills matrix?
Update a person’s row when something changes, such as finishing a course, changing roles or adopting a new tool, and review the whole grid once or twice a year. These intervals are editorial suggestions. Store a rated-on date in every cell so you can see which ratings have gone stale.
