A skills matrix is a grid with people down one side and skills across the top, and a score in every cell saying how good that person is at that skill. It answers one question that a staffing plan, an org chart and a capacity report all quietly assume somebody already knows: who can actually do this work. Most organizations answer it from memory, which is why the same four names keep appearing on every project.
Building the grid is trivial. You can do it in a spreadsheet before lunch. What decides whether the matrix is useful or decorative is the rating scale sitting underneath it. If level 3 is called "advanced" and nothing more is written down, then a cautious engineer and a confident one will score identical ability two levels apart, and every allocation decision made from those numbers inherits the error. The scale is the artifact. The grid is just how you display it.
Key takeaways
- Write behavioral anchors before you score anyone. A level is an observable claim ("has led this unsupervised on a live system"), never an adjective.
- Score self and manager independently, then look at the gap. The gap is more informative than either number, and it tells you where the matrix cannot be trusted yet.
- The output that justifies the whole exercise is the single point of failure count: skills where exactly one person can work unsupervised.
- Cap the matrix at 12 to 15 skills. A 60 skill matrix is always out of date, because nobody can re-score 60 cells per person twice a year.
- Head-count capacity plans over-count. Two people at level 2 are not one person at level 4, and pretending otherwise is how a plan looks fine and still misses.
What is a skills matrix?
A skills matrix is a table that maps each team member against each skill the work requires, with a proficiency score in every cell drawn from one shared, written scale. It is used to staff projects, find skill gaps before they become delivery risks, and see which capabilities depend on a single person. It is sometimes called a competency matrix or a skill set matrix.
That is the last definitional paragraph on this page. Everything below is the mechanics: what goes in the template, how to write a scale people score consistently, what the filled version looks like, and the two or three numbers you pull out of it that actually change a decision.
Skills matrix template, field by field
A skills matrix needs seven fields, and only three of them are the grid. The other four are what stop the grid rotting within a quarter. The right hand column below is the failure mode, because every one of these fields has a standard way of being filled in uselessly.
| Field | What it holds | How it gets filled in uselessly |
|---|---|---|
| Person | Named individual, plus their role and whether they are permanent, contract or vendor. | Roles only, no names. A matrix of roles tells you nothing about who is available on the 14th. |
| Skill | A specific capability the portfolio actually buys, phrased as work someone does. | "Communication", "leadership", "teamwork". Unscoreable, and nobody has ever been staffed on the basis of them. |
| Proficiency score | A number from the written scale, one shared scale for every skill. | A different informal scale per manager, or colored dots with no key. |
| Assessed by | Who gave the score: self, manager, peer review, or a test. | Left blank, so a self-reported 4 and a verified 4 look identical in the same column. |
| Assessed on | The date the score was set. | Omitted entirely. This is the most common single omission and it makes decay invisible. |
| Required level | The level the current portfolio needs, per skill, and how many people at it. | Not recorded, so the matrix shows what you have and never what you need. A gap needs both sides. |
| Interest | Whether the person wants to use or grow this skill. | Skipped. Then you build a development plan around someone who has quietly decided to stop doing that work. |
Swipe to see more →
The required level row is the one most templates leave out, and leaving it out is what turns a skills matrix into an inventory. An inventory is a list of what you own. A matrix is a comparison, and you cannot compare without the second number.
The rating scale is the whole design
A rating scale works when two people who have never spoken score the same person the same way. That only happens when each level is written as observable evidence rather than a word like "proficient". The scale below is a 0 to 4 anchored scale, and the anchor is always something you could be asked to produce proof of.
| Level | Label | Behavioral anchor (what must be observably true) | Can they be staffed on it? |
|---|---|---|---|
| 0 | None | No exposure. Would not know where to start. | No |
| 1 | Aware | Understands the concepts and the vocabulary. Has watched it done. Has not done it. | No, but can shadow |
| 2 | Assisted | Has done this on real work with someone reviewing every output before it went anywhere. | Only with a named reviewer |
| 3 | Independent | Has delivered this unsupervised on live work at least twice, including once when something went wrong. | Yes |
| 4 | Can teach | Level 3, plus has trained someone else to level 3, and is who the team escalates to. | Yes, and can cover others |
Swipe to see more →
Two details in that table do most of the work. The first is that level 3 requires a failure ("including once when something went wrong"), because handling the normal path is not the same skill as handling the exception, and projects fail on exceptions. The second is the last column, which converts a score into a staffing rule. A scale that does not tell an allocator what they are allowed to do with the number is decoration.
Use the same five levels for every skill. The temptation is to write a bespoke scale per skill because the skills are different, and it does produce more accurate individual scores. It also makes the matrix non-comparable across columns, which destroys the coverage analysis further down this page, and coverage is the reason to build the thing.
A filled skills matrix example
Here is a delivery team of six supporting a portfolio of integration projects, scored on the 0 to 4 scale above. The two right hand columns are the ones people skip, and they are where every real conclusion comes from.
| Skill | Priya | Marcus | Dana | Tom | Ines | Ray | Required | At or above required |
|---|---|---|---|---|---|---|---|---|
| Integration architecture | 4 | 2 | 1 | 0 | 3 | 1 | 3 | 2 |
| API design | 4 | 3 | 2 | 1 | 3 | 2 | 3 | 3 |
| Data migration | 2 | 4 | 3 | 1 | 1 | 2 | 3 | 2 |
| Mainframe interfaces | 1 | 3 | 0 | 0 | 0 | 0 | 2 | 1 |
| Security review | 2 | 1 | 4 | 1 | 2 | 1 | 3 | 1 |
| Release engineering | 1 | 2 | 2 | 4 | 2 | 3 | 3 | 2 |
| Vendor management | 3 | 1 | 2 | 2 | 4 | 1 | 2 | 4 |
| Test automation | 2 | 2 | 3 | 3 | 1 | 4 | 2 | 5 |
Swipe to see more →
Read the last column, not the grid. Three rows are in trouble. Mainframe interfaces has exactly one person at or above the required level, and that person is Marcus, who is also the only one above required on data migration. Security review depends entirely on Dana. Everything else has at least two people and can absorb a resignation, a holiday or a parallel project.
Notice what the matrix does not say. It does not say Priya is the best engineer, and it is not a performance rating. It says that if Marcus takes the six weeks of leave he is owed, two capabilities in this portfolio go to zero. That is a portfolio risk with a name and a date, which is a different class of object from "we should probably cross-train more".
How do you create a skills matrix?
Work backwards from the portfolio, not forwards from the people. List the skills the approved and pipeline work actually requires, write the scale, set the required level per skill, then score. Building it in the other order produces a list of what your team happens to know, which is an interesting document and not a useful one.
- List the skills from the work, not the CVs. Take the next two to four quarters of approved and likely projects and write down what each one needs somebody to be able to do. Phrase each as an activity. If you cannot imagine writing "needs one person at level 3" next to it, it is not a skill, it is a personality trait.
- Cut the list to 12 to 15. Merge near duplicates and drop anything only one project ever needed. This is the step everyone skips and it is the one that decides whether the matrix survives its second review.
- Write the scale with behavioral anchors. Five levels, one scale for all skills, every level an observable claim. Circulate it and argue about it before anyone scores anything, because arguing about the scale afterwards means rescoring everything.
- Set the required level and the required count per skill. Two numbers: how good does somebody have to be, and how many of them do we need at that level. The second number is what surfaces single points of failure.
- Score self and manager separately. Same scale, no conferring, then compare. Details in the next section.
- Reconcile in a conversation, not a spreadsheet. Where self and manager differ by two or more levels, talk about it against the anchors. The anchor makes this a factual conversation ("have you led one unsupervised?") instead of a negotiation about self esteem.
- Date every score and publish the matrix. A skills matrix kept private by a resource manager gets used once. Published, it gets corrected by the people in it, which is the cheapest quality control available.
Who should fill in the skills matrix?
Both the person and their manager, independently, on the same scale, before either sees the other's answer. Then you compare. Self assessment alone drifts in both directions and manager assessment alone is limited to work the manager happened to observe. The disagreement between them is a signal in its own right, not noise to be averaged away.
| Method | What it is good for | How it fails |
|---|---|---|
| Self assessment | Coverage. It is the only way to learn about skills used outside the current role. | Drifts both ways. Confident people overscore, experienced people underscore because they can see what level 4 really looks like. |
| Manager assessment | Calibration across a team, since one person applies one interpretation of the scale. | Blind to anything the manager has not personally seen, which on a matrixed team is most of the work. |
| Peer review | Accuracy on technical skills. Peers see the actual work. | Slow, and it starts measuring popularity if you let it run as a general survey. |
| Evidence or test | Settling disputed scores at the top of the scale, where the staffing consequences are real. | Expensive per skill, so it does not scale past the few skills that matter most. |
Swipe to see more →
The number worth tracking is the assessment gap: the average absolute difference between self and manager scores, per person and per skill. A gap under 0.5 levels means your anchors are doing their job. A gap above 1.0 means the scale is being read two different ways and the matrix is not yet safe to allocate from. A gap that is large in one direction for one manager usually means that manager is scoring against a private standard.
For the handful of skills where a wrong level 3 or 4 would put a project at risk, verifying is cheaper than discovering. A short standardized exercise that everyone completes the same way, or an independently vetted specialist brought in to review the work, will settle a contested score faster than three rounds of conversation, and it removes the awkwardness of a manager overruling someone's own assessment of themselves.
Single point of failure: the number this matrix exists to produce
Count the skills where exactly one person scores at or above the required level. That count, not the grid, is the output a portfolio should act on, because each one is a capability that stops entirely if a single named person is unavailable. In the worked example above the count is two: mainframe interfaces and security review.
| Single point of failure count | What it means | What to do about it |
|---|---|---|
| 0 to 1 skills | Healthy. Normal absence is absorbable. | Keep it there. Re-check when the skill list changes. |
| 2 to 3 skills | Manageable but real. Any two coinciding absences will hurt. | Name a deputy per skill and fund the time to get them to level 3. |
| 4 or more skills | The delivery plan is resting on individuals, not on a team. | Treat it as a portfolio risk with an owner. Stop taking on work in those skills until coverage exists. |
| Any skill on a Tier 1 project | Concentration risk on the work that matters most. | Escalate regardless of the total. One is too many here. |
Swipe to see more →
Treat those bands as a starting point to calibrate against your own portfolio rather than as a standard. The useful discipline is that the number is reported at all, and reported on the same slide as the delivery plan it constrains. A single point of failure that lives only in a spreadsheet gets discovered on the day the person resigns.
The related figure is bench depth: for each skill, how many people sit exactly one level below the required level. Those are the people a quarter of deliberate work turns into cover, and they are much cheaper than recruiting. A skill with nobody at required minus one has no internal path at all, and that is a hiring decision, not a training decision.
How do you use a skills matrix for resource planning?
Use it to weight capacity, not just to count it. A standard capacity plan counts available hours per person and compares them to demand, which silently assumes anybody can do anything. The skills matrix breaks that assumption, and the corrected number is almost always smaller than the one in the plan.
Take the mainframe interface work in the example. A head-count capacity plan sees six people with available hours. The matrix sees one person at or above the required level and one more at level 1 who could shadow. Effective capacity for that skill is roughly one person, not six, and the gap between those two numbers is exactly where a portfolio commits to dates it cannot hit.
| Planning question | What head-count capacity says | What the skills matrix says |
|---|---|---|
| Can we start the mainframe project in Q3? | Yes, 6 people have availability. | Only if Marcus is free. Effective capacity is one person. |
| Can we run two security sensitive projects at once? | Yes, hours are available. | No. Dana is the only person at required level, and she cannot review two in parallel. |
| What happens if Marcus leaves? | Nothing visible. Backfill one headcount. | Two capabilities drop below required. The mainframe work stops until a hire is productive. |
| Where does training money go? | No signal at all. | Mainframe interfaces and security review, in that order, at the people already at level 2. |
Swipe to see more →
This is why the skills matrix belongs next to portfolio capacity planning rather than in a filing cabinet in HR. Capacity planning tells you how many hours exist. The matrix tells you how many of those hours can be pointed at the work in front of you, and only the second number is safe to promise against.
Skills gap analysis: build, buy, or borrow
A skills gap is the difference between the required level and the highest level you actually have, per skill. Each gap has three possible answers and they differ mainly in how long they take, so the honest first question is not which is best but how many weeks you have before the work starts.
| Option | Typical lead time | Best when | Watch out for |
|---|---|---|---|
| Build (train someone already at required minus one) | One to two quarters | The skill is needed repeatedly, and somebody at level 2 wants it. | Training without protected time is a wish. Book the hours or it will not happen. |
| Buy (hire) | Two to three quarters to productive | The skill is permanent, load is steady, and no internal path exists. | The gap is open during recruitment and onboarding, not just until the offer is signed. |
| Borrow (contract or vendor) | Weeks | The need is time boxed, or you must start before a hire could land. | Contracting the same skill for four consecutive quarters is a hiring decision you have not made yet. |
| Descope or resequence | Immediate | None of the above fit the window and the date is real. | Nobody lists this as an option, which is why portfolios commit to work they cannot staff. |
Swipe to see more →
That fourth row is not padding. When a gap cannot be closed inside the window, the choice is between moving the work and missing the date, and the skills matrix is the evidence that makes moving the work a defensible recommendation rather than an admission. Bring it to the forum that owns sequencing, which is usually the portfolio board.
What is the difference between a skills matrix and a competency matrix?
In practice the terms are used interchangeably, and most templates sold as one are identical to the other. Where people do distinguish them, a skills matrix scores specific technical abilities and a competency matrix scores broader behavioral capabilities such as facilitation or decision making. The distinction matters less than whether the levels are anchored.
| Artifact | What it answers | Unit |
|---|---|---|
| Skills matrix | Who can do this work, and how well? | Person by skill |
| Competency matrix | Same question, usually with behavioral capabilities included alongside technical ones. | Person by competency |
| RACI matrix | Who is accountable, responsible, consulted and informed for this activity? | Role by activity |
| Resource breakdown structure | What categories of resource does this project consume? | Hierarchy of resource types |
| Resource histogram | How much of a resource is committed, period by period? | Quantity over time |
Swipe to see more →
The pairing that causes the most confusion is the skills matrix and the RACI. A RACI says Dana is Accountable for the security review. The skills matrix says Dana is the only person who could do it. Those are different failures when they go wrong: a RACI gap means nobody owns the work, a skills gap means nobody can do it, and no amount of clarifying ownership fixes the second one.
Can you build a skills matrix in Excel?
Yes, and for a team under about 30 people a spreadsheet is the right answer. Put people in rows, skills in columns, scores in the cells, and conditional formatting to shade anything below the required level. Add a row under the grid that counts how many people are at or above required per skill, because that row is the whole analysis.
The layout that works: freeze the top row and the first column, put the required level in a row directly beneath the headers so it is visible while you scroll, and keep the assessed date in a parallel sheet with the same shape rather than crammed into the cell. Colored dots without a legend are the most common Excel version and the least useful, because the moment the file leaves your screen nobody knows whether green means "can do it" or "can teach it".
Where the spreadsheet genuinely runs out is scale and freshness. Past roughly 50 people or 20 skills, keeping scores current becomes a job rather than a task, version conflicts start deciding what is true, and you have no history to see whether coverage is improving. That is the point to look at software, and the trigger is the update burden, not the size of the grid.
How often should a skills matrix be updated?
Twice a year for the full matrix, plus an immediate update whenever someone joins, leaves, changes role, or finishes a project that used a skill at a new level. The full re-score is the calendar item people plan for. The event driven updates are what actually keep it accurate, and they take minutes.
| Skill type | How fast the score decays | Re-assess |
|---|---|---|
| Specific tool or platform versions | Fast. Unused for a year, an honest score drops a level. | Every 6 months |
| Regulated or certified skills | Set by the certification, not by you. | On the certificate expiry date |
| Domain and process knowledge | Slow, but resets when the process changes. | Annually, or on a major process change |
| Facilitation, negotiation, leading work | Very slow. These hold once genuinely reached. | Annually |
Swipe to see more →
This is what the assessed on date is for. A score with no date cannot be aged, so a level 4 set three years ago on a platform the person has not touched since reads exactly like one set last month. Sort the matrix by date once and you will usually find a third of it is older than the skills it describes.
Five ways a skills matrix goes wrong
These are the recurring failures, in rough order of how often they show up.
- Too many skills. Sixty rows built by asking everyone what they can do. It is complete on the day it is finished and stale a month later, because re-scoring is now a week of work nobody will fund. Twelve to fifteen skills, chosen from the work.
- Unanchored levels. "Beginner, intermediate, advanced, expert" with no definitions. Every manager applies a private standard, scores are not comparable between teams, and the coverage count built on top of them is fiction.
- It becomes a performance review. The moment people believe scores affect pay, every score goes up and the matrix stops describing reality. Keep it visibly separate from appraisal, say so explicitly, and never hand the raw grid to a compensation process.
- No required level. The grid shows what you have with nothing to compare it to, so it produces no decisions. This is the difference between an inventory and a matrix.
- Built once, never used. Nobody consults it when staffing a project, so nobody maintains it. The fix is procedural: make the coverage row a required input to the staffing decision, and the matrix maintains itself because people need it to be right.
Frequently asked questions
What should be included in a skills matrix?
Named people with their employment type, 12 to 15 specific skills drawn from the work in the portfolio, a proficiency score per cell from one shared written scale, who assessed it, the date it was assessed, the required level and required headcount per skill, and each person's interest in using or growing that skill.
What rating scale should you use for a skills matrix?
A five level scale from 0 to 4 where every level is a behavioral anchor rather than an adjective, for example "has delivered this unsupervised on live work at least twice". Use the same scale for every skill so scores stay comparable across columns, which is what makes the coverage analysis possible.
How many skills should a skills matrix track?
Twelve to fifteen for a delivery team. The constraint is not what you can list, it is what you can honestly re-score twice a year for every person. Beyond about twenty skills the matrix stops being updated, and an out of date matrix is worse than none because people still trust it.
What is a skills gap analysis?
Comparing the required level for each skill against the highest level you actually hold, then counting how many people meet the requirement. The gap has four possible responses: train someone already close, hire, contract, or move the work. Which one fits depends mostly on how many weeks you have.
What is the difference between a skills matrix and a RACI matrix?
A skills matrix scores capability, mapping people against skills to show who can do the work. A RACI matrix assigns accountability, mapping roles against activities to show who owns each one. A RACI gap means nobody owns a task. A skills gap means nobody can do it, and clarifying ownership will not fix that.
Who owns the skills matrix in a project organization?
The resource or capability manager who owns the people, with the PMO owning the format and the review cadence. In a matrix organization that is usually the functional manager, since they own careers and development. Project managers are consumers of it, not owners, or the scores will bend toward the current project's needs.
How do you stop people inflating their own skill ratings?
Anchor every level to something checkable, score self and manager independently, and keep the matrix visibly separate from pay and appraisal. Most inflation is not dishonesty, it is an ambiguous scale being read optimistically. When level 3 requires two unsupervised deliveries, there is much less room to interpret.
Does a skills matrix need to include interest or willingness?
Include it as a separate column, and keep it out of the proficiency score. Someone at level 4 who no longer wants that work is a retention risk and a bad development investment, and you cannot see either if capability and willingness are collapsed into one number. It is one extra column and it changes staffing conversations.
How do you handle contractors in a skills matrix?
Include them, and flag the employment type and end date in the person field. Excluding contractors makes coverage look better than it is, since a capability held only by someone whose contract ends in March is a single point of failure with a known date. That is exactly the risk the matrix should surface.
Is a skills matrix the same as a training matrix?
No. A training matrix tracks whether required training has been completed and is current, which is a compliance record with a yes or no in each cell. A skills matrix scores how capable someone is on a graded scale. Completing a course and being able to do the work unsupervised are different claims.
Where the skills matrix sits in the wider portfolio
The matrix answers who can do the work. What you do with that answer belongs to the rest of the resource cluster. Turning capability into a staffing decision is resource allocation in project management, and the argument that happens when two projects want the same qualified person is resource contention. How much of that person is genuinely available is resource utilization, and projecting the demand before the work is approved is resource forecasting.
For the planning artifacts themselves, the skills and levels feed the resource management plan, the period by period commitment shows up in a resource histogram, and the categories the matrix scores against should line up with the resource breakdown structure. Smoothing a plan around a scarce level 4 is covered in resource leveling and smoothing.
On the organizational side, who does the scoring and who owns careers is set by the matrix organization structure, and accountability for individual activities stays in the RACI matrix. When a single point of failure needs to be treated as a portfolio risk with an owner and a date, it belongs on the risk matrix and in front of the forum described in project governance framework.
Last updated August 2026.