Skills Matrix Template for Teams

Start from the work, not from job titles. Walk through a normal quarter: the recurring tasks, the systems you run, the procedures you follow and the last few things that broke. Each one is a candidate skill. The US Government Accountability Office gives federal agencies similar advice: determine the critical skills and competencies¹ needed for current and future work, then develop strategies to close the gaps.

Name each skill narrowly enough to rate. "Networking" hides the one routing setup only one person understands, while "network and BGP routing" exposes it. Keep the list short enough to re-rate, say 8 to 20 skills for one team. Include the unglamorous ones, like a legacy system, a vendor portal or a compliance filing. Nobody volunteers to be the second person who knows those.

A rating only helps if two managers would give the same one, so describe each level by what the person can actually do. The US Office of Personnel Management's proficiency levels for leadership competencies² work this way. Their five levels run from Awareness, for someone who "requires close and extensive guidance," to Expert, for someone who "serves as a key resource and advises others."

Then give each skill a required level: the rating at which someone can own the work without oversight. For most everyday skills that is 3. Set it at 4 for work the team rarely does but can't afford to get wrong, such as a safety procedure or a regulatory filing.

Self-ratings are a useful start and a poor finish. Researchers Ethan Zell and Zlatan Krizan reviewed 22 meta-analyses in 2014³ and found that self-ratings of ability tracked measured performance only moderately. The average correlation was .29, where 1 would be perfect. Self-ratings were closer to the mark for specific skills and familiar tasks, one more reason to name skills narrowly.

So check each self-rating against recent work: who handled the last few incidents, who closed the books, whose work still needs review. Notes from your last performance review cycle often settle a rating faster than a debate. Date every rating, because an undated matrix looks current long after it stopped being true.

For each skill, count the people rated at or above its required level. Zero is a gap. One is a single point of failure: when that person is sick, on vacation or gone, the work waits. Two or more is covered. Leave learners out of the count, because three people at level 2 is a training pipeline, not cover.

Expect to find some, even on strong teams. Software engineers have a name for the project-wide count: the truck factor, the fewest people who would have to leave before a project stalls. In 2016, Guilherme Avelino and colleagues estimated it for 133 popular GitHub projects⁴: 65% came out at two or fewer.

Plenty of organizations never plan who steps in. In 2024, the Society for Human Resource Management reported that 56% of HR professionals in one of its surveys⁵ said their organization had no succession plan at all. A skills matrix asks that question for each skill, and every answer should name a person.

1. 1

   Check for a plan already in motion

   Match the skill's exact name against your development plans, not a person's reputation. A database course does not cover a routing gap. Count a plan only if it names the person, the skill, their current and target levels, and a quarter.

2. 2

   Pick the nearest successor

   Choose someone already at level 1 or 2 where you can. The climb to proficient is shorter, and a skill with one owner is the wrong place for the long route.

3. 3

   Plan real work, with backup

   In its book on site reliability engineering, Google recommends that new engineers shadow on-call shifts early and often⁶. Some teams then swap roles: the trainee handles incoming problems as the main responder, and an experienced engineer stays in the background to help. Copy that pattern for any skill, and check progress before the target quarter.

4. 4

   Decide on the rest

   For a single point of failure with no plan, choose this cycle: open a plan, hire, or accept the risk and record who decided. If it's a hire, cost it in a headcount plan before you ask for budget.

Ratings drift as people learn, leave and change roles. Re-rate quarterly if your tools or team change quickly, or twice a year if they don't. Update each successor's current level at every review, not only at the target quarter, so a plan running late shows up while there's still time to help.

Watch for ratings that never move. A rating unchanged for three cycles means either a real plateau or that nobody is checking. Both deserve a conversation at your next one-on-one.

When a sole owner gives notice, sit down with them before their last day and write up how they do the work. A role handover document gives you the outline.