Skills Matrices That Do Not Rot After Three Months

Someone builds a skills matrix. Everyone fills in their own scores from one to five over a couple of weeks. It looks great in a meeting. Three months later nobody has opened it, and the one person who does discovers half of it is wrong.

A skills matrix stays current when it is filled in from work people actually did rather than from what they say they can do, when it is small enough to maintain, and when it is used often enough that being wrong causes visible trouble. A grid that nobody consults is a grid nobody will correct.

Why self rating goes stale immediately

Three things happen at once.

People rate themselves against different internal scales. One person's four is another person's two, and neither is lying. The scores are not comparable across the grid, which is the one thing a matrix is supposed to give you.

The grid captures a moment. Someone learns a system in April and the sheet still shows the March answer. Nothing in the process makes them go back.

And the whole thing is optional. Updating it helps nobody today, so it loses to real work every single time.

Build it from work, not from opinions

The alternative is to derive the picture from evidence you already generate.

When work is assigned and completed, that is a record of who touched what. Someone who has shipped three pieces of work in the billing system this year has a real relationship with the billing system, whatever they would have put on a form. Someone whose last contact with it was eighteen months ago does not, even if they still rate themselves highly out of pride. A new joiner has no work history to derive from yet, which is why what the hiring panel already learned should reach their manager before day one.

Two columns get you most of the value:

• Has worked on it, recently. Derived from assignments and completed work in some sensible window.
• Has been the person others asked. Derived from reviews, approvals, or simply from asking a lead once a quarter who they route questions to.

Neither needs a rating scale. Recency and frequency carry more useful information than a self assessed number, and both update themselves as work happens.

Keep it small enough to survive

Most matrices die of width. Someone lists every tool, language and process the company touches, and the grid becomes forty columns that nobody can hold in their head.

Cut it to the things you actually staff against. If you have never once had to decide who works on it, it does not need a column. A short grid that is right beats a long one that is aspirational.

Review the columns themselves once or twice a year. Skills leave the business as well as people, and a column for a system you retired is worse than no column at all because it makes the rest look untrustworthy.

Use it for staffing or it will rot

This is the part that keeps everything else alive. The matrix has to be the thing someone opens when deciding who does the next piece of work.

That means it needs to answer real questions:

• Who could take this on if the usual person is away, which only holds if the leave calendar and the project board agree
• Where do we have exactly one person and no second
• Who has asked to learn this and has not had the chance
• If this person leaves, what goes with them

Notice that the last two are the interesting ones. A matrix used only to find the best available person concentrates work on the same people forever. A matrix used to find the second best person deliberately, once in a while, is how the gaps close. Freelancers belong in the same picture, alongside the pay dates and milestones they are tied to.

When the grid is used weekly, errors surface within days, because someone gets staffed on something they cannot do and says so. That feedback is what keeps it honest. No amount of quarterly reminders achieves the same thing.

The mechanics

The derived approach only works if the work records and the people records are in the same place. If assignments live in one system and the skills grid lives in a spreadsheet somewhere else, you are back to manual updates and it will rot again.

SicherOne keeps project management and HR on one set of records, which is what makes a picture like this maintainable as a byproduct of ordinary work rather than as a separate exercise. Where AI agents help draft or summarise, a person approves the output before it is used. It is sold per seat and the modules are separable.

If you are not in a position to change systems, the fallback is a small quarterly ritual: a lead sits with the last quarter's completed work in one hand and the grid in the other and corrects it. Half an hour, four times a year, based on evidence. That is still far better than a hundred self assessments nobody trusts.