Competency Framework

What is a competency framework actually made of?

It has three layers, and only the bottom one does any work. A competency is a named capability or way of operating the organization has decided matters, described in a sentence or two rather than as a task. Levels express that same competency at increasing scope, so the named thing means something different for somebody starting out and somebody running a function. Behavioral indicators sit underneath, stating what each level looks like when a person is actually doing it, in terms an observer could recognize. The indicators are the framework. The competency names and level labels above them are filing: useful for navigation, worthless as evidence. That is why a framework circulated as a list of names with one-line descriptions produces confident disagreement rather than shared judgment, and why the drafting effort belongs almost entirely at the bottom.

How is a competency framework different from a skills matrix?

They answer different questions and are routinely treated as the same document. A competency framework describes the behavior expected of anyone operating at a level, stated in advance and independent of who happens to be employed. A skills matrix records what named people hold today, capability by capability, and changes whenever somebody joins, leaves or learns something. One is a standard; the other is an inventory. The confusion carries a practical cost. An employer that has built a matrix and believes it now has a framework can say who can do what, and cannot say what a promotion to the next level would require. An employer with a framework and no matrix can describe the standard precisely and has no idea how many people meet it. Both are useful, and neither substitutes for the other.

Where does job architecture end and a competency framework begin?

Job architecture defines the structure: which families of work exist, how many levels each carries, and what those levels are called. A competency framework describes the behavior expected inside that structure. The architecture says a level exists and where it sits; the framework says what somebody operating at it does. Built in the wrong order the two collide, because level descriptors written as part of the architecture start describing behavior, the framework then describes the same behavior in different words, and managers acquire two authoritative statements of one expectation that do not quite agree. The cheap fix is deciding which document owns behavior before either is drafted, and keeping the other one referring to it rather than restating it. Restated text always drifts, and the drift is discovered during a promotion conversation rather than in a review.

Why do competency frameworks quietly stop being used?

Three failures account for most of it, and the first two show early. A framework carrying more competencies than anybody can hold in mind stops being consulted and starts being skimmed: a manager preparing for a conversation reads what fits on one screen and guesses the rest. The second failure is abstraction. Indicators written in language that sounds unarguable, describing somebody who communicates well and takes ownership, cannot be checked against anything, so two managers reading the same person against the same indicator reach opposite conclusions and neither can be shown to be wrong. What presents as a disagreement about the person is a defect in the sentence. Both are fixable while drafting, and far cheaper to fix before publication than afterwards, when every correction reads as the employer moving the standard mid-cycle.

The third failure is decisive and harder to see, because the framework looks healthy the whole time it is happening. If no decision depends on it, the framework will be refreshed on a cycle and ignored in every conversation that matters. A rating entered in [performance management software](/performance-management-software) that nobody opens before deciding a promotion, an interview guide whose competencies were chosen by whoever built the template, a development plan written in borrowed language: in each case the framework is present and inert. The test is not whether people can find it, but whether a manager who ignored it would have to explain themselves to somebody. Where the answer is no, the honest response is either to attach it to a real decision or to stop maintaining it, rather than to run another launch.

What goes wrong when hiring, appraisal and promotion each keep their own copy?

They diverge, slowly, and nobody decides to do it. Hiring needs indicators that can be evidenced in an hour with a stranger, so it shortens them. Appraisal needs indicators describing a year of work, so it lengthens them. Promotion needs indicators separating somebody performing well at a level from somebody already operating at the next one, so it sharpens the top band and neglects the rest. Each adaptation is reasonable alone, each made by a different team solving a different problem. What emerges is three documents using identical competency names for three different standards, which is worse than three openly different frameworks, because nobody can see the disagreement. Someone is hired against one definition, appraised against a second and declined a promotion against a third, and experiences that as the employer moving the line.

The fix is one definition, with each consumer deciding only how much evidence it needs, never what the words mean. Hiring can assess a subset of competencies rather than a shortened version of all of them. Appraisal can ask for evidence across a longer period against identical indicator text. Promotion can require the indicator at the next level rather than a rewritten one. The discipline holds when the definition lives in one place every consumer reads from, usually alongside the employment record inside the [HR software](/hr-software) rather than in a deck each team copies, because the copy is where the fork begins. It also needs somebody with authority to decline a rewording requested for one team's convenience, which buys a little unpopularity in exchange for the only property that made the framework worth building.

How do you write an indicator two managers will read the same way?

Describe what the behavior looks like while it is happening, then ask whether somebody in the room could have observed it. An indicator survives that check if it names an action taken, a decision made or a piece of work produced. It fails if it names a disposition, because a disposition can only be inferred, and two people inferring from identical evidence are each entitled to a different answer. Grain is the harder judgment. An indicator specific enough to be unmistakable usually describes one team's work and cannot travel outside it; one general enough to cover everybody is too loose to falsify. The workable position is a general indicator with room for a function to attach its own examples, so the standard stays common while the evidence stays recognizable to the people measured against it.

Who writes them matters more than the drafting technique. Indicators produced entirely by a central team read as an outside description of work the writers have not done, and the people being assessed can tell, which costs the framework the credibility it needs before its first difficult conversation. Indicators produced entirely inside a function generate local vocabulary and no comparability, which is the problem the framework existed to solve. The arrangement that holds names one owner responsible for the shape, the levels and the wording of the common competencies, with practitioners in each function supplying the examples that make them concrete. Wherever the finished text is stored, whether an [HRMS](/hrms) or something plainer, one version has to be the one everybody reads, and the drafts that produced it should stop circulating.

How does a framework drift away from the work it describes?

The work moves and the document does not. Tools change, a function reorganizes, work that used to sit at the center becomes something the employer buys rather than does, and the indicators describing it stay in place, still signed off, still the basis on which somebody's promotion gets decided. A framework describing work the organization no longer does is worse than having none, because it remains authoritative. Nobody treats it as obviously wrong; they treat it as a standard they are failing to meet, or assess against what the job actually involves while recording a rating against what the document says. Both responses damage what the framework was for, and the second is nearly invisible, because the records look consistent right up to the moment somebody compares them against the work.

Time-based review catches this eventually and slowly. The cheaper trigger is a signal that already exists: the evidence people bring to rating conversations. Where managers repeatedly justify a rating with work matching no indicator, the framework has fallen behind the job, and that shows long before a scheduled review would. Reorganizations are the other reliable trigger, since a function whose remit has changed is describing different work the following week. Noting where an indicator had to be stretched to fit real work, and in which function, gives the owner a short list to revisit instead of a whole document to reopen. Rewriting narrowly and often is far cheaper than the alternative, a rebuild commissioned once the framework has already lost its authority, which then inherits the skepticism the old one earned.