How to write competency descriptors

Content aligned to the Capability Guide PDF for this topic. Q2 2026 refresh.

Why do descriptors decide whether scores mean anything?

A skills matrix is only as reliable as the words behind its numbers. Vague level labels like "good at data" guarantee that two managers score the same person differently. Sharp, observable competency descriptors fix that: they turn Level 3 from a number people argue about into behaviour everyone can recognise, evidence, and defend.

CIPD's Labour Market Outlook reports that a large share of UK employers still face hard-to-fill vacancies linked to capability, not headcount alone (Chartered Institute of Personnel and Development, 2024). Clear standards for what "capable" looks like in your context make assessment and development conversations honest. Descriptors are where that standard lives. Get them wrong and every figure in the grid inherits vagueness; get them right and consistency, fairness, and audit defensibility follow.

This guide is the writing craft: weak versus strong examples, four ingredients, a full ladder for one skill, a five-step process, mistakes, and an edge case when HR and operations share ownership of wording.

What is a competency descriptor?

A competency descriptor — sometimes called a behavioural indicator — is a short statement of what a person does at a particular level of a skill. It sits behind the number, turning an abstract level into something you can watch in real work.

Behaviour, not personality. "Is a strong communicator" describes a trait; "explains complex ideas and adapts message and tone for different audiences" describes behaviour you can observe and score. Good descriptors live in the world of actions.

Meaning for each level. "Level 3" alone is empty; "completes work to standard, unsupervised, with consistent quality" tells everyone what a 3 looks like. Descriptors are the shared language that stops each manager inventing their own scale.

Evidence test. Could you point to real work that proves this descriptor? If not, rewrite until you could show an auditor a sample, a sign-off, or an observed task.

Descriptors are not job descriptions. They describe proficiency on one skill at one level — not the whole role. Keep one central behaviour per level; put secondary behaviours in notes if needed.

What separates weak from strong wording?

× Weak: "Has good data analysis skills and is confident with numbers." — "Good" and "confident" are opinions; two assessors will read them differently; no evidence trail.

✓ Strong: "Independently analyses a dataset, draws correct conclusions, and presents them clearly to non-specialists without checking." — Every part is an observable action; any two assessors would judge the same output.

The pattern: strip adjectives that describe the person; replace with verbs that describe what they do; add conditions that mark the level — "without checking", "to regulatory standard", "for different audiences".

Transformation checklist: remove "excellent", "strong", "good at"; add a verb, a concrete object, and a condition that changes between levels (supervision, checking, scope).

What four ingredients does every descriptor need?

1. Action verb — analyses, writes, leads, trains, resolves.
2. Observable object — dataset, complaint, new starter, process.
3. Standard or condition — to standard, unsupervised, for varied audiences.
4. Evidence you could point to — document, sign-off, trained colleague.

Example with all four: "Resolves escalated complaints to the required quality standard, unsupervised" — verb (resolves), object (escalated complaints), standard (required quality, unsupervised), evidence (resolved cases on file).

Verb + object + standard almost writes itself once Level 3 is anchored. Run the evidence test last: if you could not show it to an auditor, rewrite.

How do you translate generic levels into skill-specific descriptors?

The Upleashed 0–5 framework supplies generic level meanings; your job is to translate them into concrete behaviour per skill. Level 3 is almost always the pivot: capable, unsupervised, consistent quality.

Worked example. Generic Level 3: "capable, works unsupervised, consistent quality." Skill: Workshop facilitation. Specific descriptor → "Designs and facilitates client workshops to agreed outcomes, without supervisor presence, with participant feedback at target satisfaction." Level 2 adds supervision or draft-only responsibility; Level 4 adds training new facilitators or hostile-stakeholder cases.

How do you write a full ladder for one skill?

Data analysis — one descriptor per level, each a visible step in independence and scope:

Notice every level leads with a verb; independence climbs from "under supervision" to "without checking" to "trains others" to "defines standards". The Level 2→3 boundary is the unsupervised line — the dispute most calibration sessions resolve.

Publish the ladder beside the matrix tab or in a linked sheet. New managers inherit the same meaning; disagreements improve the words, not the people.

What is the five-step writing process?

Calibrate with a colleague: score two real people against the draft ladder. If you split on a level, the descriptor for that boundary is unclear — rewrite the boundary, not the person.

What mistakes make descriptors unusable?

Describing traits, not behaviour. "Excellent", "strong", "good" — not scorable.

Duplicate levels. If Level 3 and 4 read the same, raters cluster arbitrarily.

Jargon without behaviour. "Understands API architecture" — what do they actually do?

Too long. One clear sentence beats a paragraph no one reads in a scoring session.

Never updated. When tools or regulation change, descriptors must change too.

Copy-paste from another industry. Import structure, rewrite behaviour for your work.

Never calibrated. Descriptors that read well can still be applied differently — test with real cases.

What if HR owns competencies but operations owns the matrix?

Edge case: central HR publishes enterprise competency libraries while team managers maintain a local matrix. Do not maintain two conflicting definitions for the same skill name. Either map HR competencies to matrix columns with a single local descriptor set, or rename columns to match operational tasks ("Salesforce case closure") rather than abstract competencies ("Customer focus").

Where union or works council agreement requires joint sign-off, draft descriptors in a working group with one operations representative and one HR partner per function. Publish a change log when descriptors update so raters know which version applied to each scoring cycle. Version numbers in the sheet footer prevent "which 3 is this?" disputes six months later.

How do descriptors connect to rating and calibration?

Descriptors are the reference for rating and calibration. Send them 48 hours before scoring; use disagreements to improve wording, not to argue about personalities. When a descriptor repeatedly produces splits, the words are wrong — not the people.

Pair with building the matrix and competency scale 0–5 explained for framework context.

For behavioural skills, stay concrete: "synthesises conflicting views into a clear recommendation by Friday" beats "is collaborative." For technical skills, name the task, system, and quality bar: "closes reconciliations in the ledger to month-end standard without supervisor sign-off." The same four ingredients apply; only the object changes.

Store descriptors in a tab linked from the matrix, not in cell comments nobody opens. Version the tab when you change regulation or tooling so historical scores stay interpretable.

How do you workshop descriptors with practitioners?

Run forty-five minutes per skill cluster — technical, tools, compliance, behavioural — not one day for sixty skills. Draft Level 2 and Level 3 first; most disputes live on the unsupervised boundary. Test each sentence: what would we show an auditor or a new manager as proof?

Capture edge cases in footnotes: "Routine cases unsupervised; complex regulatory exceptions escalated" keeps Level 3 honest without inflating everyone to Level 4.

How do minimum standards differ from development wording?

Minimum standards are floors — safe, legal, operational must-haves. Development descriptors describe stretch beyond the floor. Put aspirational language in the required-level row or a separate development column, not inside Level 3 where raters confuse "required" with "nice to have".

How do you pilot descriptors before go-live?

Three managers score three real people independently per skill. Spread wider than one level means rewrite. Same level with different evidence stories means levels still overlap — sharpen the boundary sentence. Publish only after independent raters converge.

How do AI drafts fit the workflow?

Generators accelerate first drafts; practitioners must validate observability and level progression. Never publish AI wording without a verb test, evidence test, and pilot with three real cases. The failure mode is polished sentences that sound professional but collapse in calibration because no one can point to proof.

Store approved descriptor sets in the same folder as the matrix with a version number in the filename so raters never debate which ladder applied to last quarter's scores.

Which site tools help you write descriptors?

• 0–5 descriptor generator
• Upleashed 0–5 methodology
• Resource: competency policy 0–5
• Skills & competency glossary
• Free 5×5 mini-matrix builder
• Skills audit checklist
• How to rate employee skills

How should you score cells on the 0–5 scale?

Descriptors give each level meaning; the scale below is the shared framework they hang on.

Capability percentages use Upleashed weightings (Level 1 = 25%, Level 2 = 50%, Level 3 = 75%, Levels 4–5 = 100%; Level 0 excluded). See competency scale 0–5 explained for the full framework.

How does this guide connect to the rest of the site?

Keep write-competency-descriptors.pdf for offline briefings. Online, you get searchable structure, tables, and pointers into the wider methodology.

If descriptors drift between managers, reset them against the methodology pillar and republish from the descriptor generator.

Publish descriptors beside the grid so new managers inherit the same meaning of each level, not their own interpretation.

Frequently asked questions

What is a competency descriptor?

A short statement of observable behaviour at a given proficiency level for a skill — the words that give each level meaning so assessors score consistently and can evidence their judgement.

What makes a descriptor good?

It describes observable, evidenceable behaviour, leads with an action verb, names a concrete object, and states the standard or independence that marks the level. It passes two tests: could two assessors score it the same way, and could you point to real work that proves it?

How many levels should you write descriptors for?

Best practice is distinct proficiency levels per competency — typically Levels 1–5 on the Upleashed framework, with Level 0 marking not required. Each level should be a clear step up from the last.

How do you avoid vague descriptors?

Apply the verb test and the evidence test. Strip trait words like "good", "strong", and "confident"; replace with what the person actually does. Rewrite until you could point to real work proving each level.

Should descriptors differ by role?

Core level meanings stay consistent; objects and standards may differ by role. Use Level 0 where a skill is not required for that role.

Do you need software to write descriptors?

No. A document or spreadsheet tab works. The descriptor generator speeds drafting; the discipline is observable behaviour, not the tool.