Skip to main content
All reports

Report 10 · Builder series

Skills Taxonomy from Scratch

Which public frameworks to borrow from, the granularity rule that saves you from a 2,000-row monster, a five-step build that fits in six weeks, and the maintenance rhythm that keeps it alive.

AudienceHR, L&D and People analytics leaders
Reading time13 min read
Length8 sections, print-ready
PublishedOmie · tryomie.com
Executive summary

The short version

A shared skills vocabulary is a selection job, not a two-year consulting project.

Skills-based organization is the most-used phrase in HR strategy and the least-implemented, because step one, agreeing what the skills actually are, looks like a two-year consulting project. It is not. Public frameworks (ESCO, O*NET, Lightcast) have already done the naming for free, so your job is selection and translation, not invention.

A mid-size organization needs 60–120 skills, not 13,000. Built in six weeks, refreshed quarterly, and wired to something people feel: learning paths, internal moves, honest 1:1s. The payoff is measurable, because skills-based organizations are far more likely to retain their best people.

Evidence: 1 ESCO, O*NET, SFIA and Lightcast (2026) / 2 Deloitte (2024)

13,900Skills in ESCO

The EU's free taxonomy, a menu to select from, never to adopt whole.

1 ESCO, O*NET, SFIA and Lightcast, 2026
60–120Skills you need

The working size for a 50–500-person org. Bigger maps die of maintenance.

4 Omie, 2026
98%Top-performer retention

How much more likely skills-based orgs are to keep their best people.

2 Deloitte, 2024
39%Skills churn by 2030

Core skills changing, the reason the map must be a living document.

3 World Economic Forum, IBM and BCG, 2025

What's inside

Borrow before you build: the four public frameworks worth quarrying.
The granularity rule that keeps you off a 2,000-row monster.
A five-step build that fits in six weeks with no consultants.
The 0–4 level scale to steal verbatim, plus the calibration ritual.
A quarterly heartbeat and an eight-question self-audit with a scoring key.
Decision

The question is not what should our people learn this year. It is which skills can we name, assess, and act on.

Section 01

Borrow before you build

The free shelf.

Four public frameworks cover almost every skill your org chart contains. Treat them as quarries, not blueprints. You extract names, definitions and adjacencies, then throw away 99% of the volume.

Evidence: 1 ESCO, O*NET, SFIA and Lightcast (2026)

FrameworkWhat it isUse it forWatch out
ESCO (EU)About 13,900 skills and 3,000+ occupations, 28 languages, free.Canonical skill names and multilingual labels, ideal for European orgs.Bureaucratic granularity, and some tech skills lag the market.
O*NET (US)About 900 occupations with skills, tasks, tools and importance weightings.Role to skill starter lists: look up the occupation, harvest the top-weighted skills.US-flavored role definitions, and slow update cycles.
SFIAThe IT and digital skills framework, 7 responsibility levels.Level language for technical roles, the best-written descriptors in the field.Tech only, and licensing terms for commercial use.
Lightcast Open Skills33,000+ skills extracted from live job postings, updated continuously.Market-fresh names for fast-moving tool skills, and a check that your labels match hiring language.Noisy, with tools and brands mixed in. Curate hard.

Counts as published by each maintainer, mid-2026.

Evidence: 1 ESCO, O*NET, SFIA and Lightcast (2026)

The selection stance

Start from your ten pivotal roles, look each up in O*NET and ESCO, harvest 8–15 candidate skills per role, then dedupe across roles. You land near 80 raw candidates before the pruning in the next section. Two afternoons of work, not two quarters.

Evidence: 1 ESCO, O*NET, SFIA and Lightcast (2026) / 4 Omie (2026)

Section 02

The granularity rule

Where taxonomies go to die.

Every failed taxonomy died the same death: wrong altitude. Too coarse (communication) and nothing is actionable. Too fine (VLOOKUP) and the map outdates monthly while maintenance eats the team. The working rule is one sentence: a skill belongs on the map if you can assess it, train it, and staff with it, all three, at the same altitude.

Evidence: 4 Omie (2026)

AltitudeExampleVerdict
Too coarseMarketing, Leadership, DataCategory, not skill. You cannot assess it or close it. Use these as folder names only.
RightPaid search campaign management, Giving corrective feedback, SQL querying and joinsAssessable in an hour, trainable in weeks, staffable on a project. This is map altitude.
Too fineGoogle Ads Editor v2.7, the Q3 escalation macroTool trivia. It lives in checklists and SOPs, not the taxonomy.

The assess, train, staff test, applied to every row.

Evidence: 4 Omie (2026)

Tag skills, do not tree them

A three-level hierarchy (category, skill, behavior) is the maximum a mid-size org can maintain, and most need only two. Deep trees look rigorous in the kickoff deck and unmaintainable by the second quarter. Technical skills halve in value in about 2.5 years, so every extra layer is future maintenance debt.

Evidence: 3 World Economic Forum, IBM and BCG (2025) / 4 Omie (2026)

Section 03

The five-step build

Six weeks, no consultants.

Everything in the data, and in the mid-size organizations we work with, reduces to the same six-week sequence. The trick is to prune hard and ship something visible before the map calcifies into a spreadsheet nobody opens.

Evidence: 4 Omie (2026)

Six weeks, start to baseline

1

Week 1: Pick the ten pivotal roles

Not every role, the ten where a skill gap costs the business most. Controversial by design: forcing the list forces the strategy conversation.

2

Weeks 1–2: Harvest candidates from the frameworks

Run O*NET and ESCO lookups per role, then add a one-hour interview with the best current performer: what do you actually do that others cannot? The interview catches the org-specific glue skills the frameworks miss.

3

Week 3: Prune with the granularity test

Every candidate faces assess, train, staff. Merge synonyms ruthlessly (stakeholder management versus stakeholder communication: pick one, alias the other). Target 60–120 rows. Over 150 and you are building a monument, not a tool.

4

Week 4: Write the 0–4 level scale once

One generic scale, applied everywhere. Per-skill custom scales feel precise and guarantee inconsistency. Borrow SFIA language for technical rows if licensing allows.

5

Weeks 5–6: Baseline three teams and ship something visible

Self-rating plus manager rating on the relevant subset (10–20 skills per person, never all 100). Then wire it to something felt at once: a learning path, an internal posting, a 1:1 agenda.

Evidence: 4 Omie (2026)

Section 04

The level scale

Steal it verbatim.

One scale, five levels, applied to every row. The operational definition is the sentence that settles disputes in a calibration session, so write behaviors, not adjectives.

Evidence: 4 Omie (2026)

LevelNameOperational definition
0NoneHas not done it. No evidence. A real level, because maps with no zeros are lying somewhere.
1AwareCan follow along and execute with instructions or close supervision. Knows the vocabulary.
2PracticingExecutes standard cases independently and escalates the odd ones. The workhorse level.
3FluentHandles novel and messy cases. Others bring them their problems. Can teach it.
4ShapingSets how the org does it: builds the playbook, judges the exceptions, raises the bar.

One generic 0–4 scale, applied everywhere, never a custom scale per skill.

Evidence: 4 Omie (2026)

Calibration ritual, 30 minutes a quarter

Each manager brings one contested rating to a peer session, and the group argues it against the definitions above. Two sessions in, ratings converge. Without it, a 3 in sales and a 3 in engineering are different currencies.

Evidence: 4 Omie (2026)

Section 05

Keeping it alive

The quarterly heartbeat.

A taxonomy is a living document or a dead one. There is no shelf-stable version, because 39% of core skills change by 2030 and your technical rows churn fastest. Done right, the maintenance load is about two hours a quarter.

Evidence: 3 World Economic Forum, IBM and BCG (2025)

The quarterly heartbeat, about two hours

Add and retire pass: new skills entering the ten pivotal roles, and rows nobody has touched in two quarters. Retire means archive, not delete, because history matters for trends.
Single-holder review: every critical skill at level 3 or above held by exactly one person is a risk-register entry with a named backup plan.
Language check: do your skill names still match job-posting language? Lightcast is the free sanity check, and drift here quietly breaks hiring alignment.
Calibration session: the 30-minute ritual from the level scale.

Evidence: 1 ESCO, O*NET, SFIA and Lightcast (2026) / 4 Omie (2026)

The failure mode to avoid

A taxonomy nobody experiences is a spreadsheet with ambitions. If no learning path, posting, or 1:1 references it inside a month, it will quietly die no matter how elegant the rows are.

Evidence: 4 Omie (2026)

Section 06

Make it felt

The map earns its existence by being used.

The taxonomy is not the deliverable. The felt experiences it powers are: a learning path pointed at a real gap, an internal posting written in skill language, a 1:1 that talks about growth in the same words everywhere. Wire the map to one of these in week six, before anyone can call it a spreadsheet.

Evidence: 4 Omie (2026)

98%Retain top performers

How much more likely skills-based orgs are to keep their best people.

2 Deloitte, 2024
66%More likely to stay

Workers who prefer orgs that value skills over degrees.

2 Deloitte, 2024

Wire it to one of these first

A learning path: point daily practice at the gaps the baseline just revealed.
An internal posting: describe the role in skills, and let people see what their next move needs.
A 1:1 agenda: growth conversations in one shared vocabulary, not per-manager folklore.
A risk list: the map first deliverable is the single-holder and pivotal-gap register.

Evidence: 4 Omie (2026)

Decision

A taxonomy earns existence by being experienced, not admired.

Self-audit

Is your vocabulary shared?

Eight questions, then the one-page version.

Check every statement that is true today.

A written skills list exists: one source of truth, not per-manager folklore.
It has 60–150 rows, not 12 categories and not 800 micro-items.
Every row passes assess, train, staff, at the same altitude.
One 0–4 scale applies everywhere, with operational definitions, not adjectives.
Ten pivotal roles map to skill subsets: 10–20 per role, harvested not guessed.
Three or more teams have a current baseline: self and manager rating, younger than 6 months.
The map is wired to something felt: learning paths, postings, or 1:1s reference it weekly.
A quarterly maintenance slot exists, on a calendar, with an owner.

Evidence: 4 Omie (2026)

ScoreBandWhat it means
0–2FolkloreSkills live in managers' heads. Start with the ten pivotal roles this week.
3–4Spreadsheet eraA list exists and nothing runs on it. Wire it to one felt experience.
5–6Working mapThe vocabulary is shared. Add the calibration ritual and the heartbeat.
7–8Operating systemSkills are how work, growth and hiring get decided here. Guard the pruning discipline.

Evidence: 4 Omie (2026)

Takeaways, the one-page version

Selection, not invention: ESCO, O*NET and Lightcast name the skills for free, so harvest 8–15 per pivotal role and translate to house language.
60–120 rows is the working size: below that you cannot act, above about 150 you cannot maintain.
Assess, train, staff is the altitude test: fail any one and the row is a category or a checklist item, not a skill.
One 0–4 scale everywhere, plus a quarterly calibration ritual, beats per-skill custom scales every time.
Ship something felt in week six, and keep it alive with about two hours a quarter or it dies.

Evidence: 1 ESCO, O*NET, SFIA and Lightcast (2026) / 4 Omie (2026)

Decision

This week: name the ten pivotal roles and run O*NET or ESCO lookups for three of them. This quarter: prune to 60–120 rows, baseline one team, and wire the map to one felt experience.

Appendix

Sources and method

Every external numeric claim in this report points to one of these 2024 to 2026 sources. Forecasts and self-reported surveys are labelled so they are not mistaken for causal proof.

1
Public skills frameworks
ESCO, O*NET, SFIA and Lightcast · 2026

ESCO (European Commission): about 13,900 skills and competences, 3,000+ occupations, 28 languages, free. O*NET (US Department of Labor): about 900 occupations with weighted descriptors. SFIA: 7-level digital skills framework. Lightcast Open Skills: 33,000+ skills from live postings. Counts as published by each maintainer, mid-2026.

esco.ec.europa.eu
2
The skills-based organization research
Deloitte · 2024

Skills-based organizations are 98% more likely to retain high performers, and 66% of workers are more likely to stay at organizations that value skills over degrees.

www2.deloitte.com
3
Future of Jobs Report 2025; skill half-life research
World Economic Forum, IBM and BCG · 2025

39% of workers core skills are expected to change by 2030 (down from 44% in 2023). Average professional skill half-life is now under 5 years, and about 2.5 years for technical skills (IBM, with BCG concurring). Directional estimates.

weforum.org
4
Method note
Omie · 2026

The granularity test, the 0–4 level scale, the five-step build, the size guidance of 60–120 rows, and the quarterly maintenance heartbeat are Omie editorial framework from deployments in 50–500-person organizations.

tryomie.com