Introducing the Clu Atomic Skill Taxonomy™

Clu Labs research release, Jan 2023

Clu Labs today announces the Clu Atomic Skill Taxonomy™, a new library of more than 6,000 skills and attributes, each describing exactly one capability. This taxonomy underpins the Clu platform and the algorithms we are building.

The research question

Clu Labs set out to build a method that could compare any two skills, link skills to tasks, and model how work changes under automation, applying the same rules to every skill. For that method to give reliable results, the units it works on had to be consistent.

The working hypothesis was that a uniform method needs atomic units. If one taxonomy entry bundles two capabilities, any score, comparison, or link built on it becomes an average of two different things. That error then carries into every task and role that uses the skill.

What makes a skill atomic?

In the Clu Atomic Skill Taxonomy, a skill is atomic when it describes a single capability that carries one consistent fingerprint. “Stakeholder management and reporting” fails this test. It combines two capabilities that rely on different kinds of thinking, serve different purposes and can be automated to very different degrees. In the taxonomy, it becomes two separate skills.

The taxonomy also separates skills from attributes. A skill describes how well someone can do something. An attribute, such as a certification, a physical requirement or a specific software tool, decides whether someone can do the work at all. Skills fall into Technical, Functional, Behavioural, or Knowledge categories. Attributes sit alongside them but are scored differently.

Testing existing taxonomies

Before building, Clu Labs tested whether an existing taxonomy could serve as the foundation. Clu Labs checked a sample of 500 skills drawn from widely used public taxonomies, including O*NET and Lightcast Open Skills, against the atomicity test.

• 38% combined two or more capabilities in a single entry.

• 21% listed tools, products, or qualifications as skills.

• 17% were broad enough to cover several distinct capabilities under one name.

These taxonomies describe occupations and labour markets, and they do that well. They were not designed to support a single method applied uniformly at the skill level; they were not designed to serve as a data set for a suite of standardised calculations, and none of them was built to maintain atomicity standards throughout.

How our taxonomy was built

Clu Labs built the taxonomy from scratch with a team of organisational and behavioural psychologists, one job type at a time, sector by sector and level by level. For each job type, the team defined the tasks involved, broke those tasks into the enabling atomic skills, fingerprinted each skill, and checked it against every existing entry before adding it.

The work took eighteen months. Building on a licensed public list would have been faster, but we chose to build it ourselves because any error in the base units would repeat across every result the platform produces, and we weren't willing to take that risk.

Validation

We audited the taxonomy in six full passes and continue to iterate and refine it via our foundational algorithms. Each initial pass checked for duplicate entries, skills placed in the wrong type or category, inconsistent naming and entries that failed the atomicity test.

• Entries failing the atomicity test fell from 32% in the first audit to under 1% in the latest.

• Independent psychologists assigned the same categories and types as the taxonomy in 94% of over 1,000 sampled skills.

• Tested against 500 roles, skills derived from the task layer matched expert-built profiles in 91% of cases.

Every taxonomy change is versioned, so you can reproduce any past result from the version it was produced on.

What the taxonomy makes possible

Because every skill is atomic and described the same way, Clu can:

• measure how close two skills are even when their names share no words

• derive around 90% of the skills on a role from the tasks the role performs

• score how much of each skill AI, automation, or robotics could do

• model which roles hold each piece of critical work

All of this runs deterministically. The same input always produces the same output, and you can trace every result back to the individual skills it's built from.

The Clu Atomic Skill Taxonomy maps to every major taxonomy, so clients can easily connect Clu’s analysis to their existing systems and labour market data.

Frequently asked questions

What is the Clu Atomic Skill Taxonomy? A library of more than 6,000 skills and attributes built by Clu Labs, where each skill describes one capability only.

Why didn’t Clu use O*NET or Lightcast? Clu Labs tested both. Neither was built to support a single method applied uniformly at skill level, and neither met the atomicity standard throughout.

Is the taxonomy available to clients? It is built into the Clu platform via subscription and via an API for developers.

To see how the Clu Atomic Skill Taxonomy maps to your organisation, you can contact Clu's research team directly at labs@getaclu.io.

About Clu

Clu provides work intelligence for regulated organisations. It helps leadership teams identify and reduce the risk of their biggest workforce decisions, including restructuring, AI transformation, M&A and changes to capability models, using the structural data they already hold. Clu Labs is Clu's research arm.