AI Competency Matrix: How to Build an Employee Skills Map

A competency matrix is a table that lists the required skills and proficiency level for each role, and the verified proficiency level for each employee. AI speeds up building it in two places: it extracts skills from job descriptions, tasks, reports, and training records, and suggests an employee’s level with a link to the evidence. It should not make decisions for a person: under Article 86 of the Labor Code, an employer may not base decisions that affect an employee’s interests solely on data obtained through automated processing. The working model is a skills dictionary, a level scale with examples, an AI suggestion with evidence, confirmation by the employee and manager, and regular updates.

Prepared by AIrassvet using open sources, verified October 12, 2026. It is intended for business owners, HR directors, learning and development leaders, and frontline managers who want to understand what skills the team has and what it lacks. This is a practical guide: it includes a step-by-step plan, a matrix template, an AI prompt template, and a quality-check method. We did not test specific HR systems; the news about SAP’s deal is cited from the company’s press release, and the skills data comes from a World Economic Forum report. The labor law section is not legal advice. AIrassvet is not affiliated with the companies mentioned; the section at the end of the article describes the publisher’s services.

In Brief: How to Build a Competency Matrix with AI
Why This Topic Matters Now
Competency Matrix, Skills Map, and Professional Standard: What’s the Difference?
What AI Does Well — and What It Doesn’t
Legal Basics
Step-by-Step Build
Competency Matrix Template
AI Prompt Template for Extracting Skills
How to Check Whether AI Is Assessing Correctly
What to Do with the Finished Matrix
When a Competency Matrix Is Not Needed
Common Mistakes
Frequently Asked Questions
Primary Sources and Verification Method

In Brief: How to Build a Competency Matrix with AI

Start with 1–2 roles where skill gaps cost money: sales, customer service, manufacturing, development.
Create a skills dictionary with 8–15 items for the role, based on the professional standard, the job description, and the experience of top performers.
Define the level scale using examples of observable behavior, not adjectives like “good” and “excellent.”
Give AI the sources of evidence — tasks, reports, certificates, training results — and require a link to the evidence for each suggested level.
People confirm the level: the employee and the manager. AI suggests, but does not decide.
Check agreement between AI and managers across 20–30 employees before scaling up.
Update the matrix based on events — completed training, a new project — and review it in full at least once a year.

Why This Topic Matters Now

On October 6, 2026, SAP announced an agreement to acquire the Belgian company TechWolf , the developer of a “work intelligence” platform (SAP press release). According to SAP, TechWolf builds a “work context graph”: a model of which tasks are performed in a company, what skills people have, and what is happening in the external labor market. The data comes from existing HR and business systems. SAP plans to make TechWolf the “intelligent core” of the SuccessFactors HR suite—for skills mapping, workforce planning, and organizational redesign. Closing is expected in the fourth quarter of 2026; the deal value was not disclosed.

An important detail from the press release: SAP Autonomous Suite President Manos Swaminathan calls the skills graph “the grounding layer” for AI agent requests — that is, a factual source that agents rely on when answering questions about people and work. In other words, SAP sees a skills map not only as an HR report, but also as data that corporate AI agents should use as a basis for their answers.

The second reason is the pace of change. According to the World Economic Forum’s Future of Jobs 2025 report, employers expect that 39% of workers’ core skills will change by 2030; in 2023, that figure was 44%. The report is based on a survey of more than 1,000 employers representing over 14 million workers across 55 economies. AI and big data skills top the list of the fastest-growing skills in this report. A matrix created once and filed away becomes outdated faster than it can be put to use at this pace.

Competency Matrix, Skills Map, and Professional Standard: What’s the Difference?

Term	What It Is	What It’s For
Competency matrix	a “role × skill × required level” and “employee × skill × verified level” table	to see gaps, plan training, and assign people to projects
Skills map (graph)	links between tasks, skills, roles, and people, updated from work systems	workforce planning, AI agent responses about people and work
Professional standard	a government-approved description of job functions and qualification requirements	a basis for qualification requirements; in some cases, mandatory

Competency in this article, a skill or knowledge area that can be observed in work and verified through evidence: completed tasks, results, certificates, or peer feedback. “Communication skills” without a description of behavior are not a competency in this sense—you cannot verify or develop them in a targeted way.

What AI Does Well — and What It Doesn’t

Well:

extracts skills from texts: job descriptions, postings, task descriptions, project reports, and training programs;
normalizes wording: “1C setup,” “1C configuration customization,” and “1C programming” are reduced to a single dictionary item;
suggests a level with evidence: “Level 2 — independently led three implementations, see tasks No. …”;
finds gaps between the role requirements and the team’s verified levels;
matches training and mentors to the gaps it finds.

Poorly or risky:

assessing level based on indirect signs. The number of completed tasks shows workload, not mastery;
assessing “soft” skills from email or chat correspondence. This is both inaccurate and legally sensitive;
working without a glossary. Without a defined skills list, the model invents its own wording, and the matrix turns into a word cloud;
making decisions about people. Promotions, bonuses, transfers, or terminations based on AI evaluation without human involvement are a direct path to mistakes and disputes.

For more on the strengths and weaknesses of AI in people operations, see the article on AI in HR and Recruiting.

Legal essentials

Below are three rules most often affected by a competency matrix. For your situation, check the details with a lawyer.

1. Decisions about an employee cannot rely solely on automated processing. Clause 6 of Part 1 of Article 86 of the Russian Labor Code prohibits an employer from basing decisions affecting an employee’s interests on personal data obtained exclusively through automated processing or electronic collection. The practical takeaway: the level suggested by AI must be confirmed by a person, and the confirmation process itself must be documented.

2. Professional standards are not always mandatory. Under Article 195.3 of the Russian Labor Code, if qualification requirements for a job function are established by the Labor Code, a federal law, or another regulatory act, the professional standard is mandatory in that part. In other cases, the employer uses it as a basis for requirements, taking into account its own technologies and work organization. For the matrix, this means: in regulated roles, take the mandatory requirements from the professional standard without changes; in other roles, adapt them.

3. Employee assessments are personal data. Skill levels, reviews, and performance appraisal results relate to a specific person. Define the purpose of processing, restrict access to the matrix, and do not send this data to external AI services without a risk assessment. Options include anonymization before sending or using the model within your own environment; for more, see the article on whether data can be sent to AI.

Step-by-step setup

Define the goal. “Understand who needs training on the new CRM,” “reduce dependence on two key engineers,” “assemble a team for the project.” The goal determines the roles, skills, and depth of assessment.
Choose 1–2 roles for the pilot. A role with 10–50 employees and clear performance outcomes is better.
Build a skills glossary. Sources: the professional standard, job description, job postings, and interviews with 2–3 top employees and their managers. AI helps consolidate wording and remove duplicates, but the final list is approved by the department head. A good target is 8–15 skills per role.
Describe the level scale. A simple four-level scale: 0 — no experience; 1 — performs under supervision; 2 — works independently; 3 — trains others and improves the process. For each skill, add 1–2 examples of observable behavior at each level.
Define the evidence sources for each skill: tasks in the tracker, deals in CRM, reports, certificates, course results, mentor feedback. Note which sources contain personal data and where they can be processed.
Start extracting and suggesting levels. AI receives the glossary, scale, and evidence for one employee and returns a suggested level for each skill, with a link to the evidence or a note saying “no evidence.”
Run the validation. The employee sees the suggestion and can add evidence; the manager confirms or changes the level. A difference of more than one level is a reason for discussion, not for an automatic decision.
Build a gap map. Compare the role’s required levels with the confirmed ones: which skills exist only in one person, which are missing from the team, and who is ready for the next role.
Build in updates. A completed course, a new project, or a certificate is a reason for a targeted update; a full review is done once a year or when the role changes.

Competency matrix template

Example for the role “B2B Sales Manager.” The skills and levels are illustrative—replace them with your own.

Skill	Required level	Level 1: under supervision	Level 2: independently	Level 3: trains others	Evidence source
Needs assessment	2	conducts the meeting using the script	asks clarifying questions, logs the need in CRM	reviews colleagues’ calls, refines the script	meeting recordings, deal records
Preparing a sales proposal	2	fills in the template	builds a proposal for the client’s needs	creates templates for the team	proposal in CRM
Handling objections	2	uses ready-made responses	responds based on the situation	adds to the objections database	call reviews
Managing a deal in CRM	2	enters data after a reminder	keeps all stages on schedule	sets up team reports	change history in CRM
Product knowledge	3	answers standard questions	advises on non-standard issues	trains new hires	tests, support tickets
Price and terms negotiations	2	agrees on terms with the manager	negotiates within assigned authority	develops strategy for major deals	meeting minutes, approvals
Using the department's AI tools	1	uses them according to instructions	selects prompts for their own tasks	shares workflows with colleagues	manager feedback

For the employee, the row is supplemented with the columns “proposed by AI,” “evidence,” “confirmed by manager,” and “confirmation date.” This structure immediately shows what each level is based on.

AI Prompt Template for Skill Extraction

The template below is a starting point for a pilot. It should be run on anonymized data or in a model deployed within the company’s secure environment.

You are helping create a competency matrix for the role of “{role}”.
Skill dictionary (use only these items): {list of skills}.
Scale: 0 — no experience; 1 — under supervision; 2 — independently; 3 — trains others.
Level descriptions: {behavior examples by skill}.

Evidence for employee {employee ID}: {tasks, reports, certificates, courses}.

For each skill in the dictionary, return:
- the proposed level or “no data”;
- one or two pieces of evidence with the source ID;
- confidence: high / medium / low, and why.
Do not draw conclusions about personal qualities and do not use information outside the listed evidence.
If there is not enough evidence, write “no data” instead of guessing the level.

The three rules at the end of the prompt are the key part. They turn AI from an “evaluator” into an assistant that gathers evidence and honestly shows where it is missing.

How to Check Whether AI Is Evaluating Correctly

This is an editorial method from AIrassvet, not the result of our own testing.

Select 20–30 employees in the same role with different levels of experience.
Ask two managers to independently rate them on the same scale without seeing the AI’s suggestions.
Compare the ratings for each skill: the share of exact matches and the share of differences greater than one level.
Analyze the differences. Most often, the reason is an unclear level description or a skill for which there is no evidence in the systems. Fix the dictionary and the scale, not just the prompt.
Set the acceptance threshold in advance. For example, AI suggestions are used in work if agreement with managers for each skill is no worse than the agreement between managers themselves.
Repeat the check after changing the dictionary, scale, model, or sources.

What to Do with the Finished Matrix

Plan training: courses and mentoring for the skills with the largest gaps, not “everything for everyone.” More details in the article on AI for employee onboarding and training.
Reduce single-person dependency risk: a level 3 skill held by only one employee is a reason to transfer knowledge and document procedures in the corporate knowledge base.
Build project teams based on skill combinations, not job titles.
Prepare a talent pipeline: who is closest to the requirements of the next role.
Refine hiring requirements: which skills the team is missing, rather than “what the previous employee had.”
Give AI agents context. An agent that assigns tasks or looks for an expert inside the company responds more accurately if it relies on a verified skills map. This is the same logic SAP describes for TechWolf.
Connect it to performance reviews —carefully and separately from them; see the material on a fair feedback and development system.

When a competency matrix is not needed

The team has fewer than 5–7 people, and the manager knows each person’s strengths without a table.
Roles change every few months, and the skills dictionary will not have time to stabilize—better to start with a map of key knowledge and procedures.
There are no decisions that the matrix will support: no training, no project staffing, no talent pipeline. Then it is reporting for the sake of reporting.

Common Mistakes

Rolling it out to the entire company right away. Without a tested dictionary and scale in one role, scaling multiplies mistakes.
A scale made of adjectives. “Knows it well” cannot be verified; you need behavior examples.
Assessment without evidence. If a level has no link to a task, document, or result, it is an opinion, not data.
Automatic decisions based on AI ratings. This conflicts with Article 86 of the Russian Labor Code and undermines trust in the system.
Personal data sent to an external service without risk assessment. Employee ratings are sensitive data.
A matrix that is not updated. Given the pace of change described in the WEF report, even one year is already significant.
Hidden evaluation. If employees cannot see their levels and evidence, they cannot challenge or improve them.

Frequently Asked Questions

What is a competency matrix?

A table that lists the required skills and target level for each role, and each employee’s verified level for those skills. It shows gaps between requirements and the team’s actual capabilities and helps plan training, project staffing, and the talent pipeline.

How does AI help create a competency matrix?

AI extracts skills from job descriptions, tasks, reports, and training records, consolidates different wording into a single dictionary, and proposes an employee’s level with evidence. The employee and manager confirm the level.

Can employees be evaluated only with AI?

No. Clause 6 of Part 1 of Article 86 of the Russian Labor Code prohibits employers from basing decisions that affect an employee’s interests on data obtained exclusively through automated processing. AI-based evaluation should be used as a suggestion that a person reviews.

Is it mandatory to use professional standards in a competency matrix?

If qualification requirements for a position are established by law or another regulatory act, the professional standard is mandatory in that part—this is stated in Article 195.3 of the Russian Labor Code. In other cases, it serves as a foundation that the employer adapts to its own technology and work organization.

How many skills should be included in a matrix for one role?

A practical rule of thumb is 8–15 skills per role. Fewer, and the matrix will miss gaps; more, and it becomes hard to maintain and validate.

How often should a skills matrix be updated?

Update it selectively when events happen: completed training, a new project, or a certification. Do a full review at least once a year or whenever the role changes. According to the World Economic Forum, employers expect 39% of workers’ core skills to change by 2030.

Primary sources and validation method

We reviewed the primary sources on October 12, 2026. The information about SAP’s deal with TechWolf comes from SAP’s press release; the data on the pace of skill change comes from the WEF report; the legal rules come from the text of the Labor Code as amended on 05/25/2026. The step-by-step plan, skills matrix template, request template, and verification method are the AIrassvet editorial methodology, not the result of testing any specific HR systems.

SAP, 10/06/2026: SAP to Acquire TechWolf, Giving Enterprises Evidence-Based View of Work in the Age of AI — work context graph, plans for SuccessFactors and Joule, deal timeline.
World Economic Forum, January 2025: The Future of Jobs Report 2025 — 39% of core skills will change by 2030, based on the survey.
Russian Labor Code: Article 86 — prohibition on decisions based solely on automated processing of personal data.
Russian Labor Code: Article 195.3 — procedure for applying professional standards.

Limitations: the WEF forecast is based on employer expectations, not skill measurement, and does not isolate the Russian market; the description of TechWolf is taken from SAP’s wording; the legal section is a general overview, not advice for a specific situation.

How AIrassvet helps build a skills map with AI

If you want a skills matrix built from work data rather than annual questionnaires, while still keeping people in control, the solution should be built around your systems and personal data handling rules. On the AIrassvet website the sections directly relevant to this task are described:

RAG systems and enterprise knowledge bases: a single skills vocabulary and evidence search across tasks, reports, and learning materials.
Enterprise AI workspaces and model integration into business processes, including the skills assessment workflow.
Deploying language models on company servers when employee data cannot be sent to an external service.
Process audit, data preparation, MVP development, testing, launch, and team training.

The first step is to choose one role, describe how skills are currently assessed, where evidence is stored, what personal data restrictions apply, and by what criterion you consider the AI’s suggestions accurate enough.

Discuss the project.

Bottom line: AI gathers the evidence; people make the decision

Skills matrices are back in focus: SAP is buying TechWolf so that skills mapping can become the foundation for HR decisions and enterprise AI agents, while employers expect nearly 40% of core skills to change by 2030. AI significantly speeds up matrix building by extracting skills from work data, standardizing wording into a common vocabulary, and suggesting levels with evidence. But the decision on level and everything that follows remains with humans: that is required both by common sense and by Article 86 of the Labor Code.

Start with one role: a vocabulary of 8–15 skills, a scale with behavior examples, AI suggestions linked to evidence, employee and manager confirmation, and a consistency check across 20–30 employees. If AI aligns with managers at least as well as they align with each other, expand the matrix to adjacent roles.