TL;DR. “What should I learn next?” is probably the most personal question people now type into an AI assistant, and the answers that come back are usually a generic list of trending skills. A list cannot answer a personal question. What you need is a scoring method that combines public market data with the facts only you know. This article gives you one: the CEOtudent 4-Factor Skill Decision Matrix. It scores any candidate skill from 1 to 5 on Demand (is the market paying for it), Durability (will it hold value as AI changes the work), Leverage (does it compound with what you already have) and Cost-to-competence (how much time and money it takes to become useful). The weights are 30/25/25/20, and the matrix adds veto rules and tie-breakers. The anchors come from the World Economic Forum’s Future of Jobs Report 2025, which expects 39% of workers’ core skills to be transformed or outdated over 2025-2030, and from the US Bureau of Labor Statistics 2025-35 projections. We also joined two BLS datasets and found that occupations with “Low” relative AI exposure have the highest share projected to shrink (36.6%), while 8 of the 30 fastest-growing occupations are in the “Very high” exposure group. The takeaway: exposure to AI is not the same thing as fragility, and your ranking should not treat it that way.
Why “what’s hot” is the wrong question
A CEO does not fund a project because it is popular. Funding goes where the expected return, the time the advantage lasts, the fit with existing assets and the cost all line up. Your learning hours deserve the same discipline, because they are scarce and every hour spent on one skill is an hour not spent on another.
The case for discipline is in the data. Employers surveyed for the WEF Future of Jobs Report 2025, over 1,000 companies representing more than 14 million workers across 55 economies, expect that on average 39% of workers’ existing skill sets will be transformed or become outdated between 2025 and 2030. The report calls this “skill instability”. It is down from 44% in the 2023 edition and 57% in 2020, and the report suggests one likely reason: 50% of workers have now completed training, reskilling or upskilling measures, up from 41% in the 2023 edition. Put simply, the people who keep learning are the ones slowing the churn.
Trend lists fail for three reasons:
- They ignore your starting point. The same skill can pay off fast for one person and very slowly for another, depending on what it connects to. That idea is the core of skill stacking.
- They ignore decay. Some skills keep their value for a decade, and others lose most of it by the next software release (see the half-life of skills).
- They ignore cost. A skill that takes two years to learn can be the right move, but only if you have picked it on purpose.
The matrix below handles all three. It is built to be applied in about thirty minutes and re-run every quarter.
What the market data actually says
Before any scoring, it helps to see the two public datasets that anchor the Demand and Durability factors.
Signal 1: which skill categories employers expect to grow
The WEF ranks 26 skill categories by net increase: the share of employers who expect a skill to grow in use, minus the share who expect it to decline. Technology skills lead the ranking, but human skills such as creative thinking, resilience and curiosity are close behind.
Table 1. Skills on the rise, 2025-2030 (verified data, WEF Future of Jobs Report 2025, Figure 3.4)
| Rank | Skill category | Net increase (percentage points) |
|---|---|---|
| 1 | AI and big data | 87 |
| 2 | Networks and cybersecurity | 70 |
| 3 | Technological literacy | 68 |
| 4 | Creative thinking | 66 |
| 5 | Resilience, flexibility and agility | 66 |
| 6 | Curiosity and lifelong learning | 61 |
| 7 | Leadership and social influence | 58 |
| 8 | Talent management | 58 |
| 9 | Analytical thinking | 55 |
| 10 | Environmental stewardship | 53 |
| 11 | Systems thinking | 51 |
| 14 | Design and user experience | 45 |
| 17 | Programming | 27 |
| 18 | Marketing and media | 25 |
| 19 | Resource management and operations | 24 |
| 25 | Reading, writing and mathematics | -4 |
| 26 | Manual dexterity, endurance and precision | -24 |
Source: World Economic Forum, Future of Jobs Survey 2024, as published in the Future of Jobs Report 2025. Rows 12-13, 15-16 and 20-24 omitted for space.
Two details matter for a personal decision. First, programming (27) scores well below AI and big data (87). The report measures skill categories, and employers clearly separate “working with data and AI” from “writing code”. Second, reading, writing and mathematics sit slightly negative, and the report describes them as among the most stable skills. These are foundations, not candidates for a next-skill bet. They show up in almost every job, which is why the net change is close to zero.
Signal 2: what “AI exposure” does and does not predict
Alongside its 2025-35 projections, the BLS published a supplementary file that places all 831 detailed occupations into four relative AI exposure categories (Low, Moderate, High, Very high). The categories are built from five sources: three theoretical measures and two based on observed usage, including Claude and Microsoft Copilot data. BLS is explicit about the limits. An exposure category “is not a forecast of employment growth or decline”, it does not separate automation from augmentation, and the theoretical sources describe AI capabilities no later than mid-2023.
We joined that file to the BLS table of the 30 fastest-growing occupations and summarised each exposure group.
Table 2. CEOtudent calculation: AI exposure vs projected growth, 831 US occupations (from BLS data)
| Relative AI exposure | Occupations | Employment 2025 (thousands) | Share of employment | Projected employment change 2025-35 | Median of occupation median wages | Share of occupations projected to shrink | Among the 30 fastest-growing |
|---|---|---|---|---|---|---|---|
| Low | 213 | 30,264.7 | 17.8% | +2.5% | $49,120 | 36.6% | 4 |
| Moderate | 206 | 44,651.3 | 26.2% | +6.2% | $52,760 | 26.2% | 9 |
| High | 206 | 42,142.0 | 24.8% | +3.1% | $73,985 | 19.4% | 9 |
| Very high | 206 | 53,209.2 | 31.3% | +2.0% | $78,105 | 29.1% | 8 |
Calculated by CEOtudent from the BLS “AI exposure categories and 2025-35 employment projections” file and BLS Employment Projections Table 1.3. Employment change is the sum of detailed-occupation employment in 2035 divided by the sum in 2025. Wage column is the median across occupations (unweighted); 6 occupations without wage data excluded. For all occupations combined, BLS projects +3.5%.
The table argues against a popular shortcut: “learn whatever AI can’t touch.” The Low-exposure group has the largest share of occupations projected to shrink, and it also has the lowest wages. The Very-high group grows more slowly in total, yet it includes data scientists (+34.6%), information security analysts (+21.0%) and computer and information systems managers (+15.8%, median wage $175,140). All of these are among the fastest-growing occupations in the country.
Exposure tells you that a job’s tasks are changing. It does not tell you whether the market will keep paying for the job. That is why the matrix scores Durability separately from Demand, and why it scores Durability at the level of the skill, not the occupation.
The CEOtudent 4-Factor Skill Decision Matrix
The four factors fall into two groups. Demand and Durability describe the market, so they come from data and are roughly the same for everyone. Leverage and Cost-to-competence describe you, so they are judgments only you can make. A skill needs to do well on both groups to rank highly.
Table 3. CEOtudent editorial framework: scoring rubric (1-5 per factor)
| Score | Demand (30%): is the market paying? | Durability (25%): will it hold value? | Leverage (25%): does it compound with what you have? | Cost-to-competence (20%): how cheap is “useful”? |
|---|---|---|---|---|
| 5 | WEF net increase 65 or more, OR target occupation projected to grow at 3x the all-occupation rate (10.5% or more over 2025-35) | Principle-level skill in the judgment, direction or people layer; carries across tools and roles | Builds on two or more skills you already hold at working level and opens a new role or combination | Useful level in 1 month or less (the BLS threshold for short-term on-the-job training) |
| 4 | Net increase 50-64, OR growth 2x-3x the all-occupation rate (7.0-10.4%) | A method that AI augments rather than performs; tools change, the core does not | Builds directly on one strong existing skill | 1-3 months part-time |
| 3 | Net increase 25-49, OR growth at 1x-2x the rate (3.5-6.9%) | Durable core with a fast-churning tool layer that needs regular relearning | Neutral: neither builds on nor conflicts with your stack | 3-12 months (the BLS moderate-term band runs from more than 1 month to 12 months) |
| 2 | Net increase 0-24, OR growth between 0% and 3.4% | Tied to one platform or version; plan to relearn within 1-2 years | New domain with missing prerequisites you must learn first | More than 12 months (the BLS long-term band) |
| 1 | Negative net increase, OR target occupation projected to shrink | The skill is itself the routine task being automated (WEF lists data entry clerks among the fastest-declining roles) | No connection to your stack, and it competes for time with skills you must maintain | Requires a degree, licence or residency, plus significant money |
How to score Demand. Use whichever of the two signals you can verify for your target market, and take the higher one. The WEF category tells you what employers expect globally. BLS occupation growth, compared with the 3.5% all-occupation rate, tells you whether a specific US role is expanding. Outside the US, use your national statistics office’s projections in the same way. The most in-demand skills of 2026 covers more market signals.
How to score Leverage. Start from an honest inventory. The skill audit map walks through one. Leverage is the factor that ranks the same skill differently for two people.
How to score Cost. Estimate time to useful, not to mastery: the point where someone would pay for or depend on your output. Time-to-competence data for 20 skills gives reference points.
The formula
Skill Score = 0.30 x Demand + 0.25 x Durability + 0.25 x Leverage + 0.20 x Cost-to-competence
The maximum is 5.0. Multiply by 20 if you prefer a 0-100 scale. Demand carries the most weight because a skill nobody pays for is a hobby, which is fine but is a different decision. Cost carries the least because a high cost is often worth paying if the other three factors are strong.
Decision rules
A weighted score on its own can hide a fatal flaw. These rules come before the total:
- Veto: Demand of 2 or less. If the market signal is flat or negative, the skill is not a career bet. Learn it for its own sake, but do not count it as career capital.
- Veto: Durability of 1. Do not invest months in the task the market is automating away.
- Credential gate. A Cost score of 1 (degree or licence) is only justified with Demand 5 and a funded, dated plan.
- Tie-break within 0.2 points: pick the higher Leverage. Compounding skills produce visible results sooner, and early results keep you learning.
- Second tie-break: shorter time to a first public proof (a shipped project, a published analysis, a certification that employers ask for).
- Time-poor floor. If you raise the Cost weight above 30% because your hours are scarce, require Durability of 3 or more. Otherwise the matrix will steer you toward cheap skills that fade fast (the worked example below shows this happening).
- One core bet at a time. Fund one skill as your main learning investment and keep a small maintenance budget for the skills you already hold (see the skill decay maintenance schedule).
Worked example: same eight skills, two different people
To show the matrix in use, we scored eight candidate skills for two illustrative profiles:
- Profile A: a marketing specialist with strong writing and campaign experience and basic spreadsheet skills.
- Profile B: an operations coordinator who is strong in spreadsheets and process, with no marketing background.
Demand comes from Table 1 and BLS growth (higher signal wins). For example, operations and project management scores only 2 on its WEF category (net increase 24) but 5 on BLS, because logisticians are projected to grow 17.6%. Durability, Leverage and Cost are CEOtudent judgments made for illustration. Your own scores will differ, and they should.
Table 4. CEOtudent calculation: worked example, baseline weights 30/25/25/20
| Candidate skill | Demand (data) | Durability (judgment) | A: Leverage | A: Cost | A: Score (rank) | B: Leverage | B: Cost | B: Score (rank) |
|---|---|---|---|---|---|---|---|---|
| AI workflow automation | 5 (WEF 87) | 3 | 5 | 4 | 4.30 (1) | 4 | 4 | 4.05 (3) |
| Data analysis (SQL + Python) | 5 (WEF 87; data scientists +34.6%) | 3 | 4 | 3 | 3.85 (2) | 5 | 3 | 4.10 (2) |
| People leadership | 4 (WEF 58) | 5 | 3 | 2 | 3.60 (3) | 4 | 2 | 3.85 (4) |
| Operations and project management | 5 (WEF 24; logisticians +17.6%) | 4 | 2 | 3 | 3.60 (4) | 5 | 4 | 4.55 (1) |
| UX research | 3 (WEF 45) | 4 | 4 | 3 | 3.50 (5) | 1 | 3 | 2.75 (7) |
| Single ad-platform certification | 3 (WEF 25) | 2 | 4 | 5 | 3.40 (6) | 1 | 5 | 2.65 (8) |
| Cybersecurity fundamentals | 5 (WEF 70; info security analysts +21.0%) | 4 | 1 | 3 | 3.35 (7) | 2 | 3 | 3.60 (5) |
| Software programming | 3 (WEF 27) | 3 | 2 | 2 | 2.55 (8) | 3 | 2 | 2.80 (6) |
People leadership and operations tie at 3.60 for Profile A; Rule 4 (higher Leverage) puts leadership third. For Profile B, data analysis (4.10) and AI workflow automation (4.05) are within 0.2 points, and Rule 4 keeps data analysis ahead. No skill in this set triggers a veto. All scores recomputed by script.
Three results are worth noting:
- The same market gives different answers. Cybersecurity has one of the strongest Demand scores in the set, yet it ranks 7th for the marketer because it connects to nothing in that profile’s current work. Operations and project management ranks 4th for the marketer and 1st for the coordinator. Leverage is what turns a generic trend list into a personal decision.
- Durability separates the similar options. The ad-platform certification is cheap (Cost 5) and fits the marketer’s work (Leverage 4), but it is tied to one vendor’s interface. It still ranks only 6th.
- The highest-demand skill is not automatically the top pick. Data analysis and AI workflow automation share the maximum Demand score, but they swap places between the two profiles.
Sensitivity check: does the answer survive different weights?
A good decision tool should give stable answers when you nudge its assumptions. We re-ran both profiles under five weight sets.
Table 5. CEOtudent calculation: top-3 ranking under alternative weights (Demand/Durability/Leverage/Cost)
| Weight set | Profile A top 3 (score) | Profile B top 3 (score) |
|---|---|---|
| Baseline 30/25/25/20 | AI workflow automation (4.30); Data analysis (3.85); People leadership (3.60) | Operations and PM (4.55); Data analysis (4.10); AI workflow automation (4.05) |
| Demand-heavy 45/20/20/15 | AI workflow automation (4.45); Data analysis (4.10); Operations and PM (3.90) | Operations and PM (4.65); Data analysis (4.30); AI workflow automation (4.25) |
| Durability-heavy 20/40/20/20 | AI workflow automation (4.00); People leadership (3.80); Data analysis (3.60) | Operations and PM (4.40); People leadership (4.00); Data analysis (3.80) |
| Equal 25/25/25/25 | AI workflow automation (4.25); Data analysis (3.75); UX research (3.50) | Operations and PM (4.50); Data analysis (4.00); AI workflow automation (4.00) |
| Time-poor 20/20/20/40 | AI workflow automation (4.20); Single ad-platform certification (3.80); Data analysis (3.60) | Operations and PM (4.40); AI workflow automation (4.00); Data analysis (3.80) |
In the Equal row for Profile B, data analysis and AI workflow automation tie at 4.00 and are ordered by Rule 4 (higher Leverage first).
The first choice holds under all five weightings for both profiles. That is the signal you are looking for: when your #1 survives every reasonable weighting, commit to it. The lower positions move around. People leadership rises whenever Durability gets more weight, and the Time-poor row shows why Rule 6 exists. When you give Cost 40% of the weight, the ad-platform certification (Durability 2) jumps to second place for the marketer. The floor rule removes it. Under time pressure, people tend to choose skills that are quick to learn and quick to expire, and the rule guards against that.
The CEO + Student lens: from score to curriculum
The matrix is the CEO half of the decision: allocate scarce capital (your hours) to the highest risk-adjusted return, check that the answer holds under different assumptions, then commit. A CEO also runs a portfolio. Treat your skills the way a company treats its business lines:
- One core bet (the matrix winner) gets most of your learning hours for a defined period.
- Maintenance of skills that already earn for you gets a small fixed allowance, so they do not decay while you build the new one.
- One scouting position, a low-cost exploration of the #2 or #3 option, keeps your next move informed.
The Student half begins once the choice is made. A score does not teach you anything. What turns the choice into a skill is a sequenced plan, deliberate practice and an early public proof. For that, build a personal curriculum with AI: define “useful” as an observable output, work back from it, and put a checkpoint every few weeks.
Then re-score. Markets move, and the BLS itself republishes projections every year (the 2026-36 edition is due in 2027). Your own Leverage scores also change as you learn: every skill you add raises the Leverage of the skills next to it. That is the compounding the matrix is designed to capture.
How to run the matrix in 30 minutes
- List 5-8 candidates. Include at least one “boring” option close to your current job. It often wins on Leverage.
- Score Demand from data. Look up the WEF category and, if you have a target role, its projected growth against the 3.5% baseline (or your country’s equivalent).
- Score Durability. Ask whether the skill is the task AI performs, or the judgment that directs and checks that task. Do not use occupation exposure as a shortcut (see Table 2).
- Score Leverage and Cost honestly. These are your private inputs. If you are unsure, ask a colleague who knows your work.
- Apply the vetoes, then compute. Rank the skills and apply the tie-breakers.
- Stress-test. Re-run with Demand-heavy and Durability-heavy weights. If your #1 survives, commit. If it does not, the choice is close and Leverage should decide.
- Write down a re-score date 90 days out.
FAQ
What skill should I learn next in 2026?
It depends on your starting point, which is why a single list cannot answer it. Score your candidates on Demand, Durability, Leverage and Cost-to-competence. For many knowledge workers the matrix will point to a data- or AI-related skill, because WEF employers rank AI and big data first with a net increase of 87. But in our worked example the operations coordinator’s best move was operations and project management, because of Leverage.
Should I avoid skills in jobs with high AI exposure?
No. BLS states that its exposure categories are not forecasts of employment decline. In our join of BLS data, 8 of the 30 fastest-growing occupations are in the Very-high exposure group, while Low-exposure occupations had the highest share projected to shrink (36.6%). Score Durability at the level of the skill instead.
How is this different from a skills gap analysis?
A gap analysis tells you what you are missing for a given target. The matrix helps you choose which gap to close first, by weighing market value, longevity, fit and cost against each other.
Can I change the weights?
Yes. The 30/25/25/20 split is an editorial default. Test your result under other weights (Table 5). If you give Cost more than 30% of the weight, apply the Durability floor of 3 so the tool does not favour skills that fade quickly.
How often should I re-score?
Every 90 days, or whenever something important changes: a new role, a new tool in your field, or a new edition of the WEF report or BLS projections.
Does a higher score guarantee a higher salary?
No. The matrix improves how you allocate learning time. It does not predict outcomes. Wage and growth figures describe occupations on average, not what any individual will earn.
Sources
World Economic Forum, The Future of Jobs Report 2025, Insight Report, January 2025 (Key findings; Figure 3.4 Skills on the rise, 2025-2030).
U.S. Bureau of Labor Statistics, Employment Projections, Table 1.3 Fastest growing occupations, 2025 and projected 2035 (last modified August 27, 2026).
U.S. Bureau of Labor Statistics, AI exposure categories and 2025-35 employment projections, supplemental data file, 2026.
U.S. Bureau of Labor Statistics, Artificial Intelligence (AI) exposure categories: methodology and limitations, Employment Projections publications (last modified August 27, 2026).
U.S. Bureau of Labor Statistics, Measures of Education and Training, Employment Projections technical documentation (last modified August 27, 2026).
Note on data: Table 1 is verified data from the WEF Future of Jobs Report 2025. Table 2 is a CEOtudent calculation from BLS data. Table 3 is the CEOtudent editorial framework. Tables 4 and 5 are CEOtudent calculations in which Demand is data-derived and Durability, Leverage and Cost are illustrative editorial judgments.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
