01
Skill change is contextualSimple automation-versus-augmentation binaries miss important boundary conditions.
Digital technologies reshape skills through multiple mechanisms whose effects depend on work and institutional context.
01 / PURPOSEPOSTER COMPANION
Companion to the printed SURF 2026 poster — what the sheet could not fit
Different documents.
One shared frame.
QUICK READ
Question, finding and limit—before the detail.
QUESTION
How closely do sampled AI-skill pages align with business-school courses?
MAIN FINDING
166 of 239 screened job topics (69.5%) appear in both sets of documents.
LIMIT
This measures job-anchored textual alignment in sampled documents—not salary, capability, demand, job or course replacement, or monetary value.
- Job topic
- A textual factor derived from occupation and task text; both document sets are scored against the same fixed map.
- Shared breadth
- How many screened job topics meet the presence rule in both document sets.
- Alignment depth
- A weighted score of how consistently sampled skill pages engage topics a discipline emphasises; not curriculum share.
Occupation + task text
Curriculum documents
AI-skill pages
ONE SHARED FRAMEText can be compared.Capability still has to be tested.
TEXTUAL ALIGNMENT ≠
task performanceadoptionreplacementcausal impactdemandmonetary value
02 / LITERATUREWHY THIS QUESTION
The team’s shared literature review
Value depends on context—and on the outcome being measured.
Three findings from that reading set the terms for a narrower question: where the same job-related topics show up in both sets of documents.
02
Complement and substituteAI can complement some tasks and substitute for others within the same workflow.
The direction of change depends on tasks, technical capabilities, workflow design and organisational adaptation.
03
Value needs outcomesMarket value and realised effects are studied with outcomes, not document similarity alone.
Demand, wage premiums, quality, fees and employment are the outcome measures this literature relies on.
Alekseeva et al. (2021); Fedyk et al. (2022); Friedman et al. (2026)
THE NARROW GAP
This project takes one question out of that literature: which job topics show up in both sets of documents.
03 / DATA + METHODONE FIXED MAP + SIX STAGES
From unlike sources to one comparison
How two very different kinds of document were made comparable.
Six stages, from raw sources to one comparison. The panel beside each stage says what the current number counts.
01
SOURCE
Three inputs begin in different forms.
Occupation and task text define the comparison space. University catalogues and public AI-skill pages supply the two document sets.
02
COLLECT
The two document sets were collected separately.
Course and programme text came from official university sources. Public skillsmp.com pages were collected by a separate part of the team.
03
NORMALISE
Everything becomes plain, comparable text.
Course records kept their year and their source, and were checked for quality and completeness. Duplicate skill pages were removed before preparation.
04
FIT + PROJECT
One fixed map holds both document sets.
The method fits 361 job topics — textual factors — on occupation and task text, then places 38,764 course documents and 73,749 skill documents on that same map.
Placing a document on the map means scoring it against all 361 topics; the document itself is not changed. 40,124 counts course-panel rows, and not every row enters later analysis. 38,764 counts the documents that reached the shared map — a different unit at a different stage.
- SOURCE TEXT
- O*NET occupation and task database
- METHOD
- Cong, Liang, Zhang & Zhu (2025), textual factors
05
SCREEN
An interpretability screen keeps 239 of the 361 topics.
The rule is ours, and it is stated in full so it can be argued with: keep factors with one dominant pattern and a top word outside the filler list; the second signal must be no more than 75% of the first.
06
COMPARE
Breadth and depth answer different questions.
Breadth counts the topics present in both document sets. Depth describes where skill pages engage a course's topics most consistently.
Present in a document set = at least 10 distinct documents at an absolute loading of 0.05 or more. Shared means both sets meet that rule.
METHOD STAGE 01SOURCE
JOB DEMANDS
CURRICULA
AI-SKILL PAGES
04 / RESULTSBREADTH AND DEPTH
The breadth question, answered
SHARED BREADTH69.5%
166 of 239 screened job topics meet both document thresholds—the course and skill documents describe many of the same job-anchored topics.
166 in both 73 other screened topics
IT DOES NOT MEAN
“AI can perform 69.5% of a course or job.”
Alignment is textual overlap between documents, not evidence of what AI can do.
A different question
Descriptive alignment depth by discipline
Depth measures how consistently skill pages engage the topics a course leans on most. It is a weighted score, not the share of a course that is covered, and no significance test supports the rank order.
01
Information systems & analyticsn=2,635
9.1%
02
Marketingn=2,377
7.0%
03
Supply chain & operationsn=1,120
6.9%
04
Management & strategyn=7,321
6.6%
05
Accountingn=2,145
6.4%
06
Business law & compliancen=812
5.6%
07
Organizational behavior & HRn=1,226
5.6%
08
Salesn=72
5.6%
09
Economicsn=7,151
5.5%
10
Hospitality & tourismn=392
5.0%
11
Financen=4,685
5.0%
12
Real estaten=277
3.5%
| Rank | Discipline | Mean depth | Included course documents |
|---|---|---|---|
| 1 | Information systems & analytics | 9.1% | 2,635 |
| 2 | Marketing | 7.0% | 2,377 |
| 3 | Supply chain & operations | 6.9% | 1,120 |
| 4 | Management & strategy | 6.6% | 7,321 |
| 5 | Accounting | 6.4% | 2,145 |
| 6 | Business law & compliance | 5.6% | 812 |
| 7 | Organizational behavior & HR | 5.6% | 1,226 |
| 8 | Sales | 5.6% | 72 |
| 9 | Economics | 5.5% | 7,151 |
| 10 | Hospitality & tourism | 5.0% | 392 |
| 11 | Finance | 5.0% | 4,685 |
| 12 | Real estate | 3.5% | 277 |
Sample sizes differ a lot: Sales rests on 72 course documents and real estate on 277, against 7,321 for management & strategy. Ranks that sit close together should not be read as real differences.
Depth base: 30,213 classified course documents with at least one hit; 3,848 classified no-hit documents excluded and 4,703 unclassified omitted. Skill hit rates use 73,749 projected documents. A hit means an absolute loading of 0.05 or more.
05 / MEANINGINTERPRETATION + LIMITS
The defensible conclusion
The sampled documents share many job-anchored topics.
The map identifies shared thematic ground for closer review. Every stronger claim still needs independent evidence.
SCOPE
Sampled public AI-skill pages from one marketplace, and business-school course documents.
What the comparison shows
- 166 of 239 screened job topics appear in both sets of documents.
- Depth varies by discipline, from 9.1% in information systems and analytics down to 3.5% in real estate.
- A shortlist of topics worth checking against real capability and demand.
Capability, replacement or value outcomes
- Task performance, adoption or skill quality
- Automation, worker or course replacement
- Causal impact, demand or monetary value
Measured alignment depth is highest for information systems and analytics and lowest for real estate. That is a starting order for topic-by-topic curriculum review, not a ranking with a significance test behind it.
166 of 239 screened job topics appear in both sets of documents. That names the shortlist where capability and demand still have to be measured with the kind of outcome evidence chapter 02 describes.
WHAT IS AND IS NOT HERE
The method is open. The documents behind it are not republished.
This site explains how the study was done and what it found, using summary figures and publicly available sources. It does not republish the course descriptions or the marketplace pages the analysis read, and there are no data files, code or downloads. Every figure is shown with what it counts, what it is counted out of, and what it cannot be used to claim, so you can check what a number means before relying on it.
SURF PROJECT
Valuation of AI Skills in Online Marketplaces
SURF-2026-0437
RESEARCH TEAM
- Amer Amarneh
- Qiyuan Deng
- Shixuan Heng
- Cheng Qian
- Ziqi Yang
- Wantong Zhao
SUPERVISORJia Hao Bryan Chai
CO-SUPERVISORJoshua Lee Sze Ray
PROGRAMMESummer Undergraduate Research Fellowship 2026
DESIGN + IMPLEMENTATIONAmer Amarneh
Companion website design and development · poster design lead · business-school curriculum collection automation and verification
PROJECT VOLUNTEERS
- Fanli Xu
- Lingxi Li
- Siyu Li
- Yifei Jiang
SELECTED RESEARCH REFERENCES
- Cong, Liang, Zhang & Zhu (2025), Textual Factors
- Grimshaw & Miozzo (2026), How Are Skills Changing with Digital Technologies?
- Tschang & Almirall (2021), Artificial Intelligence as Augmenting Automation
- Alekseeva et al. (2021), The Demand for AI Skills in the Labor Market
- Fedyk et al. (2022), Is Artificial Intelligence Improving the Audit Process?
- Friedman, Sutherland & Vetter (2026), Technological Investment and Accounting
IN ONE SENTENCE
166 of 239 screened job topics appear in both sets of documents—a map of where to look next, not a measure of what AI can do.