1Overview

The Intellectica Skills Demand–Supply Intensity Index™ (SDSI Index) measures how much demand–supply pressure a specific skill domain carries relative to the wider knowledge-work economy in a given country. It provides business leaders and HR decision-makers with a comparable, interpretable metric for ranking talent pressure across skill domains.

The index answers a single question: how does demand–supply pressure in a specific skill domain compare with the wider knowledge-work economy? It produces a percentage-based score indexed at 100%, where 100% represents the average pressure across the tracked reference universe. Values above 100% indicate greater-than-reference pressure; values below 100% indicate less.

The SDSI Index is a cross-sectional instrument. Eleven skill baskets are tracked simultaneously within a country, covering technology, business functions, design, and professional services, so that relative pressure can be compared across domains from the first release onward. The time series builds forward with each monthly publication.

1.1Why this index exists

Existing labour-market indicators tend to be either too broad (economy-wide unemployment rates) or too narrow (individual company hiring metrics). Business leaders making talent-strategy decisions need a middle layer: skill-specific demand–supply signals that can be compared across domains and tracked over time. The SDSI Index fills this gap by translating LinkedIn Talent Insights data into a standardised, interpretable metric.

1.2Audience

The primary audience is business leaders and HR decision-makers — professionals asking questions such as: should we be concerned about finding AI talent in Greece? Which skill domains are experiencing the greatest hiring pressure? Is the talent scarcity we feel in our sector specific to us, or general? The SDSI Index provides the quantitative basis for these conversations.

1.3Design intent: a comparative instrument

The index is built to support comparison — between skill domains within a release, and for each domain across releases. It is not built to quantify absolute hiring difficulty. The distinction matters because the underlying inputs are counts of professional profiles and open job postings, and postings list skills more liberally than profiles do. A basket whose skills are widely listed as secondary requirements will therefore register a higher absolute level than the genuine scarcity of that domain warrants, while its position relative to other baskets remains informative.

Section 4.7 states the resulting reading rules explicitly, and Section 9 records the limitation. In summary, the index is designed for ranking and tracking; it is not designed to support claims that hiring in one domain is a specific multiple more difficult than in another.

2Data Sources

The SDSI Index uses a single data source to maintain methodological simplicity and to eliminate cross-source comparability problems.

2.1LinkedIn Talent Insights

LinkedIn Talent Insights is a proprietary analytics platform providing near-real-time data on the professional workforce. It is the sole data source for the SDSI Index.

• Update frequency. Near real-time — professional profiles and job postings update continuously.

• Access type. Paid subscription.

• Historical depth. None. The platform provides current-state snapshots only; there is no facility to query values for past dates. The index begins live from launch and builds forward. No backtested values are produced or implied.

• Population coverage. The platform over-represents white-collar, urban, tech-adjacent professionals. Coverage is strong for knowledge-work domains and structurally incomplete for manual trades, the informal economy, and non-digital sectors. This defines the scope boundary of the index and is the reason the benchmark is a knowledge-work reference universe rather than an unfiltered national count.

• Skill-association mechanics. Job postings are matched to skills by the platform's own association logic, which is subject to revision by the provider. A revision changes measured posting counts for skill-filtered queries without any change in underlying hiring activity. The index treats the stability of this association as a monitored assumption rather than a given; see Sections 7.3 and 9.

2.2Measured inputs per release

Six quantities are measured for each country in each release. All are recorded on a single collection date so that every value within a release is mutually consistent.

3Components & Dimensions

The SDSI Index is a measurement framework applied across 11 skill baskets spanning technology, business functions, design, and professional services. Each basket represents a distinct professional domain, and the framework produces an SDSI score per basket, enabling cross-sectional comparison.

Two metrics are calculated per basket: the Tightness Ratio (absolute demand–supply balance) and the Skills Demand–Supply Intensity score (pressure relative to the reference universe).

3.1Skill basket construction principles

Each skill basket is a curated list of LinkedIn skill tags representing a professional domain. The following principles govern basket construction.

• Domain essentiality. A skill belongs in a basket if it is essential to working in that domain, even if it is also used in other fields. The test: is this skill a core part of the practitioner's toolkit in this domain? Python is not exclusively an AI skill, but it is the language in which virtually all AI work is done.

• Domain diagnosticity. A skill qualifies only if it is a technology or practice of the domain, not a general competency the domain merely uses. Social Media Marketing is marketing — its claimants are doing marketing work wherever they sit. Negotiation is not sales — it is a deal competency claimed across legal, procurement, and executive populations. General business competencies inflate a basket's population with professionals from unrelated domains and are excluded regardless of how naturally they associate with the domain.

• Homonym check. A tag must mean the same thing across the platform's population. Tags that carry a different meaning in another profession contaminate the basket with an unrelated population and are excluded. Sourcing, for example, is claimed on the platform predominantly as a recruiting term (candidate sourcing) rather than a procurement practice.

• Taxonomy-level consistency. Skills are selected at mid-level specificity — not umbrella terms (not “Artificial Intelligence” as a catch-all) and not niche techniques (not “Hyperparameter Tuning”). Preferred level: Machine Learning, Deep Learning, Computer Vision, NLP, Generative AI.

• Dominance monitoring. If a single skill accounts for more than 40–50% of a basket's professional count when queried independently, it is flagged as dominant. Dominant skills are not automatically excluded if they pass the essentiality and diagnosticity tests, but the basket's SDSI is tracked in parallel with and without the dominant skill at each release. Sustained divergence between the two values triggers reconsideration of the skill's inclusion.

• Basket size target: 10 ± 3 skills (7–13 per basket). Fewer than 7 creates fragility; more than 13 dilutes the basket toward the general population. Domains are not required to hold equal counts.

• Longevity over trendiness. Skills expected to remain meaningful for 5+ years are preferred. Emerging skills may be included and are flagged as candidates for the 6-month review cycle. Brand-name skills are included only where they function as de facto industry categories with 10+ years of stability.

• Selection-time checklist. Each skill is assessed against six questions: does it pass domain essentiality; is it diagnostic of the domain rather than a general competency; does its tag mean the same thing across the platform's population; is it at mid-level specificity; does it trigger dominance monitoring; is it likely to remain relevant in 5+ years.

3.2Basket overview

The index tracks 11 skill baskets covering 105 skills across the knowledge-work economy. The union of these 105 skills also defines the Reference Skill Universe against which every basket is benchmarked; see Section 5.1.

3.3AI & Data Science (13 skills)

3.4Cybersecurity (10 skills)

3.5Cloud Computing & DevOps (10 skills)

3.6Software Engineering (11 skills)

3.7Digital Marketing (8 skills)

3.8Finance & Accounting (9 skills)

3.9Human Resources & Talent Management (9 skills)

3.10Sales & Business Development (9 skills)

3.11Supply Chain & Logistics (9 skills)

3.12UX/UI & Product Design (9 skills)

3.13Legal & Compliance (8 skills)

4Calculation Formula

The SDSI is a location quotient: a basket's share of demand divided by its share of supply, measured against a common reference population. It is computed in three steps.

4.1Step 1 — Tightness Ratio (TR)

The Tightness Ratio measures the absolute demand–supply balance for a skill basket in a country. It is the number of active job postings per professional in that skill domain.

Where TR(s) is the Tightness Ratio for basket s; JP(s) is the count of active job posts matching at least one skill in basket s; and TP(s) is the count of professionals holding at least one skill from basket s.

4.2Step 2 — Reference Tightness Ratio (TR-Reference)

The same calculation applied to the Reference Skill Universe: the union of all 105 tracked skills, queried once as a single deduplicated population. This is the benchmark against which every basket is measured. Section 5 sets out the reference universe and the reasoning behind its selection.

4.3Step 3 — Skills Demand–Supply Intensity (SDSI)

The SDSI expresses a basket's demand–supply pressure relative to the reference universe, indexed to 100%. The reference universe always equals 100%.

4.4Equivalent form

The same quantity can be written as a ratio of shares, which is the form that makes the construction's logic visible. The two expressions are algebraically identical, not alternative measures:

Read in the share form, the index is a basket's share of reference demand divided by its share of reference supply. This form also makes a robustness property explicit: if the data source revises its skill-association logic such that all skill-filtered posting counts scale proportionally, JP(s) and JP(r) move together and the ratio is unchanged. Section 5.2 quantifies this.

4.5Rounding convention

Published SDSI values are rounded to whole percentages (213%, not 213.2%). Tightness Ratios are reported as percentages to one decimal place (6.4%). Further decimal places imply false precision, since the underlying counts fluctuate daily.

4.6Score interpretation

The SDSI is unbounded and can range from near 0% to well above 250%. The bands below describe relative pressure against the reference universe. They are labelled in relative terms deliberately: a band describes where a basket sits among tracked knowledge-work domains, not how difficult hiring is in absolute terms.

4.7How to read an SDSI value — and how not to

The index supports three readings and excludes one.

• Supported — ranking. Which domains carry the most demand–supply pressure in this country this month, and in what order. This is the primary reading and the headline of every release.

• Supported — banding. Whether a domain sits above, at, or below the tracked knowledge-work average, using the bands in Section 4.6.

• Supported — tracking. Whether a domain's relative position is rising or falling across releases, provided the measurement-stability checks in Section 7.3 have passed for the periods compared.

• Not supported — absolute hiring difficulty. An SDSI of 213% does not mean that hiring in that domain is 2.13 times harder, or that a vacancy takes 2.13 times longer to fill. The index measures postings per professional as recorded by the data source, and job postings list skills more liberally than professional profiles do. Baskets whose skills appear frequently as secondary requirements therefore carry an upward-biased absolute level. The ratio accurately reflects the recorded inputs; the bias arises from the asymmetry in how demand and supply are recorded.

Intellectica commentary accompanying each release observes this discipline, and Section 8.2 states it as an editorial rule.

5Baseline Protocol

5.1The Reference Skill Universe

The Reference Skill Universe (RSU) is the benchmark population of the index. It is defined as the union of every skill tracked in the framework — the 105 skills listed in Sections 3.3 to 3.13 — queried as a single population so that both professionals and job postings are deduplicated within it.

Every basket is a subset of the RSU, so each SDSI is a proper location quotient: a part measured against the whole that contains it. The RSU is queried with the same filter logic used for the baskets, ensuring the numerator and denominator are subject to identical matching mechanics. As measured on 21 August 2026 for Greece, the RSU comprised 395,168 professionals and 11,906 active job posts, a reference tightness of 3.0%.

5.2Why the benchmark is skill-anchored

An unfiltered national posting count was evaluated and rejected as the benchmark, for two reasons.

First, fragility. Basket demand JP(s) is a skill-filtered count. An unfiltered national count is not subject to the platform's skill-association mechanics, so a revision of those mechanics by the provider moves every basket numerator while leaving such a denominator untouched — propagating the revision unattenuated into every published score. A skill-anchored benchmark is subject to the same mechanics as the baskets, so a proportional revision cancels:

Second, scope. The data source covers a knowledge-work population rather than the full national labour force, as recorded in Section 2.1. An unfiltered national count as denominator would imply a comparison — against the whole labour market — that the underlying data cannot support. The RSU states the comparison the data can support: pressure in one tracked domain against pressure across all tracked domains. Its boundary is also a diagnosticity boundary: because every RSU skill has passed the selection principles in Section 3.1, the benchmark population is professionals holding at least one domain-diagnostic skill, rather than a general white-collar pool. The unfiltered market ratio is retained, as macroeconomic context and as a stability diagnostic, in Sections 6.2 and 6.3.

5.3Self-anchoring design and implications

The index is self-anchoring. The RSU tightness ratio is recomputed from fresh data with every release; there is no historical reference period, no anchor drift, and no scale-factor maintenance. An SDSI of 213% means 2.13 times the reference universe's tightness in that release, not relative to a fixed past month.

Because the benchmark is recomputed each period, month-over-month movement in an SDSI reflects a change in relative position rather than in absolute level. A basket's SDSI can fall while its own Tightness Ratio rises, if pressure across the reference universe rose faster. The auxiliary Tightness Ratio in Section 6.1 supplies the absolute context needed to distinguish these cases, and every release reports both.

5.4Reference universe versioning and re-basing

Because the RSU is the denominator of every score, a change to its composition shifts every published value. Its versioning is therefore governed explicitly:

• The RSU is versioned independently of the basket set. RSU v2.0 is fixed as the union of the 105 skills documented in this version. Adding a new basket to the published ranking does not automatically alter the RSU; the new basket is scored against the existing RSU until the RSU is itself revised.

• Any change to RSU composition is a major-version event. It requires a new version number, public disclosure, and a statement that values before and after are not directly comparable.

• Re-basing is not retroactive. The data source does not support historical queries, so prior periods cannot be recomputed under a revised RSU. A re-based series begins forward from the version in which the revision takes effect, and the prior series remains available at its own version.

6Auxiliary Metrics

6.1Tightness Ratio (TR)

The Tightness Ratio is published alongside each basket's SDSI. Where the SDSI gives relative position, the TR gives the absolute demand–supply signal: active job postings per professional in the domain.

The TR is reported with the same caution that applies to the SDSI's absolute level. It counts postings that match at least one basket skill, and postings list skills more liberally than profiles do, so the ratio should be read as postings-per-professional as recorded by the source rather than as a measure of vacancies per available candidate.

6.2Market Tightness Ratio (TR-Market)

The unfiltered national ratio, TR(m) = JP(m) ÷ TP(m), is published as macroeconomic context. It is not the benchmark for any SDSI calculation. It answers a separate question — how tight is the country's posting activity overall — and provides the reader with the macro backdrop against which relative pressure is interpreted.

6.3Reference Coverage Ratio (RCR)

The Reference Coverage Ratio expresses the tracked reference universe as a share of the unfiltered national posting count. It is published as a diagnostic and monitored as one of the index's primary measurement-stability checks.

Under stable measurement the RCR moves gradually, reflecting genuine shifts in the composition of national hiring toward or away from tracked knowledge-work domains. A step change in the RCR between consecutive releases, unaccompanied by a corresponding movement in professional counts, indicates that the relationship between skill-filtered and unfiltered counts has changed at source rather than in the labour market. Section 7.3 defines the consequences.

6.4Two-dimensional interpretation

Reading the SDSI together with the Tightness Ratio separates domain-specific pressure from broad conditions:

7Data Collection & Measurement Stability

7.1Collection

Each monthly release is built from a single LinkedIn Talent Insights snapshot per country, taken on a fixed collection date within the publication month. The same date applies to every basket and to the reference universe in that country-month, so all values within a release are mutually consistent.

Three categories of measurement are recorded per snapshot: the eleven per-basket pairs TP(s) and JP(s); the reference universe pair TP(r) and JP(r); and the unfiltered national pair TP(m) and JP(m). From these, the formulas in Section 4 yield TR(s), TR(r), and SDSI(s), together with the auxiliary metrics of Section 6. Raw counts and derived values are archived with their collection date to build the forward series. Past values are never recomputed retroactively, since the data source does not support historical queries.

7.2Consistency requirements

• Baskets and the reference universe are queried with identical filter logic and identical geographic scope, so that numerator and denominator remain comparable.

• All queries in a release share one collection date. Values collected on different dates are not combined into a single release.

• The reference universe is queried as one population rather than assembled by summing baskets, which would double-count postings matching several baskets.

7.3Measurement stability monitoring

Because the index is self-anchoring, comparability across releases rests on the stability of measurement at source. Four checks are evaluated before each release is published.

• Reference coverage check. The Reference Coverage Ratio (Section 6.3) is compared against its trailing history. A step change inconsistent with the concurrent movement in professional counts is treated as evidence that skill-to-posting association has changed at source.

• Per-basket drift check. Each basket's month-over-month change in JP(s) is compared against its trailing distribution, and against the movement in TP(s) for the same basket. A large divergence between demand-side and supply-side movement across many baskets simultaneously indicates a measurement change rather than a hiring shift.

• Rank-stability check. The cross-sectional ordering of baskets is compared against the previous release. A sharp drop in rank agreement, unexplained by identifiable labour-market events, indicates that measurement rather than hiring has reordered the baskets — the ranking is the index's primary output, so this check guards it directly.

• Thin-signal check. Any basket producing fewer than 50 job posts in a country is published with an explicit low-confidence flag.

Where the first three checks indicate a measurement change at source, the affected release is not published as a continuation of the existing series. Publication is held, the change is documented, and the release is either issued with an explicit discontinuity notice or deferred until the new measurement level is confirmed stable across subsequent periods. Where a change is confirmed to be persistent, the index is re-based under the procedure in Section 5.4 and a major version is issued.

8Publication Format

8.1Release components

• Headline cross-sectional ranking: all 11 baskets ordered by SDSI, each with its value and its relative-pressure band label

• Auxiliary Tightness Ratio alongside each basket

• Reference universe counts and Reference Tightness Ratio for the period

• Market Tightness Ratio and Reference Coverage Ratio as macroeconomic and diagnostic context

• Score-interpretation legend for the bands in Section 4.6

• Trend chart of SDSI over time per basket, from the third release of a country onward

• Intellectica commentary: two to three sentences on the most notable relative movements or comparisons

• Low-confidence flags on any basket below the thin-signal threshold, and a discontinuity notice where one applies

8.2Interpretation discipline

Commentary and headline copy accompanying each release describe rank, band, and direction of movement. They do not characterise an SDSI value as a multiple of hiring difficulty, time-to-fill, or candidate scarcity, for the reasons set out in Section 4.7. Where a release quotes a value, it is paired with its band label rather than with an absolute-difficulty gloss.

8.3Channels and formats

The index is published as an Intellectica-branded dashboard on the Intellectica website, with supporting distribution via LinkedIn and an email digest. Each release is issued in two self-contained formats generated from the same data state: a static variant for inline publication and print, and an interactive variant for embedding. Both carry source attribution and a link to this methodology document, which is hosted at a stable versioned URL so that prior versions remain available as later ones are published.

8.4Cadence

Monthly. Each release is published within the first week of the calendar month and reports the preceding month's snapshot. Where the stability checks in Section 7.3 hold a release, the delay and its reason are stated in the following publication.

9Limitations & Considerations

• Benchmark scope is knowledge-work, not the whole economy. 100% denotes the average of the tracked reference universe — the 105 skills across 11 knowledge-work domains — not the national labour-market average. The index does not measure pressure relative to the whole economy, and comparisons to national labour statistics are not like-for-like.

• Coverage bias in the source population. The data source over-represents white-collar, urban, tech-adjacent professionals. Values are reliable for knowledge-work domains and structurally incomplete for manual trades, hospitality, construction, and the informal economy. These domains are outside the index's scope rather than under-measured within it.

• Project management is not covered. The platform's available tags for the domain are predominantly general management competencies that fail the diagnosticity principle; a basket built on them measures general management rather than project management, and testing confirmed its population is heavily diluted with non-practitioners. The domain is excluded until a diagnostic basket (certifications, methodologies, tooling) can be validated; see Section 11.

• Absolute levels are upward-biased for baskets rich in secondary skills. Job postings list skills more liberally than professional profiles do. A basket whose skills frequently appear as secondary or desirable requirements — cloud platforms being the clearest case — registers a higher absolute level than the domain's genuine scarcity warrants. Relative position and movement remain informative; the absolute multiple should not be read as hiring difficulty. See Section 4.7.

• Dominant-skill dilution in AI & Data Science. Python is retained as essential to AI work but is held by large populations doing no AI work, which inflates the basket's supply side and may understate the domain's relative pressure. The basket is tracked with and without Python at each release, and the divergence is reviewed at each 6-month cycle.

• Skill-association stability is an assumption, not a guarantee. Measured posting counts depend on the source's logic for associating postings with skills, which the provider may revise. Benchmarking against the reference universe neutralises proportional revisions but not revisions that affect baskets unevenly; those are detected and disclosed rather than corrected. A confirmed revision constitutes a discontinuity in the series.

• No historical baseline and no backtest. The source provides current-state snapshots only. The index begins live from launch and builds forward. No backtested values are produced or implied, and no re-based series is reconstructed backwards.

• Snapshot, not flow. The source reports the current stock of professionals and currently open postings. It does not capture hiring velocity, time-to-fill, offer acceptance, or salary pressure. The index measures structural pressure, not hiring outcomes.

• OR-logic skill filters and cross-basket overlap. A professional or posting matching any one basket skill is counted. Broadly-skilled professionals and multi-skill postings therefore appear in several baskets at once. Basket values are not statistically independent of one another, and basket counts are not additive — no calculation in this methodology sums them.

• Daily fluctuation in counts. Professional and posting counts shift daily. A single monthly collection date mitigates this but introduces timing sensitivity; a release reflects the state of the source on one date, not a monthly average.

• Interpretation bands are provisional. The band boundaries in Section 4.6 are initial estimates and will be re-derived from the observed cross-sectional distribution after six releases.

• Thin-signal risk in smaller markets. Legal & Compliance and UX/UI & Product Design may produce limited counts in smaller countries. Baskets below 50 job posts are published with an explicit low-confidence flag.

10Versioning & Updates

10.1Versioning convention

Minor versions (x.1, x.2 …). Adding or removing individual skills within an existing basket without altering the reference universe, adding a new skill basket scored against the existing reference universe, documentation improvements, cosmetic refinements. No structural effect on the formula, the benchmark, or the interpretation framework.

Major versions (x+1.0). Changes to the SDSI formula; changes to the composition of the Reference Skill Universe; changes to the benchmark definition; modification of interpretation bands; changes in publication frequency; fundamental redefinition of basket construction principles; and re-basing following a confirmed measurement discontinuity.

10.2Component updates

Skill baskets are reviewed every six months. Triggers for component change:

• A skill shows a decaying professional count or negligible posting activity across multiple consecutive periods

• A new skill emerges with strong, continuous signal that fits an existing basket's domain

• Technological or market shifts render a skill obsolete or change its domain association

• A dominant skill's with-and-without values diverge persistently, indicating a distortive effect

• A skill is found to fail the diagnosticity principle — widely claimed across unrelated populations — or the homonym check, and therefore to contribute more noise than signal

10.3Adding new skill baskets

Adding a basket is a minor-version event. The new basket is scored against the existing Reference Skill Universe and enters the cross-sectional ranking from its first month of publication; no phased inclusion is required, because each basket is calculated independently and no basket's value depends on any other. Its skills are incorporated into the reference universe only at the next major version, which keeps the benchmark — and therefore the comparability of all published values — stable between major releases. Until then, a newly added basket's own skills are not represented in the denominator; the effect is small relative to a 105-skill universe and is disclosed at the release in which the basket is introduced.

10.4Major-version communication

Major methodology changes are announced with the release in which they take effect, using the following form:

11Roadmap

• Interpretation-band recalibration. Re-derive band boundaries from the observed cross-sectional distribution after six releases.

• Dominant-skill tracking formalisation. Publish the AI & Data Science with-and-without-Python parallel series once sufficient history exists, and extend the treatment to any basket whose dominant skill shows persistent divergence.

• Project Management (redesigned). Construct and validate a diagnostic project-management basket built on certifications, methodologies, and tooling (e.g., PMP, PRINCE2, Kanban, project-management software tags) rather than general management competencies. Publish only if signal, population plausibility, and diagnosticity tests pass.

• Demand-side filter calibration. Test whether requiring a minimum of two matched skills, or anchoring on designated core skills, further reduces secondary-skill bias in posting counts without collapsing counts in the smaller baskets.

• Cross-country expansion. Apply the framework to additional countries, each with its own reference universe measured on the same skill list, enabling comparison of relative skill pressure across geographies.

• Sub-basket analysis. Within large domains, explore sub-basket breakdowns — within AI & Data Science, for example, core ML against generative AI against ML infrastructure.

• Additional baskets. Evaluate Green Energy & Sustainability, Data Engineering, and other domains as source coverage and signal warrant.

• Integration with the Intellectica AI Pulse Index. Explore complementarity between talent-market pressure and broader AI adoption and discourse signals.