Bridging the Education–Employment Gap: Linking Labour Market Needs with University Programmes Through AI-Assisted Insights and Micro-Credentials

Abstract

This study explores how artificial intelligence (AI)-driven labour market analytics can support the alignment of university curricula with emerging skill demands and inform the development of targeted micro-credentials. A mixed-methods approach was applied, combining AI-assisted analysis of approximately 30,000 online job advertisements in Latvia with ESCO-based skill mapping, curriculum analysis, and expert interviews. The study develops and validates a three-layer analytical framework integrating labour market demand, professional standards, and programme learning outcomes. Three occupations—Personnel Specialist, Finance Manager, and Organisation Manager—were analysed at Riga Technical University as proof-of-concept cases. The findings demonstrate that strict one-to-one ESCO matching overestimates curriculum gaps because labour market and educational actors often describe competencies at different levels of abstraction. Composite matching significantly improves alignment estimates by identifying functionally equivalent competencies embedded across curricula. Nevertheless, the analysis reveals a persistent under-representation of digital competencies across all programmes, confirmed by industry experts. Interviews further identify a “pedagogical transfer gap”, where formally acquired competencies are insufficiently applied in practice, and highlight employer support for high-quality micro-credentials focused on technical upskilling. The study contributes an AI-assisted curriculum-monitoring framework that combines large-scale skill extraction, semantic alignment, and stakeholder validation, offering universities a practical tool for evidence-based curriculum renewal and lifelong learning development.

1. Introduction

The accelerating digital transformation of economies and workplaces is reshaping skill requirements across sectors, challenging higher education institutions (HEIs) to maintain alignment between academic programmes and the realities of employment (Kopackova et al., 2024). The persistent skills mismatch remains a major concern affecting graduate employability and institutional relevance (Goulart et al., 2022). Nelson et al. (2025) argue that traditional university curricula often fail to respond to evolving work–life realities, leading to systematic misalignments between graduate qualifications and occupational needs. Similarly, Geng (2025) emphasises that employability in the twenty-first century requires a broader skillset that includes adaptability, digital literacy, and problem-solving capabilities beyond formal qualifications. These findings underscore the need for HEIs to adopt data-informed and evidence-based approaches to curriculum design, ensuring that programme outcomes reflect the competencies increasingly sought in the labour market.

At the same time, advances in artificial intelligence (AI) and data analytics present new opportunities for bridging the education–employment gap through continuous monitoring of labour market trends (Sanguino et al., 2025). Research highlights that digital intelligence and contextualised skill development are essential for improving work readiness (Yaya et al., 2025; Lāma & Lastovska, 2025), while soft skills remain integral to graduate employability and career adaptability (Urkia-Basterra et al., 2025). Furthermore, flexible learning pathways—such as micro-credentials—have emerged as effective mechanisms for linking education to industry expectations (Siafu, 2024; Sheerin & Brittain, 2023).

As regards Latvia, the mismatch between higher education output and labour market needs has become a prominent concern for policymakers. Latvian Educational Development Guidelines for 2021–2027 emphasise that the future development of the education system must ensure balanced, future-relevant skills acquisition and effective collaboration between the education sector and economic industries (Ministry of Education and Science, 2021). At the same time, several research studies (Guznajeva et al., 2023; Savrina & Martisune, 2021) admit that despite significant demand even for jobs with medium and low-level qualifications, Latvia’s educational system does not adequately support the acquisition of essential skills due to issues such as under-funding of higher education—specifically, insufficient public investment in academic staffing, infrastructure, and programme development—and weak links between academia and industry (Slišāne et al., 2021). Micro-credentials are considered one of the ways to address this issue by providing a faster response from the HEIs to the labour market’s needs (Līce, 2021).

Against this background, this study contributes to the ongoing debate on how universities can use AI-assisted labour market analytics to inform curriculum renewal and design modular learning opportunities that enhance employability and support the digital and sustainable transformation of higher education. The overarching research question guiding the study is as follows: How can AI-assisted analysis of labour market data be used to align university curricula with emerging skill demands and support the design of targeted micro-credentials?

To address this question in a structured and operationalisable way, three sub-questions are defined:

RQ1: What is the nature and extent of alignment between labour market skill demand, professional standards, and university programme learning outcomes when assessed through a unified ESCO-based framework?

RQ2: To what extent does the choice of matching approach—strict one-to-one versus composite ESCO L3 matching—affect the diagnosis of skills mismatch, and what does this imply for how curriculum gaps should be interpreted?

RQ3: Which specific competence gaps persist after composite matching, and how can these inform the design of targeted micro-credentials that meet stakeholder quality expectations?

The aim of the study is to develop and validate an AI-assisted analytical framework that links labour market analytics with higher education curricula to support evidence-based curriculum renewal and the targeted design of micro-credentials. To achieve this aim, three objectives are pursued: (1) to synthesise existing literature on skills mismatch, AI-assisted labour market analytics, and micro-credentials in order to establish the theoretical foundations and identify the research gaps addressed by this study; (2) to apply the three-layer framework empirically to three high-demand occupations at Riga Technical University, comparing labour market demand, professional standards, and programme learning outcomes through ESCO-based exact and composite matching; and (3) to validate the quantitative findings through structured expert interviews with programme directors and industry representatives, and to translate the validated gaps into concrete micro-credential priorities. The study was conducted in the Latvian higher education context, drawing on a dataset of approximately 30,000 online job advertisements and focusing on three ESCO-classified occupations—Human Resources Officer (ESCO 2423.3), Finance Manager (ESCO 1211), and Managing Director/Chief Executive (ESCO 1120) (European Commission, 2026)—as proof-of-concept cases spanning short-cycle, first-cycle, and second-cycle study levels.

2. Literature Review

The three bodies of literature reviewed in this study—skills mismatch and the education–employment gap, AI-assisted labour market analytics, and micro-credentials—are not independent strands but form an integrated rationale for the analytical framework proposed here (see Figure 1). Skills mismatch research establishes why alignment between education and employment is a persistent structural problem and what dimensions of mismatch matter most. AI-assisted labour market analytics address how large-scale, real-time evidence on employer skill demand can be systematically extracted and standardised. Micro-credentials scholarship then provides the policy instrument through which identified gaps can be addressed in a targeted and agile way. Together, these three areas converge on a shared question: how can universities continuously monitor alignment between their programmes and the labour market, and how can they respond when gaps are substantive? The three-layer analytical framework developed in this study—integrating empirical demand from job advertisements, normative demand from professional standards, and educational supply from programme learning outcomes—is designed precisely to operationalise this question in practice.

2.1. Skills Mismatch and the Education–Employment Gap

Over the last decade, an increasing number of policymakers and researchers have focused their attention on the issue of skill mismatches and the education–employment gap, which have significant consequences for the labour market. Skills mismatches are a complex phenomenon that encompasses both individual-level mismatches between an individual’s education and skill level and labour market requirements, as well as structural mismatches between the education system and the labour market as a whole. Skills mismatch refers to a situation where there is a mismatch between the skills required in the labour market and the knowledge and skills acquired by workers, resulting in structural mismatches between the education system and the needs of the labour market, which affects both individual employment levels and overall economic productivity (McGuinness et al., 2018; McGowan & Andrews, 2017; Comyn, 2019; Cedefop, 2010). These studies emphasise that skills mismatch is not a random and temporary phenomenon, but rather a persistent challenge based on the slow ability of the education system to change and respond quickly to changes in the labour market. Consequently, the mismatch between educational programs and labour market needs is considered one of the most significant factors driving skills mismatches.

Several factors contributing to the emergence of skills gaps have been identified in the literature: the slow adaptation of education systems to labour market requirements, insufficient cooperation between educational institutions and employers, rapid technological development, globalisation, demographic trends, and the aging of society (Penesis et al., 2017; McGuinness et al., 2018; Brunello & Wruuck, 2021; Varadarajan et al., 2023). OECD reports describe skills mismatch as a widespread problem, especially in developing countries (Brun-Schammé & Rey, 2021; OECD, 2023a). In particular, they highlight the gap between the technical skills acquired in educational institutions and soft skills (cooperation, critical thinking, effective communication, etc.), which employers are increasingly valuing (Brun-Schammé & Rey, 2021).

Skill mismatch is examined in several dimensions in the literature. Literature distinguishes between vertical mismatch (over-education or under-education), when formal qualifications are higher or lower than those required for a particular job, and horizontal mismatch (field-of-study mismatch), when an employee’s field of study does not match the job requirements. However, their qualification level is appropriate, meaning that the person is not working in their field of specialization. In addition, a skills gap is identified when employees lack specific technical or transversal skills that are clearly required by employers (Di Pietro & Urwin, 2006; Sala, 2011; McGuinness et al., 2018; Salas-Velasco, 2021). The authors emphasise that the gap between education and the labour market is not only quantitative (too many or too few graduates), but mainly qualitative, namely, a mismatch between the type and depth of graduates’ skills and knowledge and the needs of employers (McGowan & Andrews, 2017; Comyn, 2019). McGuinness et al. (2018) systematise skills mismatches, clearly distinguishing between vertical and horizontal mismatches, skill gaps, skill deficits, and skill obsolescence, while noting that these dimensions are measured differently and are weakly correlated with each other (McGuinness et al., 2018).

Empirical data confirm that this mismatch is not insignificant. OECD and Eurostat estimates show that approximately 30% of employees are overqualified, i.e., they work in jobs that do not require their formal educational qualifications, while 20–25% of employees are underqualified because they lack the necessary professional knowledge and skills (Matching Skills and Jobs, 2023; Eurostat, n.d.). Skills mismatches at the macro level can be assessed by comparing the structure of job advertisements that require a specific level of education with the educational structure of the working-age population (Brunello & Wruuck, 2021). These indicators confirm the structural nature of the skills mismatch described above and indicate that the problem affects a significant proportion of workers.

A significant contribution is made by Asai et al. (2020) in their study using PIAAC data, which shows that skills mismatches arise both from deficiencies in labour market mechanisms (e.g., job search, information gaps, support problems) and from macro-level supply and demand mismatches, which are exacerbated by technological change. In addition, the authors emphasise that skills mismatch indicators depend on the measurement approach chosen and propose new combinations of indicators that better reflect the mismatch in skills utilisation. It follows that labour market policy responses must be differentiated: in some cases, the main emphasis should be placed on improving labour market institutions and active labour market policies, while in others, on structural changes in the education system. (Asai et al., 2020). Faustino et al. (2025) shift the emphasis from individual “wrong” choices to institutional ones. The authors examine the extent to which educational institutions can flexibly restructure programs in line with changing labour market needs. They point out that a combination of three elements is essential, namely more flexible programs, close cooperation between educational institutions and industry, and better integration of vocational education into higher education (Faustino et al., 2025).

Furthermore, the Cedefop and Eurostat report on evidence gathering from online job advertisements demonstrates how online job advertisements can be used as a dynamic source of skills intelligence by linking employer requirements to the ESCO skills taxonomy (Matching Skills and Jobs, 2023). This approach enables the identification of skills in high demand at a given time and place more quickly. Data-driven solutions for strengthening the link between higher education and industry are also based on such skills analytics. Andonovikj et al. (2024) propose a combination of generative language models (LLaMA-2) and Sentence-BERT embeddings to analyse course descriptions in business and economics study programs and job advertisement texts in Slovenia, linking the acquired competences to the ESCO taxonomy (Andonovikj et al., 2024). The result is a regular curriculum audit that identifies skills under-represented in study programs and provides a policy tool—systematically reviewing and updating programs in line with labour market needs.

A similar approach is represented by Faisal et al. (2025) in their publication, which describes a machine learning-based decision support framework for aligning higher education programs with labour market needs (Faisal et al., 2025). Using multi-criteria decision-making methods and classification algorithms, the authors show how “weak spots” in courses can be systematically identified based on industry expert and labour market data and offer recommendations for adapting study programs and courses. Overall, these studies demonstrate a shift from episodic employer involvement to a continuous, data-driven process of study program improvement.

At the same time, the broader literature on curriculum design and competence-based education indicates that demand-driven alignment must be balanced with pedagogical principles: learning outcomes are most effective when they are constructively aligned with teaching activities and assessment (Biggs, 1996), and when they support the development of both technical and transversal competencies in authentic professional contexts (Urkia-Basterra et al., 2025; Yaya et al., 2025). This perspective implies that data-driven gap analysis should inform, but not replace, academic judgement about how competencies are taught, practised and assessed.

2.2. AI-Assisted Labour Market Analytics

AI-assisted labour market analytics have rapidly emerged as a global tool for understanding workforce trends in the digital age (e.g., see Shron et al., 2025; Mahmud et al., 2024; Purohit et al., 2024). By applying AI tools to large datasets—such as online job postings, resumes, and professional profiles—these approaches can generate real-time labour market intelligence on in-demand skills and occupations (Colombo et al., 2019; Khaouja et al., 2021). Many countries and institutions are leveraging big data from job portals and social networks to monitor skill requirements and employment trends at scale, a task that traditional surveys often cannot accomplish with the same timeliness or granularity (Manroop et al., 2024; Ramasubbareddy et al., 2021; Nomura et al., 2017).

To enable cross-regional analyses, standardised frameworks for classifying jobs and skills are increasingly integrated into AI analytics. For example, the European Skills, Competences, Qualifications and Occupations (ESCO) framework and the U.S. O*NET database provide multilingual, structured taxonomies of occupations and skills, ensuring semantic consistency when parsing unstructured labour market data (Chiarello et al., 2021; Le Vrang et al., 2014). These frameworks serve as a foundation for mapping and comparing skill demand across different sectors and countries, thereby enhancing the global interoperability of labour market insights.

A key advantage of AI-assisted labour market analytics is its scalability and ability to process information in real time. Advanced systems can continuously crawl and analyse millions of job advertisements and other labour market signals, updating insights on emerging skills or declining occupations far more quickly than conventional labour market information systems (Tzimas et al., 2024). This scalability supports evidence-based decision-making in both industry and public policy.

Predictive analytics can forecast skills shortages or emerging competency needs, enabling policymakers and educational institutions to proactively adjust curricula and training programmes (Chiarello et al., 2021). In practice, this means aligning vocational education and higher education offerings with the evolving needs of the labour market—an integration that can improve graduates’ employability and address skill gaps on a national or even international scale. Indeed, pilot projects in Europe and elsewhere have demonstrated that AI-assisted analysis of job market data can inform the development of skills strategies and education policy, ensuring that workforce development initiatives are grounded in up-to-date market intelligence (Tzimas et al., 2024). Such analytics can also enhance career guidance and job matching: by linking granular skill data with job profiles, AI tools help identify transferable skills and alternate career pathways, thereby improving matching efficiency between job seekers and vacancies (Fareri et al., 2021). Overall, the infusion of AI into labour market analysis offers a more dynamic, fine-grained understanding of global labour trends, supporting a range of stakeholders from policymakers to employers in making data-driven decisions.

Notwithstanding its promise, AI-assisted labour market analytics faces several limitations. One major concern is data bias and representativeness. The big data sources typically used—online job postings, professional networking sites, etc.—may not fully represent the entire labour market, especially in regions or sectors where Internet presence is limited or where informal employment is prevalent. Consequently, analyses might be skewed towards high-demand urban or digital economy jobs, overlooking smaller industries or rural employment (Tzimas et al., 2024). A prominent example of this challenge is detailed in an OECD (2021) analysis of data from Burning Glass Technologies, a major commercial aggregator of online job vacancies. The report confirms that this data, while vast, is not representative of the whole economy. It significantly over-represents “high-skilled, white-collar occupations” (e.g., professionals) and under-represents “blue-collar,” public sector, and agricultural jobs (Cammeraat & Squicciarini, 2021, p. 13). Furthermore, the data reflects employer demand (a “list of desired skills”) rather than the actual skills required for a role, which can introduce its own bias into the analysis (Cammeraat & Squicciarini, 2021, p. 11). Moreover, biases present in historical data can be perpetuated by AI models: for example, if certain groups have been under-represented or stereotyped in past hiring practices, predictive algorithms might inadvertently reinforce those patterns (Khaouja et al., 2021).

Another significant challenge is skill ambiguity and semantic complexity in human language. Job advertisements often use inconsistent or vague terminology for skills and roles—for instance, one employer’s “client relations” might be equivalent to another’s “customer service,” and many soft skills are described in highly varied terms (Siekmann & Fowler, 2017; Panzaru & Grama, 2025). This lack of standardised language makes it difficult for algorithms to accurately extract and classify skills. AI systems rely on comprehensive dictionaries or ontologies, and while frameworks like ESCO aim to harmonise skill definitions, gaps remain. Studies have highlighted that soft skills in particular are inherently challenging to categorise and often go by many names, leading to discrepancies in automated skill identification (Panzaru & Grama, 2025). Even with sophisticated AI techniques, context is needed to disambiguate skill names and ensure they map correctly to standardised categories—a task that sometimes requires human expert validation alongside automated methods (Chiarello et al., 2021).

Also, various ethical challenges surround big data analytics (Manroop et al., 2024). For example, AI algorithms may show rampant racial biases (National Institute of Standards and Technology, 2020). These practices may result in the profiling of individuals based on protected characteristics such as age, race, gender, and socio-economic status. They also raise significant ethical concerns, particularly when data collected from social media platforms about jobseekers are used to predict employability, hiring patterns, turnover risks, and similar outcomes.

Finally, technical constraints such as the need for constant updating are non-trivial: the labour market continually evolves with new job titles and competencies (especially in fast-growing fields like AI itself), so analytical models and taxonomies must be regularly maintained to stay relevant (Fareri et al., 2021). If not carefully managed, these challenges—data bias, incomplete coverage, and semantic ambiguities—can limit the reliability of AI-assisted labour market insights.

2.3. Micro-Credentials as Agile Responses to Labour Market Needs

Micro-credentials have emerged as a pivotal mechanism for addressing the challenge of persistent and widening skills mismatch between the supply of higher education and the dynamic demands of the labour market, offering structured learning opportunities that are fundamentally agile and designed to increase institutional responsiveness and relevance (OECD, 2021, 2023b; Schutte & Kyriazi, 2025; Siafu, 2024; Shanahan & Organ, 2022; Varadarajan et al., 2023).

Micro-credentials are conceptualized as focused qualifications defined by the European Council as a “record of the learning outcomes that a learner has acquired following a short learning experience” (Council of the European Union, 2022, p. 5). Unlike traditional degrees, micro-credentials are shorter in duration, more targeted in subject matter, and flexible in delivery (OECD, 2024; Varadarajan et al., 2023). Their emergence gained significant traction in the aftermath of the COVID-19 pandemic, which amplified the demand for online, self-paced learning (Brown et al., 2021; Tamoliune et al., 2023). These small, targeted learning units effectively align education with the growing industry trend toward competency-based and skills-first hiring (OECD, 2024; Uzosike et al., 2025). They serve as tangible and verifiable evidence of acquired capabilities, often presented as digital badges (Oliver, 2019). To ensure trust, comparability, and recognition, the Council of the European Union (2022) specifies that micro-credentials must be built upon transparent, clearly defined standard elements, including learning outcomes, assessment methods, and workload. Key principles emphasise that they must be quality-assured, learner-owned, portable, and stackable toward larger qualifications (Council of the European Union, 2022).

Micro-credentials are highly valued by stakeholders for promoting flexible learning pathways and enhancing employability (Schutte & Kyriazi, 2025; Varadarajan et al., 2023). Their benefits can be understood from the perspectives of the three primary stakeholders (see

Table 1. Key benefits and quality conditions of micro-credentials for stakeholders (created by the authors based on literature overview).

Stakeholder	Core Benefits	Key Conditions for Value
Learners	Flexibility, accessibility, and affordability relative to full degrees; appeal to non-traditional learners and mid-career professionals (OECD, 2021; Yılık, 2025). Stackable credentials allow customised pathways and documentation of non-formal learning (Epaphras & Wachira, 2025; Tamoliune et al., 2023).	Credentials must be portable, stackable toward larger qualifications, and formally recognise competencies from co-curricular or informal contexts (Council of the European Union, 2022; Ward et al., 2023).
Employers	Just-in-time upskilling and reskilling channel; verified evidence of actual skills for skills-based hiring in fast-moving technical sectors (Bruguera et al., 2025; Gauthier, 2020; Uzosike et al., 2025). Major technology companies have established non-degree entry routes (OECD, 2023a).	Credentials must be externally benchmarked rather than self-certified; value concentrated in specific technical domains, not generic soft skills (Maina et al., 2022).
HEIs	Diversification of learning offer; agile response to fast-changing skill needs; support for modularisation of qualifications and pedagogical innovation through small-scale experimentation (Kato et al., 2020; OECD, 2021; Tamoliune et al., 2023).	Design must be informed by real-time labour market data and co-developed with employers; quality assurance must be transparent and aligned with NQF levels and ESCO taxonomies (ETF, 2023; OECD, 2023b; Uzosike et al., 2025).

).

A key opportunity for micro-credentials, given their focus on labour market relevance, is to have their design directly informed by current labour market needs, a function increasingly facilitated by AI-assisted analytics. The agility of micro-credentials is maximized when their content is directly informed by AI-assisted labour market analytics, establishing a crucial feedback loop between demand and provision (Uzosike et al., 2025). This data-driven approach is particularly suited to micro-credentials rather than full degrees, as traditional curriculum renewal is a slow, bureaucratic, and resource-intensive process (Yılık, 2025), whereas micro-credentials can be deployed in months to match “just-in-time” AI-identified trends (Siafu, 2024). This allows HEIs to use micro-credentials to address the granular, discrete skill gaps (Piróg & Hibszer, 2024). Furthermore, competency-based education platforms increasingly use AI to recommend personalized micro-learning pathways based on prior learning assessments, job market trends, and career aspirations, promoting individualized learning (Uzosike et al., 2025).

However, research draws attention to several aspects to be considered in this AI-assisted approach. An over-reliance on dynamic job market data could risk creating micro-credentials that can quickly become obsolete (Yılık, 2025). Furthermore, critics warn that “unbundling” education based on AI-identified keywords risks “fragmenting the curriculum” and divorcing skills from their deep, disciplinary context, which is essential for critical thinking (Wheelahan & Moodie, 2021). Therefore, it is suggested that the speed of AI-assisted design must be balanced against pedagogical rigour and robust quality assurance (Hou et al., 2024).

Despite the innovative potential of micro-credentials, their widespread adoption and value are conditional upon solving a crisis of trust through standardisation (Varadarajan et al., 2023). A key challenge is ensuring that flexible, digitally delivered micro-credentials maintain credibility. This requires a multi-pronged approach to building a trusted ecosystem. First, research indicates educational providers must collaborate actively with external stakeholders, including employers, trade unions, and professional bodies, for the co-design and validation of micro-credentials (ETF, 2023; OECD, 2023a). Second, micro-credentials must align with National Qualifications Frameworks (NQF) and utilise standardized skill taxonomies like ESCO to ensure consistency and portability (Council of the European Union, 2022; ETF, 2023; Uzosike et al., 2025). This involves clearly documenting learning outcomes, workload (e.g., ECTS credits), assessment protocols, and the quality assurance process used (Council of the European Union, 2022; Līce, 2021). Finally, as quality and trust are paramount, the development of transparent digital verification systems (like digital badges) is crucial for portability and authenticity, allowing learners to easily transfer and stack credentials across different providers and jurisdictions (McGreal et al., 2022; Uzosike et al., 2025).

3. Methodology

This study adopts a mixed-methods design integrating AI-assisted labour market analysis with ESCO-based skill mapping to examine the alignment between higher education programmes, professional standards, and labour market requirements. The methodological framework rests on the assumption that, in principle, no substantial discrepancies should exist between (1) competencies required in the labour market, (2) competencies defined in officially approved professional standards, and (3) competencies embedded in university programme learning outcomes. To test this assumption, the study operationalises a three-layer comparison model distinguishing between empirical demand, normative demand, and educational supply—three analytically distinct representations of the same underlying competence space, each produced by a different institutional actor and expressed through a different document type.

Within this framework, AI tools were applied in two analytically distinct stages: extraction of skill expressions from job advertisements, and identification of candidate matches to ESCO Level 3 (L3) categories. All final mapping and classification decisions were performed manually by the research team. This division between automated processing and expert validation is central to the study’s design: AI enables scalable analysis of large unstructured datasets that would be impractical to code manually, while expert oversight ensures conceptual accuracy, guards against misclassification, and preserves the interpretive rigour expected of mixed-methods research.

The framework integrates three interconnected layers—empirical labour market demand derived from job advertisements, normative demand defined by professional standards, and educational supply represented by programme learning outcomes—compared systematically at the ESCO L3 skill level. Comparisons are further refined by distinguishing between exact matches, composite matches, and true gaps, enabling a more nuanced and transparent assessment of alignment than strict one-to-one matching would permit. Quantitative alignment indicators are subsequently validated through structured expert interviews with programme directors and industry representatives, ensuring that data-driven results are not only methodologically sound but also contextually relevant and institutionally actionable.

The research methodology, including all data collection procedures, was approved by the Decision of the RTU Research Ethics Committee No. 04000-10.1-e/32 on 15 April 2026.

The overall process logic is illustrated in Figure 2; the main phases, analytical techniques, and outputs are summarised in

Table 2. Summary of research methodology (created by the authors).

Phase	Objective	Tools/Techniques Used	Output
1. Data Collection	Build a representative dataset of labour market needs in Latvia	Web scraping from CV Online and NVA using automated scripts	Raw dataset of ~30,000 job ads with job titles, descriptions, and metadata (research period: 1 June 2025–31 December 2025)
2. Preprocessing & Translation	Ensure linguistic consistency for analysis and ESCO mapping	DeepL neural translation, manual review of 5000 samples, segmentation for processing	English-translated, cleaned job ads with verified terminology
3. Occupational Classification (ESCO)	Classify jobs using a structured, auditable taxonomy	Claude (Anthropic LLM), Python 3.12 pipeline, JSON schema, reasoning fields, batch audit	ESCO-coded dataset with role type, seniority, and occupation data
4. Study Programme Learning Outcomes Analysis	Extract and structure educational competencies from programme learning outcomes	Manual review and standardisation of learning outcomes, ESCO L3 mapping	Supply profile (ESCO L3 skills)
5. Professional Standard Analysis	Extract and structure normative competency requirements	Manual analysis of professional standards, ESCO L3 mapping	Normative demand profile (ESCO L3 skills)
6. Skill Extraction & Standardisation	Identify and unify skill references across the dataset	AI-assisted manual classification, fuzzy-matching inheritance, SQLite skill mapping	Standardised list of ESCO Level 3 skills, free of duplication (Empirical demand profile)
7. Three-Layer Curriculum Comparison	Analyse alignment across empirical demand, normative demand, and educational supply	Review of RTU programme learning outcomes by using Claude, exact and composite matching algorithms; Jaccard similarity and coverage calculations	Skill gap analysis (representational vs. substantive) across all three analytical layers
8. Expert Validation	Contextualise and validate findings with stakeholder input	Structured interviews, directed thematic analysis, and triangulation	Qualitative insights from programme directors and employers
9. Framework Refinement	Translate findings into institutional guidance	Synthesis of all data, limitations analysis	Final AI-assisted framework and targeted micro-credential priorities

.

3.1. Data Sources and Sample

The empirical basis of the study consists of four data sources corresponding to three selected occupations: Personnel Specialist (PS), Finance Manager (FM), and Organisation Manager (OM). These occupations were selected purposively to satisfy four criteria simultaneously: (i) each is offered by Riga Technical University (RTU) with a corresponding study programme, ensuring that educational supply data are available; (ii) each is mapped to an officially approved professional standard developed under the Professional Education and Employment Tripartite Cooperation Sub-Council (PINTSA), enabling normative demand analysis; (iii) each is associated with a distinct ESCO occupational code, allowing structured cross-source comparison; and (iv) the three occupations span the three study levels relevant to higher education curriculum analysis—short-cycle, first-cycle, and second-cycle—enabling comparative observation across qualification levels. Beyond these criteria, the selected occupations represent high-demand roles in the Latvian labour market and reflect typical career progression within the business and management domain, capturing the transition from operational to strategic competencies.

Labour market demand was captured through automated web scraping of online job advertisements from CV Online and the National Employment Agency (NVA) portal, yielding a raw dataset of approximately 30,000 advertisements with job titles, descriptions, and metadata (research period: 1 June 2025–31 December 2025).

The full dataset was subsequently deduplicated and subjected to occupational classification via the AI-assisted ESCO pipeline described in Section 3.2, which assigned a 4-digit ESCO occupation code to each advertisement. The three focal occupations were then extracted from this coded dataset on the basis of the four purposive criteria described above, yielding the occupation-specific subsets reported in

Table 3. Mapping of selected occupations (ESCO classification) to RTU study programmes across study levels (created by the authors).

Study Level	Position	ESCO Code	Number of Job Ads	Study Program RTU	Qualification
Short-cycle professional studies	Human Resources Officer	2423.3	198	Entrepreneurship and management	Personnel Specialist
First-cycle (professional bachelor) studies	Finance Managers	1211	132	Entrepreneurship and management	Finance Manager
Second-cycle (professional master) studies	Managing Directors/Chief Executives	1120	124	Leadership and management	Organisation Manager

: 198 advertisements for Human Resources Officer (ESCO 2423.3), 132 for Finance Manager (ESCO 1211), and 124 for Managing Director/Chief Executive (ESCO 1120). Advertisements that could not be reliably assigned to an ESCO code, or that were flagged as title-only classifications (i.e., no job description text was available), were excluded from the skill extraction analysis.

To ensure linguistic consistency for ESCO mapping, all advertisements were translated from Latvian into English using DeepL neural translation, followed by manual review of a 5000-sample subset to verify terminological accuracy and segment texts for further processing. Normative demand was derived from professional standards approved by PINTSA, which define the expected competence profiles for each occupation. Educational supply was represented by the Programme Learning Outcomes of three corresponding study programmes at RTU, extracted and standardised through manual review. The findings from these three quantitative sources were validated and contextualised through structured interviews with six experts comprising programme directors and industry representatives, whose recruitment and interview procedure are described in Section 3.3.

3.2. AI-Assisted Skill Extraction, ESCO Mapping, and Alignment Metrics

In the interest of full reproducibility, the complete prompt templates, JSON output schemas, quality-control logic, and workflow documentation for all AI-assisted steps described in this section are provided in Supplementary Materials (DeepL, 2025). The present section summarises the pipeline structure and design rationale; Supplementary Materials should be consulted for the verbatim prompts and technical specifications.

An AI-assisted text analysis pipeline was applied to extracted skill-related information from unstructured job advertisements. Specifically, Claude (Anthropic LLM) was used within a Python 3.12 pipeline incorporating a JSON schema with reasoning fields and batch audit functionality to classify occupations and identify candidate skill expressions at the ESCO L3 level. The pipeline employed the following prompt structure: each job advertisement text was submitted with an instruction to identify expressed competencies and map them to the closest ESCO L3 descriptor, returning the result as a structured JSON object with fields for the extracted skill phrase, the candidate ESCO L3 label, the ESCO code, a confidence indicator, and a reasoning field explaining the mapping rationale. This reasoning field was used during manual audit to verify and, where necessary, correct AI-generated candidate matches. Extracted skill expressions were subsequently standardised and cleaned to eliminate duplication, semantic variation, and non-skill elements—such as generic role descriptors or organisational requirements—using fuzzy-matching inheritance and an SQLite-based skill mapping database. This step produced a coherent empirical demand profile expressed in standardised ESCO L3 terms.

The selection of ESCO as the unifying semantic framework for this study was made on the basis of four criteria, evaluated against the principal alternatives—O*NET and SFIA. O*NET (Occupational Information Network), developed by the US Department of Labor, provides a detailed taxonomy of occupational requirements widely used in English-language labour market research (Handel, 2016); however, its coverage of European occupational structures and its multilingual applicability are limited, making it poorly suited to a Latvian labour market context where job advertisements, professional standards, and programme documentation are produced in Latvian and mapped to European qualification frameworks. SFIA (Skills Framework for the Information Age) offers a structured competence taxonomy specifically designed for digital and information technology roles (SFIA Foundation, 2021); while its granularity is valuable for ICT-focused studies, its scope is by design sector-specific and does not cover the broad business, management, and HR occupational profiles central to the present study. ESCO, by contrast, satisfies all four selection criteria simultaneously.

First, it provides multilingual coverage—currently available in 27 languages—enabling consistent skill mapping across Latvian-language source documents and English-language analytical outputs without loss of conceptual integrity.

Second, it is explicitly designed for semantic interoperability between heterogeneous data sources, making it appropriate for a study that compares three institutionally distinct document types—job advertisements, professional standards, and programme learning outcomes—that represent the same underlying competence space through different vocabularies and at different levels of abstraction (Le Vrang et al., 2014).

Third, it is the reference taxonomy mandated by the European Commission for cross-border skills intelligence and is directly integrated into Latvian professional standard development through the PINTSA framework, ensuring alignment between the analytical layer and the regulatory layer of the study.

Fourth, its hierarchical three-level structure (L1–L3) supports the composite matching procedure central to this study, allowing higher-level competence expressions in job advertisements to be decomposed into their constituent L3 descriptors within a single, internally consistent taxonomy. On this basis, ESCO was selected as the most appropriate framework for the present study’s analytical context, data sources, and geographic scope. Benchmarking the composite matching results against alternative taxonomies such as O*NET is identified as a direction for future comparative research. The complete prompt templates, output schemas, and pipeline documentation supporting the AI-assisted classification and skill mapping procedures are provided in Supplementary Materials.

All competencies—derived from job advertisements, professional standards, and programme learning outcomes—were mapped to the ESCO framework at Level 3, which served as a unifying semantic layer enabling direct comparison between heterogeneous data sources and ensuring conceptual consistency across the three analytical layers. This use of ESCO is consistent with its design rationale as a framework for semantic interoperability rather than strict lexical one-to-one matching (Le Vrang et al., 2014). That rationale is particularly relevant here because online job advertisements systematically combine skill formulations of different granularity—from broad, category-level expressions such as “management skills” or “working with computers,” which correspond to ESCO Level 1 areas rather than discrete L3 competencies, to highly specific operational tasks—whereas professional standards and programme learning outcomes are typically formalised at a more uniform level of abstraction (Cammeraat & Squicciarini, 2021; Khaouja et al., 2021). For each occupation, the mapping process produced three comparable skill profiles: an empirical demand profile derived from job advertisements, a normative demand profile derived from professional standards, and an educational supply profile derived from programme learning outcomes.

To assess alignment between skill profiles, the following indicators were calculated for each pairwise comparison: Supply (unique L3 skills in the educational profile), Demand (unique L3 skills in the comparator profile), Intersection (skills present in both), Union (skills present in either), Coverage (Intersection/Demand), Jaccard index (Intersection/Union), Gap (skills in demand absent from supply), and Oversupply (skills in supply absent from demand). These metrics allow for both similarity measurement and structural analysis of mismatches. Additionally, to identify broader structural patterns, L3 skills were aggregated into ESCO Level 1 (L1) categories for descriptive analysis of competency distribution across major skill areas—including management, information, communication, and digital competencies—though L1 aggregation was not used for the calculation of alignment metrics.

To address the limitations of strict one-to-one ESCO L3 matching, a composite matching approach was applied. The ESCO ontology itself acknowledges the existence of semantically overlapping skills through the concept of “OverlappingSkillGroup,” defined as collections of skill concepts with overlapping semantics (European Commission, 2020), and empirical studies confirm that strict lexical matching provides a limited measure of correspondence between unstructured labour market data and structured taxonomies (Chiarello et al., 2021; Fareri et al., 2021; Khaouja et al., 2021). Composite matching allows a skill identified in one data source to be considered covered when it corresponds functionally to one or more ESCO L3 skills in the compared profile, addressing the structural asymmetry in skill granularity described above by using ESCO’s built-in hierarchy (L1–L2–L3) to link higher-level expressions to the specific L3 competencies they encompass.

The procedure followed three steps.

First, all skills were standardised and mapped to ESCO L3 categories to ensure semantic consistency.

Second, potential composite relationships were identified based on semantic similarity and functional equivalence—for example, the broad expression “processing information” in a job advertisement was linked to the more specific ESCO L3 skills “managing information” and “analysing and evaluating information and data.”

Third, candidate composite matches identified by AI-assisted semantic analysis were reviewed independently by two members of the research team, each assessing whether the proposed functional equivalence relationship was conceptually defensible or constituted a false equivalence. Disagreements were resolved through structured discussion; where consensus could not be reached, the match was excluded from the effective intersection and retained in the unresolved demand gap. No disagreements required adjudication by a third party.

To illustrate the resolution process: one disputed composite match involved the job-advertisement expression ‘financial reporting’ and the ESCO L3 skill ‘preparing financial statements’. One reviewer considered these functionally equivalent; the other noted that reporting encompasses a broader set of communication acts than preparation alone. Following the discussion, the match was excluded as ambiguous and retained in the unresolved demand gap, consistent with the conservative approach applied throughout.

No formal inter-rater reliability coefficient was calculated, as assessing functional equivalence between competence expressions across institutionally distinct document types constitutes an expert interpretive judgment rather than a closed-category coding task; the conservative exclusion of all partial and ambiguous matches from effective alignment calculations serves as a structural safeguard against the subjective inflation of alignment scores.

Composite matches were classified into three categories: (1) exact matches (direct equivalence between standardised ESCO skill concepts), (2) validated composite matches (functional equivalence across multiple skills confirmed through expert review), and (3) partial or ambiguous matches (excluded from the effective intersection to maintain conservative estimates). Only exact and validated composite matches were included in the corrected coverage and effective Jaccard calculations; partial or ambiguous matches were retained within the unresolved demand gap. Compared to related approaches—including AI-based semantic similarity scoring (Andonovikj et al., 2024) and statistical clustering of skill terms (Piróg & Hibszer, 2024)—the present procedure adopts a more conservative alignment strategy by retaining ESCO L3 as a fixed reference taxonomy and not counting partial matches as effective alignment.

3.3. Validity, Reliability, and Qualitative Validation

Several steps were taken to ensure the validity and reliability of the analysis. The use of ESCO as a standardised taxonomy ensured semantic consistency across heterogeneous data sources. AI-assisted extraction was combined with data cleaning and manual standardisation procedures to reduce noise and ambiguity in skill identification. The use of multiple complementary indicators—Coverage, Jaccard index, Gap, and Oversupply—provided a robust, multi-perspective assessment of alignment rather than reliance on any single metric. Ethical approval for the study was obtained prior to data collection; all interview participants provided informed consent, and all data were anonymised before analysis.

To triangulate the quantitative findings and ensure their contextual relevance, the study incorporated a qualitative validation phase through structured expert interviews. Participants were purposefully sampled from two stakeholder groups: programme directors from RTU responsible for the development and quality assurance of the three analysed study programmes, and industry representatives from sectors with high demand for the selected occupations. Specifically, two programme directors and four industry representatives in senior managerial or HR roles were recruited through institutional networks and professional contacts associated with RTU and its industry advisory bodies. PD1 was responsible for two of the three analysed programmes (Personnel Specialist and Finance Manager, both within the “Entrepreneurship and Management” programme), while PD2 was responsible for the Organisation Manager programme (“Leadership and Management”). Industry representatives were senior managers and HR professionals with 11 to 29 years of experience in sectors directly corresponding to the three selected occupations. A sample of six participants was considered appropriate for the study’s purpose of contextualising and qualifying quantitative findings rather than independently validating the broader national ecosystem; the sample was not designed to be statistically representative but to provide expert insight from the two key stakeholder groups directly engaged with curriculum design and graduate employment.

To consolidate the transparency of the mixed-methods design, the AI model functioned exclusively as an execution tool under a fixed, human-designed protocol at each stage. Occupational classification decisions, composite match validations, and thematic codes were all subject to independent review by two researchers, with disagreements resolved through structured discussion. No classification, match, or code was accepted on the basis of AI output alone without human confirmation. Full prompt documentation is provided in Supplementary Materials.

Coding was conducted at the passage level: each meaningful unit of participant speech addressing one of the five pre-defined themes was coded independently by two researchers working from verbatim transcripts. Researchers coded all six transcripts separately, then compared their codes in a structured reconciliation session. Disagreements—primarily in cases where a passage addressed more than one theme simultaneously, or where content did not fit the a priori framework—were resolved through discussion until consensus was reached; no cases required adjudication by a third party. The three emergent cross-cutting themes (T6–T8) arose from a second inductive pass over codes that had not been accommodated by the a priori framework: these codes were grouped by conceptual similarity and confirmed as themes when the pattern was independently present across at least three participants from different stakeholder groups. Representative quotations were selected on the criterion of illustrative clarity—the passage most directly and succinctly expressing the coded theme—with preference given to quotations corroborated by multiple speakers or representing a divergent view that qualified the quantitative findings.

All six interviews were conducted online between 8 and 10 April 2026, lasting between 20 and 30 min. All sessions were recorded with participants’ consent, transcribed verbatim, and anonymised prior to analysis.

Interview transcripts were analysed using directed (deductive) thematic analysis following the procedure described by Braun and Clarke (2006). Five themes were defined a priori based on the interview protocol and the study’s objectives: (T1) current alignment mechanisms between universities and industry, (T2) perceived skills gaps in graduate profiles, (T3) the validity of job advertisement data as a proxy for labour market demand, (T4) institutional agility and barriers to curriculum change, and (T5) attitudes toward micro-credentials. The interview protocol consisted of eight to ten semi-structured questions per theme, with follow-up probes used to elicit elaboration. Each transcript was independently coded by one researcher and reviewed by a second, with disagreements resolved through discussion. The analysis additionally revealed three emergent cross-cutting themes not defined in advance: (T6) a pedagogical transfer gap—the discrepancy between competencies formally present in curricula and those effectively demonstrated by graduates in professional practice; (T7) demand profile limitations—the constraints of job advertisements as a complete representation of employer requirements; and (T8) learner characteristics as a confounding variable, particularly relevant to the Organisation Manager programme whose students are predominantly practising managers. Responses were compared within and across stakeholder groups and across the three occupations to identify convergences, divergences, and occupation-specific patterns, and the resulting insights were mapped against the quantitative skills gap findings to identify areas of confirmation and qualification.

4. Results

4.1. Quantitative Findings from ESCO-Based Alignment Analysis

The study focuses on three occupations spanning different qualification levels. Their ESCO classification, corresponding job advertisement data, and linked RTU study programmes are presented in Table 3. Throughout the analysis, ESCO occupational titles are used for labour market data, while Latvian professional standard terminology is retained for study programmes: Human Resources Officer (ESCO 2423.3) aligns with Personnel Specialist, Finance Manager (ESCO 1211) with Finance Manager, and Managing Director/Chief Executive (ESCO 1120) with Organisation Manager. The analysis was conducted on the full set of extracted skills without prior reduction or exclusion of heterogeneous or overlapping skill expressions, in order to preserve the real-world structure of job advertisements. Subsequent standardisation, ESCO-based mapping, and composite matching were applied at the analytical stage to ensure comparability while retaining this underlying complexity. In each of the three pairwise comparisons that follow, the term empirical demand refers to the unique ESCO L3 skills extracted from job advertisements, normative demand refers to the unique ESCO L3 skills derived from professional standards, and supply refers consistently to the unique ESCO L3 skills derived from study programme learning outcomes (PLOs).

4.1.1. Alignment Between Labour Market Demand and Study Programme Learning Outcomes

A large-scale dataset of labour market demand in Latvia was constructed through automated web scraping of job advertisements, from which empirical demand profiles were developed for each of the three selected occupations. Study programme learning outcomes were independently analysed and mapped to ESCO skills, producing supply profiles for structured comparison.

Table 4. Alignment indicators between labour market demand and study programme learning outcomes across the three analysed occupations (authors’ calculations).

Alignment Indicator	Human Resources Officer/Personnel Specialist (ESCO 2423.3)	Finance Manager (ESCO 1211)	Managing Director/Chief Executive/Organisation Manager (ESCO 1120)
Supply (unique L3 from learning outcomes)	53	54	37
Demand (unique L3 from job advertisements)	96	78	85
Intersection (exact matches)	48	41	34
Union	101	91	88
Coverage (exact)	50%	53%	40%
Jaccard index (exact)	0.48	0.45	0.39
Oversupply (supplied but not demanded)	5	13	3
Initial gap (demanded but not directly matched)	48	37	51
Gaps resolved through composite matching	22	20	38
Partial/ambiguous gaps (excluded)	10	7	9
True gaps (unmet demand after composite matching)	16	10	4
Effective intersection (exact + composite)	70	61	72
Effective gap	26	17	13
Corrected coverage (composite)	73%	78%	85%
Corrected Jaccard index (composite)	0.69	0.67	0.82

presents alignment indicators based on both exact and composite matching across the three occupations.

Under exact one-to-one matching, initial coverage ranges from 40% (Organisation Manager) to 53% (Finance Manager), and Jaccard similarity indices from 0.39 to 0.48—indicating moderate formal correspondence across all three programmes. The initial gap figures (48, 37, and 51 demanded skills not directly matched) might suggest substantial curriculum deficiencies; however, qualitative classification of these gaps reveals that not all represent genuine competence absences. A share of unmatched items consists of overly generic or non-standardised job advertisement formulations—such as “management skills”—that do not correspond to discrete ESCO L3 competencies and should be treated as artefacts of job advertisement language rather than substantive curriculum gaps. A further share reflects behavioural attributes or terminological differences rather than formal competencies, and some items are context-specific requirements not universally applicable across roles. The Organisation Manager case illustrates this most clearly: “using foreign languages” appears as a gap under strict matching, yet language acquisition is embedded in the curriculum and simply not articulated as a discrete ESCO L3 outcome.

After applying composite matching—which allows a skill expressed at a higher level of aggregation to be considered covered when it corresponds functionally to one or more ESCO L3 skills in the supply profile—coverage increases to 73% (Personnel Specialist), 78% (Finance Manager), and 85% (Organisation Manager), with Jaccard indices rising to 0.69, 0.67, and 0.82, respectively. The increase is most pronounced for the Organisation Manager programme, where job advertisements tend to express competencies in fragmented and integrative forms that are particularly ill-suited to strict one-to-one matching. Oversupply remains low across all three cases (3–13 skills), indicating that the programmes provide focused competence profiles without excessive redundancy. The nature of the remaining true gaps varies by qualification level: at the short-cycle level, residual gaps are more concrete and operational; at the first-cycle level, they are more specialised but often implicitly covered by broader learning outcomes; and at the second-cycle level, the true gap is minimal and largely attributable to highly contextual or externally acquired competencies such as language proficiency or niche regulatory knowledge.

4.1.2. Alignment Between Professional Standards and Study Programme Learning Outcomes

Professional standards approved by PINTSA were analysed and mapped to ESCO Level 3 skills to construct a normative competence profile for each occupation. Figure 3, Figure 4 and Figure 5 present the gap and oversupply profiles for the Personnel Specialist, Finance Manager, and Organisation Manager programmes, respectively. Across all three cases, identified gaps are concentrated in management and information-related competencies, while oversupply is most visible in communication skills—a structurally consistent pattern across study levels despite differences in absolute magnitude.

Table 5. Comparison of three RTU study programmes (ESCO-based analysis) (created by the authors based on calculations).

Alignment Indicator	Short-Cycle SP, Personnel Specialist	First-Cycle SP, Finance Manager	Second-Cycle SP, Organisation Manager
Supply (unique L3)	53	54	36
Demand (L3 from PQR)	43	48	39
Intersection	26	26	27
Union	70	76	48
Coverage	60%	54%	69%
Jaccard index	0.37	0.34	0.56
Gap (unmet demand for skills)	17	22	12
Oversupply (excess skills)	27	28	9

Note: Coverage = Intersection/Demand; Jaccard index = Intersection/Union (a measure of overall similarity between supply and demand skill sets, ranging from 0 to 1); Gap = skills present in demand but absent from supply; Oversupply = skills present in supply but absent from demand.

summarises the alignment indicators across the three programmes. Coverage ranges from 54% (Finance Manager) to 69% (Organisation Manager), with the Personnel Specialist at an intermediate 60%. The second-cycle Organisation Manager programme demonstrates the strongest overall correspondence (Jaccard 0.56), while the first-cycle Finance Manager programme exhibits the weakest alignment (Jaccard 0.34) and the largest gap (22 skills). Importantly, the discrepancies are structural rather than merely quantitative: in all three cases, the simultaneous presence of unmet demand and oversupply indicates that programme learning outcomes do not fully mirror the competence profiles defined in professional standards. The pattern is particularly evident in the Finance Manager and Personnel Specialist programmes, where oversupply remains high despite low intersection. These results indicate that the existence of a professional standard does not automatically ensure close correspondence with educational supply; the translation of formally defined competence requirements into programme-level learning outcomes is partial and uneven across study levels.

Table 6. Distribution of ESCO L1 skill areas across study programmes (percentage share) (created by the authors based on calculations).

L1 Area	Short-Cycle SP, Personnel Specialist	First-Cycle SP, Finance Manager	Second-Cycle SP, Organisation Manager
Assisting and caring	5.66%	5.56%	5.56%
Communication	32.08%	29.63%	33.33%
Information skills	32.08%	33.33%	27.78%
Management skills	24.53%	25.93%	30.56%
Working with computers	5.66%	5.56%	2.78%

presents the distribution of ESCO L1 skill areas across the three supply profiles. The overall structure is broadly consistent: communication, information, and management skills together account for the largest shares across all programmes, each representing approximately one quarter to one third of the total skill set. The short-cycle and first-cycle programmes show near-identical L1 distributions, while the second-cycle programme displays a slightly higher share of management skills (30.56%), a lower share of information skills (27.78%), and the lowest share of digital competencies (2.78%). Digital competencies account for only 5.66% and 5.56% of supply-side L3 skills in the short-cycle and first-cycle programmes, respectively—a consistently low figure that is noteworthy given the centrality of digital skills in the current labour market.

4.1.3. Alignment Between Labour Market Demand and Professional Standards

In the final pairwise comparison, the skills identified in job advertisements are compared with the corresponding professional standards.

Table 7. Alignment between labour market demand (job advertisements) and professional standards based on ESCO L3 skills across study levels (created by the authors based on calculations).

Alignment Indicator	2423.3 (Personnel Specialist)	1211 (Finance Manager)	1120 (Organisation Manager)
Unique L3 (Job Ads)	96	78	85
Unique L3 (Professional Standard)	43	48	39
Exact Intersection	15	24	28
Union	124	102	96
Standard Coverage (%)	35%	50%	72%
Jaccard Index	0.12	0.24	0.29
Only in Professional Standard	28	24	11
Only in Job Advertisements	81	54	57

Note: Standard coverage is calculated as the proportion of professional standard skills represented in job advertisements (Intersection/Professional standard). The Jaccard index is calculated as the ratio of the intersection to the union of the two skill sets.

presents the initial exact-matching results, which reveal a relatively low level of direct alignment at lower qualification levels, consistent with the structural asymmetry noted in Section 4.1.1: professional standards represent competencies in a more formalised and aggregated manner than the heterogeneous, granular skill expressions found in job advertisements.

After applying composite matching (

Table 8. Composite-adjusted alignment between labour market demand and professional standards based on ESCO L3 skills across study levels (created by the authors based on calculations).

Alignment Indicator	2423.3 (Personnel Specialist)	1211 (Finance Manager)	1120 (Organisation Manager)
Effective intersection	38	31	32
Standard coverage (%)	88%	65%	82%
Jaccard index	0.38	0.33	0.35

Note: Effective intersection includes both exact and composite matches, where a skill identified in job advertisements is considered covered if it corresponds to one or more ESCO L3 skills in the professional standard. The union remains unchanged, as it represents the total set of unique skills across both datasets. Corrected Jaccard index is calculated using the effective intersection.

), alignment increases substantially across all three occupations. For the Personnel Specialist, the effective intersection rises from 15 to 38 skills, increasing coverage from 35% to 88%. Coverage for the Finance Manager and Organisation Manager occupations increases to 65% and 82%, respectively. This confirms that a significant share of the initially identified mismatch between job advertisements and professional standards reflects differences in skill granularity and aggregation rather than genuine competence absence.

4.1.4. Synthesis of Alignment Across Labour Market Demand, Study Programme Outcomes, and Professional Standards

Table 9. Synthesis of alignment indicators across the three pairwise comparisons (Coverage, %) (created by the authors based on calculations).

Pairwise Comparison (Exact/Composite Matching)	Personnel Specialist	Finance Manager	Organisation Manager
Labour market ↔ Study programme learning outcomes	50%/73%	53%/78%	40%/85%
Labour market ↔ Professional standards	35%/88%	50%/65%	72%/82%
Professional standards ↔ Study programme learning outcomes	60%/—	54%/—	69%/—

Note: Coverage indicates the proportion of demanded skills represented in the corresponding source. The dash (—) indicates that composite matching was not applied to the professional standards ↔ Study programme learning outcomes comparison in the present analysis; this represents a methodological limitation discussed in the next section. Coverage values for that comparison (60%, 54%, 69%) reflect strict one-to-one ESCO L3 matching only and may therefore underestimate the true level of alignment, consistent with the patterns observed in the two other comparisons.

synthesises the key alignment indicators across all three pairwise comparisons and the three analysed occupations.

Three consistent patterns emerge from the synthesis. First, strict one-to-one ESCO L3 matching systematically underestimates the true level of alignment in all comparisons where composite matching was applied, with initial coverage figures driven substantially by differences in skill granularity and representation rather than substantive competence absence. Second, composite matching increases observed alignment markedly across all layers: coverage rises from 50%, 53%, and 40% to 73%, 78%, and 85% in the labour market demand versus supply comparison, and from 35%, 46%, and 39% to 88%, 65%, and 82% in the labour market demand versus professional standards comparison. A consistent application of composite matching to the professional standards versus programme learning outcomes comparison—not yet performed in the present study—is expected to yield a similar increase, given the structural asymmetry between those two sources. Third, the three domains differ systematically in how they represent competencies: job advertisements express skills in granular, task-oriented, and often overlapping terms; programme learning outcomes provide a structured and operationalised representation; and professional standards adopt an abstract, aggregated competence-based formulation. This structural differentiation accounts for a large share of the initially observed gaps and is most pronounced at lower qualification levels, where the asymmetry between operational job advertisement language and generalised professional standard formulations is greatest. The one finding that persists across all comparisons, even after composite matching, is the consistent under-representation of digital competencies in programme learning outcomes (2.78–5.66% across all three programmes), indicating a substantive rather than representational gap warranting targeted curriculum action.

4.2. Qualitative Findings from Expert Interviews

Six structured interviews were conducted with two programme directors (PD1, PD2) and four industry Representatives (IR1–IR4) to contextualise and validate the quantitative results.

Table 10. Summary of qualitative findings from expert interviews by thematic area (created by the authors).

Theme	Key Convergence	Key Divergence
T1. Current alignment mechanisms	Both groups value industry involvement and recognise the role of existing formal mechanisms such as professional standard development, advisory groups, and state examination commissions.	Programme directors emphasise existing structures as sufficient; industry representatives find current engagement inadequate and call for more structured, regular, and practice-oriented collaboration formats (e.g., hackathons, IR1; greater involvement in programme design, IR2).
T2. Perceived skills gaps	Both groups agree that adaptability and willingness to learn matter more than specific technical knowledge at entry level.	Industry perceives significantly larger and more specific gaps—particularly in digital and applied competencies—than programme directors. PD2’s assessment of no significant gaps represents a disconfirming case, potentially explained by the atypical student profile of the OM programme (practising managers).
T3. Validity of job advertisement data	Job advertisements capture hard and technical skills better than soft skills, attitudes, or context-specific requirements; both groups treat them as a useful but imperfect signal.	Industry respondents identify specific distortion mechanisms: minimal-information advertisements understate requirements (IR1), while ‘ideal employee’ listings overstate them by combining multiple roles (IR1, IR2). Programme directors rely primarily on other signals (advisory groups, direct employer contact) and are less engaged with this limitation.
T4. Institutional agility and barriers	Both groups identify accreditation bureaucracy and methodological commission processes as barriers to timely curriculum updates.	Programme directors regard informal workarounds—adapting course content without formal revision—as adequate compensation. Industry respondents identify a more fundamental structural problem: the study process itself does not always require sufficient student commitment (IR4), pointing to a pedagogical transfer gap beyond curriculum content.
T5. Attitudes toward micro-credentials	All respondents agree that micro-credentials require clearly defined quality standards and measurable outcomes to be credible. Micro-credentials for specific, well-defined technical competencies (e.g., labour law, data analysis, AI tools) are viewed as potentially valuable.	PD2 opposes micro-credentials as fragmenting the coherence of study programmes; industry conditionally supports them. Key nuance: industry views micro-credentials as continuing professional development instruments for existing specialists, not as entry-level substitutes for degree-level education (IR3). Generic soft-skill micro-credentials are not valued (IR4).

summarises the key convergences and divergences across the five pre-defined thematic areas.

Regarding perceived skills gaps (T2), the divergence between stakeholder groups is particularly striking. While both groups agree that adaptability and willingness to learn matter more than specific technical knowledge at entry level, industry respondents identify significantly larger and more specific competence deficits than programme directors—particularly in digital and applied skills. IR2 reported that Finance Manager graduates effectively need to be retrained from the start, citing specific deficits in applied data skills and the ability to distinguish management from financial accounting. IR3 and IR4 identified high demand for AI tools, automation, and strategic HR thinking in the Personnel Specialist profile. PD2’s assessment that no significant gaps exist in the Organisation Manager programme represents a disconfirming case, plausibly explained by the atypical student profile of practising managers whose prior professional experience compensates for formal curriculum gaps. On the validity of job advertisement data (T3), both stakeholder groups treat them as useful but imperfect signals: IR1 estimated that advertisements reflect only 60–80% of actual job duties, while IR2 noted that listings often describe standard requirements rather than the actual role. Regarding micro-credentials (T5), acceptance is conditional and differentiated across respondents. Industry representatives endorse micro-credentials only when they target specific, well-defined technical competencies, meet externally benchmarked quality standards, and serve as continuing professional development tools for existing specialists rather than substitutes for degree-level education. PD2’s strong scepticism—arguing that micro-credentials cannot eliminate skill gaps and risk fragmenting the coherence of programmes—represents an important institutional counterpoint.

Beyond the five pre-defined themes, three cross-cutting observations emerged inductively. First, multiple industry respondents independently identified a pedagogical transfer gap (T6): competencies formally present in curricula are not always effectively acquired or applied by graduates in professional contexts. IR1 noted that graduates “are afraid to come forward with proposals and initiatives,” and IR4 argued that the study process itself does not consistently demand full student commitment—suggesting that the gap lies not only in curricular content but in how competencies are taught, practised, and assessed. Second, the interviews reinforced the limitations of job advertisement data as a complete proxy for employer demand (T7), consistent with the quantitative findings in Section 4.1. Third, the atypical learner profile in the Organisation Manager programme (T8)—where students are practising managers—highlights that programme-level gap analyses should account for actual student characteristics, as prior professional experience may compensate for formally unaddressed curriculum gaps.

4.3. Triangulation

Mapping the qualitative findings against the quantitative results reveals three areas of convergence and one important qualification. The low representation of digital competencies across all programmes (2.78–5.66% of supply-side L3 skills; Table 6) is independently corroborated by industry interviews, in which AI, data analysis, Power BI, and process automation were consistently identified as critical skill deficits across all three occupational profiles (IR2, IR3, and IR4). PD2’s observation that AI-related competencies are embedded across courses rather than articulated as discrete learning outcomes suggests that part of the observed under-representation is a visibility gap—a documentation failure rather than a teaching failure—though this does not fully account for the industry-reported deficit in graduates’ applied digital skills. The highest quantitative alignment (85% composite coverage, Organisation Manager) corresponds to PD2’s assessment of no significant curriculum–employer gap, while the lower alignment figures for Finance Manager (78%) and Personnel Specialist (73%) align with the more pronounced practical skill deficits reported by IR2, IR3, and IR4. Finally, the qualitative findings qualify the quantitative demand profile: since job advertisements may reflect only 60–80% of actual duties or may overstate requirements by combining expectations from multiple roles, the measured gap figures should be interpreted as approximations rather than precise deficit measures, reinforcing the need to read quantitative alignment indicators alongside stakeholder evidence.

5. Discussion

The primary contribution of this study is the development and proof-of-concept validation of a three-layer analytical framework that integrates empirical demand from job advertisements, normative demand from professional standards, and educational supply from programme learning outcomes within a unified ESCO-based taxonomy. While prior studies have applied AI and ESCO-based approaches to the alignment of labour market demand with academic curricula (Andonovikj et al., 2024; Faisal et al., 2025), these models are typically limited to a two-layer demand–supply comparison. The present framework extends this by introducing professional standards as a third analytical layer, enabling simultaneous diagnostic assessment across three institutionally distinct representations of the same underlying competence space. A further methodological contribution lies in the introduction of composite matching alongside exact matching, which produces a transparent and defensible range of alignment estimates—exact matching as a conservative lower bound, composite matching as a more contextually accurate upper bound—rather than a single point estimate that conceals the interpretive choices embedded in skill comparison. Compared to related approaches, including AI-based semantic similarity scoring (Andonovikj et al., 2024) and statistical clustering of skill terms (Piróg & Hibszer, 2024), the present procedure adopts a more conservative strategy by retaining ESCO L3 as a fixed reference taxonomy and excluding partial or ambiguous matches from effective alignment calculations. The novelty of the framework therefore lies not in any single technique but in the combination of three-layer architecture, dual matching strategy, and integrated qualitative validation—a combination that enables the framework to function as a diagnostic instrument for institutional dialogue rather than a black-box verdict on programme adequacy. Critically, neither Andonovikj et al. (2024) nor Piróg and Hibszer (2024)—the two most directly comparable studies—incorporate professional standards as an analytical layer: both compare labour market demand with educational supply only, leaving the normative regulatory dimension entirely outside the diagnostic. By introducing professional standards as a third, institutionally grounded comparator, the present framework can distinguish not only whether a programme covers what employers currently demand, but whether it meets the formally approved occupational competence profile that governs qualification recognition—a distinction with direct implications for accreditation, curriculum governance, and the credibility of gap diagnoses as a basis for institutional action.

The empirical findings refine the prevailing narrative on skills mismatch by demonstrating that a substantial share of apparent misalignment between education and labour market demand is attributable to differences in how competencies are represented, aggregated, and expressed across institutional domains rather than to their substantive absence. While earlier research has documented persistent structural gaps between education and employment (McGuinness et al., 2018; Comyn, 2019), the present analysis shows that measurement approach itself shapes how mismatch is diagnosed. This has a direct methodological implication: studies that rely exclusively on strict one-to-one taxonomic matching are likely to overstate curriculum deficiencies, particularly at lower qualification levels where the asymmetry between granular job advertisement language and aggregated programme outcomes is most pronounced. The substantial alignment increases observed after-composite matching—from 40–53% to 73–85% in the labour market demand versus supply comparison—empirically demonstrate the magnitude of this overestimation. At the same time, the results do not suggest that alignment problems are illusory. The analysis identifies selective and uneven translation of competence requirements across the three analytical layers, consistent with the literature describing the slow adaptation of education systems to labour market change (Brunello & Wruuck, 2021; Penesis et al., 2017). Professional standards, in particular, appear to function primarily as stabilising normative reference points rather than dynamic reflections of real-time employer demand, introducing a temporal and semantic lag between empirical skill needs and formal competence definitions. The progression toward stronger alignment at higher qualification levels—where competencies become more integrative and strategic—corresponds with established theoretical perspectives on competence development (Geng, 2025) and is consistent with the atypical student profile of the second-cycle Organisation Manager programme, whose practising-manager students may compensate for formal curriculum gaps through professional experience.

The finding that digital competencies are persistently under-represented in programme learning outcomes (2.78–5.66%)—independently corroborated by all four industry respondents—is the most policy-relevant result of the study. This is one of the few patterns that survives composite matching and thus represents a substantive rather than representational gap. It aligns with broader international evidence that digital transformation is outpacing curriculum adaptation in higher education (Kopackova et al., 2024; Geng, 2025). The interview evidence introduces an important nuance: PD2 reported that AI-related skills are taught across courses rather than designated as discrete learning outcomes, suggesting that some of the measured under-representation reflects a documentation gap—a failure to make tacit teaching practice visible in formal outcome statements—rather than entirely missing instruction. Addressing this gap therefore requires both new content and more explicit and transparent competence formulation in programme documentation.

The pedagogical transfer gap identified inductively through the interviews represents a conceptual contribution that extends the scope of curriculum alignment debates. Industry respondents consistently reported that graduates possess formal knowledge but lack the confidence, initiative, and applied judgement to deploy it in unfamiliar professional situations. This points to a dimension of the skills problem that no taxonomic or document-based analysis can capture: the quality of pedagogical processes through which competencies are developed, practised, and assessed. This resonates with work-integrated learning and competence-based education research emphasising that learning outcomes guarantee competence acquisition only when coupled with teaching and assessment practices that demand active application in authentic contexts (Urkia-Basterra et al., 2025; Yaya et al., 2025). The pedagogical transfer gap concept thus connects curriculum alignment to pedagogical effectiveness—a linkage largely absent from the existing skills mismatch literature.

On micro-credential design, the study derives three empirically grounded principles from the intersection of quantitative gap analysis and stakeholder validation. First, micro-credentials should target clearly defined, specific technical competencies rather than generic or soft-skill formulations—the latter are not valued by industry respondents as credentialling objects (IR4). Second, they should align with externally recognised quality benchmarks rather than relying on university-issued certificates alone, since employer trust is conditional on external validation (IR2). Third, they should function as continuing professional development instruments for existing specialists rather than as entry-level substitutes for degree-level education (IR3), positioning them as additions to rather than replacements of formal programmes—a positioning consistent with the Council Recommendation on micro-credentials (Council of the European Union, 2022) and with the broader literature cautioning against the fragmentation of curriculum coherence through unbundling (Wheelahan & Moodie, 2021; Varadarajan et al., 2023). Applying these principles to the substantive gaps identified in this study, the framework generates illustrative micro-credential priorities for each occupation: data analysis, business intelligence tools, and AI applications in recruitment for the Personnel Specialist; management accounting, cost structure analysis, and applied business intelligence for the Finance Manager; and strategic digital transformation and AI-supported decision-making for the Organisation Manager. These are presented as illustrative outputs of the framework rather than prescriptive recommendations, demonstrating the type of evidence-based, occupation-specific guidance the diagnostic process can produce for any institution that applies it.

The study also confirms the continued importance of expert judgement and stakeholder validation as complements to AI-assisted analysis. Job advertisement data, while scalable and real-time, systematically understate some requirements, overstate others by combining expectations from multiple roles, and over-represent white-collar professional occupations (Cammeraat & Squicciarini, 2021). The five-step institutional process through which the framework operationalises this complementarity is described in the Methodology and summarised in the Conclusions.

6. Conclusions

This study developed and applied a three-layer AI-assisted analytical framework linking empirical labour market demand, normative professional standards, and educational supply through a unified ESCO-based taxonomy, with the aim of identifying substantive curriculum gaps and informing the design of targeted micro-credentials. The proof-of-concept application to three RTU study programmes—Personnel Specialist (short-cycle), Finance Manager (first-cycle), and Organisation Manager (second-cycle)—confirms the operational feasibility and analytical value of the framework and yields four substantive conclusions.

First, a large share of apparent misalignment between labour market demand, professional standards, and programme learning outcomes is attributable to differences in how competencies are represented across institutional documents rather than to their genuine absence. Composite matching substantially increased observed alignment across all comparisons where it was applied, demonstrating that curricula are more responsive to labour market demand than strict one-to-one taxonomic matching suggests. This has direct implications for how policymakers and institutional leaders should interpret skills gap diagnostics: a single metric based on strict matching overestimates the gap and may generate unnecessary and costly curriculum revision.

Second, the one finding that persists as a substantive gap across all programmes and all comparisons—even after composite matching—is the consistent under-representation of digital competencies in programme learning outcomes (2.78–5.66%). This is independently confirmed by industry respondents across all three occupational profiles and constitutes the clearest priority for curriculum action. Addressing it requires both the introduction of new content and more explicit formulation of digital learning outcomes in programme documentation, since part of the gap reflects a visibility failure—digital skills taught implicitly across courses but not articulated as discrete outcomes—rather than entirely absent instruction.

Third, the pedagogical transfer gap—the discrepancy between competencies formally specified in curricula and those effectively demonstrated by graduates in professional contexts—represents a dimension of the skills problem that neither the quantitative framework nor any document-based analysis can capture. Closing this gap requires attention to pedagogical processes, teaching methods, and assessment practices that demand active competence application, not only to curriculum content. This finding points to a productive direction for future research connecting curriculum alignment with work-integrated and competence-based pedagogies.

Fourth, micro-credentials can add value in addressing the substantive gaps identified by the framework, but only under specific conditions: they must target well-defined technical competencies, meet externally benchmarked quality standards, and function as continuing professional development instruments rather than degree substitutes. Generic soft-skill micro-credentials are not valued by industry respondents and risk undermining the credibility of the broader micro-credential ecosystem.

Several limitations of the study should be acknowledged. The framework was applied to three occupations within a single institution and one national context, constituting a proof-of-concept rather than a generalisable validation. The exploratory nature of this validation means that the framework’s performance across different institutions, disciplines, and national qualification systems remains to be tested; cross-institutional and cross-disciplinary application is the most important direction for future work. Online job advertisements over-represent white-collar, high-skilled occupations and provide an approximation of employer demand rather than its full complexity. The qualitative sample of six participants—appropriate for contextualising and qualifying quantitative findings—is insufficient for drawing broader conclusions about the national ecosystem. Composite matching was applied to two of the three pairwise comparisons; its consistent application to the professional standards versus programme learning outcomes comparison is a methodological priority for future analysis. The ESCO-based framework captures alignment between intended learning outcomes and demand, not actual graduate competence acquisition, and the study does not include the perspectives of students and graduates—the ultimate beneficiaries of curriculum decisions—whose experience of the gap between formal learning outcomes and professional preparedness would provide an important complementary lens. Benchmarking composite matching against alternative semantic alignment techniques, such as AI-based similarity scoring, is a further direction for future work.

Future research should extend the framework to course-level learning outcomes to enable more granular diagnostic guidance, apply composite matching consistently across all three pairwise comparisons, and conduct cross-institutional and cross-disciplinary comparisons to establish the generalisability of the three-layer approach. Longitudinal studies examining the effectiveness of micro-credentials designed on the basis of such diagnostics, and research foregrounding the student and graduate perspective on the pedagogical transfer gap, would substantially strengthen the evidence base for AI-assisted curriculum monitoring.

Overall, this study demonstrates that AI-assisted, three-layer analysis of real-time labour market dynamics—combined with composite matching, ESCO standardisation, and stakeholder validation—can serve as a practical institutional tool for continuous curriculum monitoring and targeted micro-credential design. In the Latvian context, the framework can support coordinated action across higher education institutions, the Ministry of Education and Science, PINTSA, and employer representative bodies such as the Employers’ Confederation of Latvia (LDDK), each of which contributes to a different layer of the alignment ecosystem. More broadly, the study’s core argument—that skills mismatch is shaped not only by differences in supply and demand but by how competencies are measured, represented, and translated across institutional systems—offers a reframing with implications for curriculum policy, quality assurance, and labour market analytics well beyond the Latvian context.