Industry · HR & skills matching

A skills matching platform that helps supply and demand truly understand what skills mean.

We build custom skills-matching platforms for organisations that want to take the match between people, roles, projects and learning paths seriously: large employers with an internal talent marketplace, staffing agencies and recruiters who want to go beyond keyword matching, training institutes that want to link learning pathways to the labour market, and sector-specific labour market platforms for healthcare, engineering and construction. Built around ESCO and O*NET taxonomies, AI-driven, AI Act-compliant, and with code ownership that stays with your organisation.

Use caseInternal talent marketplace
Use caseAI-driven recruitment
Use caseSkill gaps & learning paths
Use caseSector-specific labour market platforms

The skills economy in numbers.

~14.000
Skills in the ESCO taxonomy (the EU standard for skills)
~17.000
Occupations in O*NET with underlying skill profiles
~70%
Of employers in EU studies report persistent skill gaps in core technology
HIGH-RISK
AI Act classification for AI in recruitment and selection (Annex III, Art. 6)

Source: European Commission ESCO classification, U.S. Department of Labor O*NET, World Economic Forum Future of Jobs, Regulation (EU) 2024/1689 (AI Act).

Keyword matching was designed for yesterday's labour market.

Most matching systems on the Dutch market, whether the large ATS packages or the in-house systems HR has built up over the years, still rely on some form of keyword matching: the CV contains the word "Python", the vacancy asks for "Python", so it's a hit. Anyone who has ever heard a hiring manager complain that the system produces too much noise knows the problem. The software doesn't understand that Pandas is part of Python, that data analysis in R is a closely related competency, or that five years as a senior data engineer at a SaaS scale-up implies a deeper set of skills than a CV on which "Python" appears ten times.

On the other side of the market, the same problem plays out in mirror image. An internal employee in a talent marketplace gets no suggestions that fit their actual skill profile, only roles in which the exact words from their CV appear. A student at a training institute sees no learning paths that fit their target role. A healthcare worker on a sector-specific labour market platform sees no alternative roles at another care provider, even though their skill set overlaps seventy per cent with that role.

The other extreme, a large AI vendor such as Eightfold, Gloat or a comparable international player, asks you to bend your organisation to fit their template. Their taxonomy, their matching algorithm, their audit trail. Code ownership sits with them, your data too, and if the contract ends your operation grinds to a halt. Since 2026 there is a further complication: AI in recruitment and selection is classified as HIGH-RISK under the AI Act, with documentation, bias-testing and audit obligations that you, as the deployer, must be able to demonstrate, not the vendor.

A custom skills-matching platform takes a different approach: the system follows your skills taxonomy, your ontology, your matching criteria and your audit requirements. We build that layer on top of proven taxonomies such as ESCO and O*NET, with a custom ontology layer where your organisation-specific roles, positions and projects have a place. For the AI layer, we draw on custom AI development and the custom LLM integrations we have delivered for other clients: vector embeddings for skill similarity, RAG for context-rich matches and a classic text-matching layer as a fallback.

Skills-matching software that fits any context.

Each context has a different audience, different taxonomy choices and different audit requirements. Below, for each use case, is what we typically build — from internal talent marketplaces to sector-wide labour market platforms.

Internal talent marketplaces for large employers

At a large employer with multiple business units, sites or international departments, internal mobility is an underused channel. The vacancy is advertised externally while someone in another department already has exactly the right skills profile — but nobody knows, because the talent data is scattered across an HRIS, a learning platform, performance reviews and a few stray Excel files. An internal talent marketplace brings that data together around a skills layer: an enriched profile for each employee, a skills requirement for each open role or project, and between the two an AI-driven match score with an explainable rationale.

Our module set for internal marketplaces builds on an employee skills profile that is enriched from the HRIS, learning platform, certifications and the employee's own self-reflection, an open roles feed where department managers publish vacancies and projects, and a match engine that gives each employee a prioritised list of suggestions — with the option for employees to explore anonymously or on an initial, exploratory basis. For the marketplace layer itself, we draw on our experience in building a marketplace platform — two-sided flows, moderation flows, transparent ranking.

One difference from the Eightfolds and Gloats of this world: the system does not make the choice for the employee. It makes the choice visible, supports the manager with a well-founded shortlist, and the employee stays in control. We build an audit trail of match decisions that records which skills were weighed, what weight the algorithm gave them and which decision a human made. That makes the organisation AI Act-compliant and allows a rejected candidate to understand later on why they did not progress.

  • Skills profile from multiple sourcesHRIS, learning platform, certifications, self-reflection — no more standalone forms.
  • Open roles and projects feedDepartment managers publish directly, with no detour via central HR.
  • Anonymous exploration modeEmployees can look around first without their manager knowing who is looking.
  • Explainable match scoreFor each match you can see which skills count and how heavily they are weighted.
  • Audit trail per decisionAI Act-compliant documentation of match and rejection decisions.

Which skills-matching platforms we typically build.

Skills matching is broader than recruitment alone, and that shows in the kinds of platforms we build for clients. For one large employer, an internal talent marketplace is the focus: employees explore roles and projects within the group, department managers get a well-founded shortlist from within the organisation before a vacancy goes out externally, and HR gets an overview for the first time of where the skills sit in the organisation and where the shortages are. This often comes with secondment and project hiring flows — an employee works temporarily in another business unit without changing employer — where the system tracks cost allocation, the return route and skills development.

For staffing agencies, secondment firms and recruitment agencies, we build an AI-driven matching platform between candidates and clients. This is where the skills-matching layer meets the broader recruitment context, and where we build on our experience of developing recruitment software. The candidate pool is larger, the match frequency higher and the speed requirement tougher: a recruiter wants to see a prioritised shortlist within seconds of a vacancy coming in. This is where a vector-embeddings architecture comes into its own: a CV and a job description are translated into vectors in a semantic space, and their cosine similarity gives a richer picture of match relevance than any keyword search.

A skills-matching platform often needs to work alongside an existing ATS: the recruiter keeps working in their ATS, but gets a match score per candidate directly, an explainable rationale and an audit trail for the AI Act. For clients with an established ATS stack, such as Bullhorn, Connexys, Recruitee or Homerun, we connect the matching platform through standard interfaces and keep the workflows in the ATS leading. The skills layer enriches; it does not replace.

A third group is training institutes and learning providers. Here the match is not about vacancies but about learning paths: given a target role or target skills profile, which modules get the learner there fastest? Which certifications demonstrably give an edge in the labour market, and which modules are already mandatory for a profession under the collective labour agreement or statutory qualification requirements? We build these platforms so that each learner gets a skills gap analysis per learning track, a recommended path and ongoing feedback. For institutes with a re-integration or retraining role, we link the gap analysis to a labour-market feed: which jobs are opening up in a region that fit this learning path?

A fourth group is sector labour-market platforms and guilds: in healthcare, engineering and construction, public-private platforms are emerging to tackle staff shortages at sector level. Here we deal with multiple employers at once, a labour-market fund that co-finances, a sector collective labour agreement that sets standards for roles and certifications, and an ambition to combine career changers, returners and internal movers. For healthcare, there is an NEN 7510 layer on top, and in many sectors an eIDAS signature flow for training certificates.

A final group is employer-branding pages and talent communities: an employer builds an open community where professionals register with a skills profile, receive content that fits their career direction, and are matched to vacancies at a later stage. The common thread: the software follows your taxonomy and your decision rules. Vendor-independent, AI Act-compliant, with code ownership resting with your organisation.

Built within the laws that apply to matching, AI and HR data.

Skills matching touches several frameworks at once: the AI Act, GDPR, sector compliance and information security. From the first sprint we work within those frameworks, not as a final audit round before go-live.

AI Act: HIGH-RISK

Recruitment and talent matching under Annex III

The EU AI Act classifies AI used in recruitment, selection, task allocation and employee monitoring as HIGH-RISK (Annex III, Art. 6 of Regulation (EU) 2024/1689). For you as a deployer, that means a risk-management system, a data-governance policy, technical documentation, logging, human oversight, transparency towards the candidate and a conformity assessment before production. We build that documentation layer as an integral part of the system. Also read about our approach to custom AI development.

GDPR

Privacy by design for CV and skills data

CVs, skills, performance data and learning history are personal data, and combining them quickly amounts to profiling under Article 22 GDPR. We deliver a DPIA for every project, separate training and production data, minimise to the necessary fields and build data subject rights (access, rectification, restriction, erasure) as first-class flows. Automated decisions without human intervention are prohibited under the GDPR; our platforms are built by default so that a person makes the final decision.

Bias evaluation

A structural check for fairness and disparate impact

A matching algorithm trained on historical data risks reproducing historical inequality. We build a bias evaluation pipeline into the platform that measures, for every release, whether the algorithm produces substantially different outcomes across groups: gender, age, surname origin and educational background. Where it does, we reweight features, revise the training data or escalate to a manual decision. It is a build step in the release pipeline, not a one-off audit after the fact.

ESCO & O*NET

International skills taxonomies as a foundation

We use the European ESCO classification as the source taxonomy for the European labour market and the American O*NET database as a rich supplement for competencies, tasks and related occupations. On top of that we add our own ontology layer, in which your organisation-specific roles, positions, certifications and project types find their place, linked to ESCO/O*NET wherever possible. This gives you multi-language support and a connection to public labour market platforms.

eIDAS

Qualified electronic signatures on certificates

For training institutes and sector platforms that issue certificates and declarations of competence, we integrate eIDAS-compliant electronic signatures: qualified where the law requires it, advanced where that is sufficient. Certificates are stored in a machine-readable format, so that in a later matching process they can count directly as proof of a skill.

NEN 7510

Information security for healthcare contexts

For sector labour market platforms in healthcare, NEN 7510 is the information security framework. We design within that framework: role separation, authorised access, logging where applicable and certification-ready on delivery. For financial institutions and government, we align with ISO 27001.

Employee monitoring

No continuous assessment under the radar

The GDPR and the Dutch Works Councils Act are clear: continuous monitoring of employee performance is only permissible under strict conditions, with the consent of the works council and a substantiated legal basis. Our internal marketplaces only measure skills that employees have actively provided and publicly visible performance data, and they show employees when their profile is being enriched.

Ownership & source code handover

The taxonomy, the models and the data stay with your organisation

We deliver code, skills ontology and matching models in a repository managed by your IT organisation, which continues to exist without Appfront. No mandatory licences, no hidden hosting lock-in on the embeddings store, no compulsory maintenance contract. We can provide hosting and fine-tuning if you want it, but that is your choice.

Seamlessly connected to your HR, ATS and learning stack.

We integrate with the systems your organisation already uses. The skills matching layer complements them and rarely replaces them. Your HRIS, ATS and learning platform remain the sources of truth.

AFAS Profit
HRIS & payroll
Workday
Talent & HR
SAP SuccessFactors
HXM platform
Bullhorn
Recruitment ATS
Connexys
Recruitment ATS
Recruitee
Hiring platform
Homerun
Hiring & ATS
ESCO API
EU skills taxonomy
O*NET Database
Occupations & competencies
OpenAI / Mistral
LLM & embeddings
LinkedIn Talent API
External talent pool
Cornerstone / Moodle
Learning & LMS

Standard integrations, no project-specific connections.

We've standardised the integrations mentioned above across multiple skills-matching projects. On a new engagement, we set up the integration using the same abstraction layer: fewer bugs at the interfaces, faster turnaround in the build phase and a data architecture your own IT team can maintain. The integration layer is a mappable adapter, so a migration from Workday to SAP or from Bullhorn to Recruitee is a configuration change at the interface, not a rebuild of the matching platform. For the details, our page on smart API integrations is the right place.

For the AI and embeddings layer, we work model-agnostically: OpenAI, Mistral, Anthropic or a locally hosted open-source LLM. The choice depends on data residency requirements, cost profile and latency. For healthcare and government, a locally hosted model is often the right choice; for commercial recruiters, a hosted API often offers better price-performance. For the embeddings store, we work with pgvector, Pinecone or Weaviate, depending on data residency and existing infrastructure. For the broader context, see our approach on the custom LLM integrations page.

One difference from the big AI vendors: with us, the model weights (where possible), the API credentials, the embeddings store and the taxonomy belong to your organisation. We configure and build, but we don't make Appfront indispensable in the chain. If your IT partner or your LLM vendor changes, there's no hidden cost quietly running in the background, and the fine-tuning of the matching model moves with you to the new environment.

From taxonomy workshop to go-live in clear steps.

A skills-matching project has its own rhythm. Five phases that are recognisable for every client, whether it's an internal talent marketplace, a staffing agency or a sector-specific platform.

01 · Audit

Taxonomy and data mapping

Workshops with HR, recruitment and the business: which skills count, which roles exist, which data sources are available, which ESCO/O*NET integration fits, and which organisation-specific ontology extension is needed.

02 · Design

Model architecture and bias baseline

Which LLM, which embeddings store, which fallback path if the AI fails, and which bias tests are already part of the baseline. Plus: AI Act documentation structure, DPIA template and audit trail schema.

03 · Build

Short sprints, a demo every sprint

Every sprint we deliver a working flow that recruiters, talent managers and compliance officers can test. The first working match demo is up and running within just a few sprints.

04 · Cutover

Pilot in one business unit or vacancy type

First one department or one role type, then we broaden the rollout. With every rollout: training for recruiters and hiring managers, communication to employees and candidates, and handover to your functional management team.

05 · Maintenance

Continuously evolving with you

With every ESCO update, AI Act guideline or new sector collective agreement, we adapt the system. Ongoing bias monitoring, retraining on new data, and first-line support within the same working day.

What the industry says about AI-driven skills matching.

European Commission — ESCO

"The European classification of skills, competences and qualifications, ESCO, provides a common language that strengthens mobility between sectors, countries and providers of education and employment, a foundation for digital matching applications in the European labour market."

World Economic Forum — Future of Jobs

"Half of all workers worldwide will need to be reskilled or upskilled in the coming years; organisations that set up their internal talent marketplaces seriously and link them to a skills graph retain more control over their labour market position."

AI Act — Regulation (EU) 2024/1689

"AI systems used for recruitment and selection, task allocation based on performance, and monitoring of employees are categorised as high-risk, with corresponding requirements for risk management, data governance, technical documentation and human oversight for the deployer."

Answers for the HR or talent director considering a custom build.

The questions we hear most from HR directors, talent managers and sector labour-market platforms, answered from the practice of our own projects.

Why not an off-the-shelf package such as Eightfold, Gloat or a comparable international player?
For an organisation with a simple use case and few sector-specific rules, a SaaS vendor is often the right answer. We only come into the picture when complexity increases: multiple business units with their own job frameworks, a sector collective labour agreement (CAO) with specific certification requirements, an internal ontology that doesn't fit into ESCO, or an AI Act audit requirement that the vendor doesn't adequately cover. A second reason is ownership: with a custom build, the code, the taxonomy, the embeddings and the hosting stay with your organisation, whereas with a SaaS vendor they sit on their side, including everything you build up in model fine-tuning on your historical data.
Can you integrate with our existing HRIS, ATS or learning platform?
Yes. We have standardised adapters for AFAS Profit, Workday, SAP SuccessFactors, Bullhorn, Connexys, Recruitee, Homerun and most common learning platforms. For less common systems, we build an adapter case by case. A later migration between vendors is a configuration change at the integration layer, not a rebuild. Also read about smart API integrations and specifically ATS integrations.
How do you make sure the system complies with the AI Act?
Under the AI Act, AI used in recruitment and selection is classified as HIGH-RISK (Annex III). For you as the deployer, that means a risk management system, a data governance policy, technical documentation, logging of decisions, human oversight and transparency towards the candidate. We build that documentation layer as an integral part of the system: for every match decision, an audit trail showing which skills were weighed, which features the algorithm used and which decision a human ultimately made. A rejected candidate has the right to a meaningful explanation, and that explanation must come from the system itself, not from a reconstruction after the fact.
How do you deal with bias and discrimination?
We build a fixed bias evaluation pipeline into the release flow. For every new model version, we measure whether the algorithm produces substantially different outcomes across groups: gender, age, surname origin, educational background. Where it does, we reweight features, revise the training data, or introduce a human review as a mandatory step. We document this (what the test measures, what the thresholds are, what the results are) as part of the AI Act documentation. This isn't meant to be a separate audit after the fact, but a build step in the matching model's pipeline.
Which skills taxonomy do you use as a basis?
By default, we work with ESCO as the source taxonomy for the European context and with O*NET as a rich supplement at competency and task level. On top of that, we add our own ontology layer in which your organisation-specific roles, positions, certifications and project types find a place, linked to ESCO/O*NET where possible. For sector platforms, we link to qualification frameworks (CREBO, NLQF, professional association registers). For multinationals, this gives direct multilingual support and international interoperability.
Do you work with LLMs and embeddings, or with classic matching?
Both, and it's precisely that combination that makes the difference. We build a vector embeddings layer for semantic skill similarity and a classic text-matching layer as a fallback and second opinion. For the LLM layer, we work in a model-agnostic way: OpenAI, Mistral, Anthropic or a locally hosted open-source LLM. In data-residency-sensitive contexts, a local model is often the right choice. Our approach is described on the custom LLM integrations page.
How do we handle information security and compliance in a healthcare context?
For sector-wide labour-market platforms in healthcare, NEN 7510 is the governing framework. We design and build the platform within that framework: separation of roles, authorised access per role, logging where applicable, and certification-ready at delivery. For the privacy layer, we combine this with GDPR-compliant flows for access, rectification, restriction and erasure. For certificates and declarations of competence, we integrate eIDAS-compliant electronic signatures — qualified where the law requires it, advanced where that suffices — linked to a trust service provider of your choice.
Do you also build a talent marketplace for internal employees?
Yes — it's one of the most common use cases. We build an internal talent marketplace with an enriched employee skills profile, an open roles and projects feed where department managers publish directly, and an AI-driven matching engine with explainable reasoning. Often with an anonymous exploration mode: employees can explore options without their manager knowing who is looking. For the marketplace layer, we draw on our experience with building marketplace platforms.
What does a custom skills-matching platform cost?
A tightly scoped first implementation (one role type, one HRIS integration, one business unit) is a different project from an organisation-wide rollout with multiple collective labour agreements, multiple ATS integrations and a sector-specific audit requirement. We work with a fixed planning phase in which we get the scope clear — only then do we give a concrete price for the complete build, including integrations, AI Act documentation and a bias-evaluation pipeline.
How do you get started in an organisation without its own AI team?
With one use case and existing data. We run the taxonomy workshop, deliver the first pilot and set up a dashboard that lets your HR or talent team see the match results themselves. Only once the pilot has proven itself do we broaden the scope. You don't need your own AI team to get started. We deliver the software, monitoring and documentation so that a mid-market IT department can take it over without building a complete AI stack. For the broader context, we draw on our approach to custom AI development.
Share LinkedIn Email

Talk to us about your skills-matching platform.

A half-hour introductory call with your HR director, talent manager or sector board member. We listen, ask questions about your skills taxonomy, AI ambitions and compliance context, and give you an initial direction. No obligation.

Edit content