Key Takeaways
It can look as if adopting an AI learning platform will automatically complete skills-based HRD. In reality, the opposite is true. A platform is a container for data. If the organization has not decided which skills connect to which jobs, work, and outcomes, learning recommendations, AI coaching, and capability diagnosis remain superficial features.
CompTIA’s 2026 Workforce and Learning Trends reports that 83% of organizations place a high priority on addressing skills concerns, and 62% of HR professionals and IT leaders expect AI training budgets to increase over the next year. Yet the same research states that only 34% of companies have formal, organization-wide upskilling or reskilling programs. Interest and budgets are growing, but operating structures for defining and accumulating skills are still insufficient.
Therefore, HRD’s 2026 question must be “What decisions will our organization use skills data for?” before “Which AI learning platform should we buy?” Skills data is not a record of training completion. It is HR operating data that connects job change, learning pathways, work application, internal mobility, and performance validation.
Skills-Based HRD Starts With a Data Model, Not a Platform
When organizations begin skills-based HRD, they often first review LMSs, LXPs, and AI learning recommendation tools. Tools are, of course, necessary. But when tools come first, an organization can build a system that recommends a lot of learning content while failing to build a system that explains which skills are needed for which work transitions.
Deloitte’s 2026 Global Human Capital Trends explains that competitive advantage is shifting from static workforce placement to orchestrating people, skills, data, and technology in real time. The key here is not technology itself, but the connection structure. People’s current roles, required skills, changes in work, data, and technology use must be connected in one flow.
CompTIA’s survey includes respondents related to workforce development, including HR professionals and IT leaders. In this sample, priority for addressing skills concerns is high, but only 34% have formal organization-wide upskilling or reskilling programs. This gap is difficult to explain as a platform shortage alone. The more fundamental problem is that organizations have not decided what unit to use for defining skills, what data to use for verification, and which HR decisions to connect them to.
A skills data model must answer at least four questions. First, which skills matter by job? Second, how will the current skill levels of employees be confirmed? Third, how will actual work application after learning be assessed? Fourth, how will the results be connected to placement, internal mobility, role transition, and performance management?
The First Data Is Not “Job Title,” but the Connection Between Work and Skills
The starting point for skills data design is not the job title. Even within the same job title, automated work, judgment work that humans must continue to perform, and AI-supported validation work can differ. As discussed in Article 5, reskilling candidate selection should also be based on changing work rather than the entire job. The same principle applies to data design in Article 6.
CompTIA identifies current training formats including job role-based training at 64%, foundational AI skills training at 64%, compliance and security at 62%, workflow-related training at 62%, and advanced AI training at 53%. These figures show that learning is already moving toward the units of job role and workflow. Therefore, skills data is insufficient if it merely attaches a list of competencies below a job title.
In practice, it is useful to break jobs into three levels. The first is the job title: for example, recruiter, learning professional, or sales manager—the role units the organization already uses. The second is key work. For a recruiter, this might include candidate sourcing, interview operations, recruiting data analysis, and onboarding linkage. The third is the skills required for each unit of work. These need to be broken into smaller units such as data interpretation, interview design, stakeholder communication, AI tool use, and privacy judgment.
Designed this way, the platform’s learning recommendations also change. Instead of “recommend AI training to recruiters,” the system can recommend strengthening data validation and candidate-experience design skills that remain after candidate sourcing is automated. Skills data should be a map of work change, not a classification table for a training catalog.
The Second Data Is Skill Proficiency Change, Not Learning History
Many organizations’ LMSs contain course history, completion rates, satisfaction scores, and test scores. But these alone cannot explain which skills employees have actually improved or which work they can now take on. In skills-based HRD, the needed data is not “the employee took a course,” but “a specific skill proficiency moved from one level to another.”
SHRM’s 2026 AI in HR report states that 56% of respondents do not formally measure the success of HR organizations’ AI investments. Survey populations and industry composition may differ from Korean companies, but this shows that measurement structures are not keeping pace as AI adoption and learning investments increase. The same applies to AI learning platforms. If only completion rates remain, it is hard to explain investment outcomes.
Skill proficiency data does not need to start with excessive complexity. A four-level model can be enough. Level 1 is conceptual understanding, Level 2 is performance with a guide, Level 3 is independent performance, and Level 4 is coaching others or proposing work improvements. The key is that these levels must connect to real work behaviors, not course names.
For example, if “generative AI use” is managed as a single skill, the data becomes vague. For HRD professionals, it should be divided into work-specific behaviors such as drafting learning needs analyses, structuring learning content, summarizing survey responses, and reviewing draft training-effectiveness reports. Only then can pre- and post-training diagnosis, manager feedback, and project application cases connect into one skills-change dataset.
The Third Data Is Work Application After Training
If skills data remains only inside HRD, it may support a report that “training was well run,” but it cannot explain that “the organization’s way of working changed.” The organization must also record which work the skill was applied to after training, which projects employees were assigned to, and which role transitions followed.
TalentLMS’s 2026 L&D Report focuses on an HR manager sample and covers learning design, budgets, priorities, and performance measurement. It presents supporting measures of L&D success: business impact at 37%, career growth outcomes at 31%, and training satisfaction at 28%. The fact that business impact and career growth outcomes appear alongside satisfaction is important. It signals that learning data must connect to actual work and internal mobility.
Work-application data does not need to begin with grand productivity metrics. It can start with small data points: work applied within 30 days after training, the output produced, manager confirmation, peer feedback, project assignment, and the unit of work automated or improved. The key is not to break the link between “training completion” and “work change.”
For example, after AI training, HR should check whether employees reduced report-drafting time, improved customer-inquiry classification standards, automated repetitive work, or created criteria for reviewing meeting-summary quality. Only when this data accumulates can the performance indicators discussed in the next article be designed around application, mobility, and role transition rather than completion rates.
HRD Practitioner Checklist: Six Data Questions Before Platform Adoption
Before reviewing an AI learning platform or LXP, HRD must first organize its data questions. CompTIA’s 2026 finding that 62% of HR professionals and IT leaders expect AI training budgets to increase shows growing investment pressure. But the fact that only 34% of companies have formal upskilling or reskilling programs is also a warning to check the data structure before investing.
First, what unit does our organization use to define skills? If job titles, job competencies, work, behavioral indicators, and tool-use capabilities are mixed together, search and recommendation quality will be low even after data is loaded into a platform.
Second, is there a connection table between skills and work? Broad expressions such as “AI capability” need to be replaced by work-behavior phrases such as “summarizes open-ended training satisfaction responses and categorizes improvement tasks.”
Third, how will current levels be diagnosed? The organization must decide whether to use only self-diagnosis, add manager confirmation, review task performance, or look at project outputs. Without a diagnostic method, learning recommendations remain recommendations based on individual interests.
Fourth, how will change after learning be recorded? Completion status, diagnosis-score change, work-application cases, manager confirmation, and project assignment can serve as minimum candidate data.
Fifth, which HR decisions will use this data? The required level of data differs depending on whether it will be used only for learning recommendations, or also for internal mobility, role transitions, succession candidate pools, project assignment, and workforce planning.
Sixth, how will privacy and evaluation risks be managed? Skills data is useful for employee growth support and placement decisions, but it loses trust if used for opaque evaluation or labeling. The purpose of data collection, access rights, retention period, and personal feedback approach must be set before the platform contract.
What HR Should Watch Next
The core of upskilling and reskilling in 2026 is not deploying more learning content. HR needs to be able to explain which work is changing, which skills are needed, who is at which level, and what work employees applied their learning to afterward. That explainability is the role of skills data.
A platform can collect and display this data well. But without a data model, the platform only repeats completion rates and content recommendations. Conversely, if the flow of work-skill-diagnosis-learning-application-validation is organized, a pilot can begin even with a small LMS or spreadsheet.
The next article will examine which performance indicators this data should lead to. Reskilling outcomes are hard to explain with completion rates alone. Metrics such as internal mobility, role transition, work application, productivity improvement, manager evaluation, and project assignment must be included. Skills data is the foundation for building those performance indicators.
