한국어 
日本語 
English 
LMS vs LXP — what is the difference, and which one should you choose?
Last updated: 2026-07-14
Definitions and boundaries, the four-layer architecture of next-generation learning operations, interoperability standards, and a decision table by organizational condition. Criteria, not vendor rankings.
learning systems (Josh Bersin)
operations architecture
(xAPI, LTI, Open Badges, etc.)
An LMS manages completion. An LXP designs discovery.
An LMS (Learning Management System) distributes courses built by the L&D team and aggregates completion and audit evidence. An LXP (Learning Experience Platform) is designed around learners discovering, creating and being recommended content. As of 2026, however, this distinction is less a product category than a difference in how learning is operated. In the global market, LMS, LXP and microlearning are converging into a single learning operating system. The real question is not "LMS or LXP" but how your organization supplies content, and what it measures.
This page does not rank vendors. It presents definitions and boundaries, the four-layer architecture, interoperability standards, the five-stage evolution, and a decision table by organizational condition — each with a verification method and a cited source.
The four-layer architecture, interoperability standards and global vendor landscape on this page describe capabilities that the global AI HRD/L&D market demands. They are not a list of TouchClass features. What TouchClass can actually evidence today is stated, with ratings, in the table below: Where TouchClass sits in this frame. Skill graph support is partial (△).
Definitions and boundaries
-
It distributes courses designed by the L&D team, evaluates completion conditions, and aggregates evidence. The unit of work is the course; the headline metrics are completion rate and time spent. For work such as statutory compliance training — where you must show an auditor who completed what and when — an LMS remains the precise tool. In TouchClass operating data, compliance-led operation was the single most common usage pattern across more than 100 companies. See Completion management.
Unit: the course · Metric: completion rate -
Learners discover, consume and produce content themselves, and the system recommends what to learn next based on skill gaps and job context. The unit is not the course but the learner and the skill, and content comes from multiple sources — the L&D team, frontline employees and external libraries. This means an LXP is not something you switch on by buying a feature: it only works if your content supply structure changes. See Social learning.
Unit: learner & skill · Multi-source content -
Josh Bersin describes the separate LMS, LXP and microlearning categories converging into one (Dynamic Enablement). External research frames the learning portal as being redefined into an AI-native operating system connected to HCM, performance, knowledge management and collaboration tools. The practical implication is simple: classify candidates not by product label but by whether they are course-catalog-centric or skill-and-data-centric.
LMS + LXP + microlearning → one category -
Research converges on a shift in metrics — from time spent and completion rate to proficiency gain, speed of role transition, internal mobility, productivity, revenue contribution and risk mitigation. The problem is that very few organizations actually measure these. Only 8% of companies reported measuring Kirkpatrick Level 4 outcomes (ATD, 2024). In TouchClass operating data across 107 companies, roughly 9% had quantitatively evidenced a learning-to-outcome link, and fewer than 5 companies had evidenced revenue or productivity effects. See the operating data report.
Level 4 evidence: fewer than 5 of 107 companies -
The shared conclusion across the research is that the bottleneck is not a lack of technology but data quality, governance, trust, change management, and legal and ethical constraints. Recommendation quality comes from clean skill-role-content mapping plus exposure and outcome logs — not from better prompts. Meeting transcripts, collaboration messages, free-text performance reviews and compensation-adjacent data are classified as high-risk inputs: it is safer not to feed them into recommendations without explicit notice, consent, access control and audit. Keep notification channels separate from analytical inputs.
Separate high-risk data from recommendation inputs
Level 4 outcomes (ATD 2024)
link (N=107)
The four-layer architecture of next-generation learning operations
Global research describes the next-generation LMS/LXP not as a single application but as four stacked layers. This is a frame of capabilities the market demands, not a feature list for any one product. In an RFP, state explicitly who owns what at each layer.
-
Skill taxonomies and ontologies, content metadata, the LRS (Learning Record Store), the data warehouse and the vector store live here. A skill graph — skills, roles, content and learners connected as nodes and relationships — is the core of this layer. With the half-life of technical skills at roughly 2.5 years (Deloitte, 2024), how skill data is normalized and refreshed becomes the real issue. The verification question is a single one: does the contract state who owns the skill data and how it is normalized?
TouchClass: skill graph is partial (△) -
RAG retrieval, recommendation engines, content generation, assessment and quiz generation, coaching and role-play, and agent orchestration. AI-driven learning recommendations have been reported to lift completion rates by 35% (McKinsey, 2024), and 83% of L&D professionals named AI personalization their top priority (LinkedIn, 2025). Recommendation quality, however, is decided by the layer beneath, not by the model. TouchClass AI capabilities are documented on AI features.
Recommendation quality comes from data, not the model -
The LMS/LXP screen, work messengers, browser extensions, mobile and in-app embeds are where learning is actually consumed. "Learning in the flow of work" means learning is delivered inside the workflow rather than requiring a portal visit. In Korea, roughly 60% of the workforce are deskless workers (Ministry of Employment and Labor, 2024), which makes mobile the primary channel in practice. Channel choice is also segmented: external research indicates Teams is dominant in large enterprises, education and regulated industries, while Slack is strong in IT and digital organizations.
~60% of the Korean workforce is deskless (MOEL 2024) -
RBAC, consent management, policy engines, audit logs, explainability, privacy and security. This layer already appears as a purchasing condition in Korean RFPs — source attribution, hallucination control, opt-out from LLM API training, prevention of sensitive-data egress, permission/log/download controls, and a human-in-the-loop review step are written into requirements. TouchClass holds ISMS-P and ISO/IEC 27001:2022 certification, and does not use knowledge assets created or provided by customers during AI service usage as AI model training data. See Security.
Customer knowledge assets are not used for model training
recommendations (McKinsey 2024)
personalization top priority (LinkedIn 2025)
that is deskless
of expectations
Seven interoperability standards — what to write into the RFP
Standards are the most frequently omitted item in an LXP procurement. Without them, learning data disappears the moment you leave the vendor.
The seven below recur across the global market, each with its official source.
| Standard | What it standardizes | What to verify in the RFP | Official source |
|---|---|---|---|
| xAPI (Experience API) | A record format for learning experiences, including activity outside the portal | Can data be exported to an LRS? Is the statement schema published? | xapi.com |
| LRS (Learning Record Store) | The store that persists and queries xAPI statements | Is the LRS built in or integrated? Who owns the data? | adlnet.gov |
| LTI | The interface for connecting external learning tools to a platform | LTI version, tool registration flow, how permissions are passed | 1edtech.org |
| Open Badges 3.0 | A verifiable format for digital badges and credentials | Can a badge be verified outside your organization? | 1edtech.org |
| CLR 2.0 | Bundles learning and competency history into one verifiable record | Can learning history move with the person when they leave? | 1edtech.org |
| HR Open Standards | Data exchange between HR systems — includes a skills proficiency API | The contract for exchanging skill data with HRIS/HCM | hropenstandards.org |
| O*NET · Lightcast Open Skills | Open job and skill taxonomies | Are skill names defined in-house, or mapped to an open taxonomy? | onetonline.org · lightcast.io |
Standard list source: global HRD/L&D market research frame (xAPI/LRS, LTI, Open Badges 3.0, CLR 2.0, HR Open Standards skills proficiency API, O*NET, Lightcast Open Skills). Check the official sources above for current specifications.
Five stages in the evolution of corporate learning systems (Josh Bersin)
LMS and LXP are not different products so much as different moments on the same axis.
The five stages below trace how the role of the system has shifted.
| Stage | Period | Role of the system | The question of that era |
|---|---|---|---|
| E-Learning & Blended | 1998–2002 | LMS as E-Learning Platform | Can classroom training move online? |
| Talent Management | 2005 | LMS as Talent Platform | Can learning connect to talent management? |
| Continuous Learning | 2010 | LMS as Experience Platform (70-20-10) | Can we handle learning outside the course? — the origin of the LXP |
| Digital Learning | 2018 | LMS invisible, data driven | Can people learn without noticing the system? |
| Learning in the Flow of Work ● We are here |
2020– | Learning inside the workflow — the AI-embedded stage | Is learning delivered into the work context? |
Source: Josh Bersin, five stages in the evolution of corporate learning systems · Dynamic Enablement (convergence of the LMS, LXP and microlearning categories). The "question of that era" column restates each stage's system role as a practical verification question.
Decision table — when an LMS is enough, and when you need an LXP
An LXP is not a feature. It is a content supply structure. Switch one on in an organization where the frontline does not create content, and you get an empty feed.
This table is not a vendor recommendation. It is a way to locate the stage you are actually in.
| Your current condition | What you need now | Evidence · baseline | What to verify first |
|---|---|---|---|
| Compliance evidence is priority one Almost no always-on content |
An LMS is enough Completion & audit automation |
In our operating data, compliance-only operation was the most common pattern (40 companies), and its off-season MAU sits at 5–10%. | Run completion rules, non-completer segmentation and audit report export live in the demo |
| The L&D team supplies all content No frontline authoring (Level 1) |
LMS + AI authoring An LXP is premature |
Level 1 (fully supplied) is the largest group at ~40% of content self-sufficiency. Switch on an LXP here and there is nothing to fill it with. | Measure how long one administrator actually takes to build one piece of content |
| The frontline has started authoring (Level 2 collaborative · Level 3 distributed) |
Time to introduce LXP elements Social learning & curation |
Level 2 is ~35% and Level 3 ~20%. This is where learner-generated content actually appears. | Authoring permissions, review workflow, learning curve of the authoring tool |
| Learner-generated content is self-sustaining (Level 4) |
A full LXP + skill data | Fewer than 5% of companies reach Level 4. Social learning was the hardest of the five usage patterns to sustain (20 companies). | Controls for content quality, popularity bias and misinformation spread |
| You manage role transitions & internal mobility as metrics | Skill taxonomy + skill graph | Research converges on metrics shifting from completion to proficiency, role transition and internal mobility. Technical skill half-life is ~2.5 years (Deloitte 2024). | Ownership of skill data; whether it maps to an open taxonomy (O*NET, Lightcast) |
| Most of your workforce is deskless | Experience layer first Mobile & in-workflow delivery |
About 60% of the Korean workforce is deskless (MOEL 2024). A design that assumes a portal visit never reaches them. | The actual mobile app, on-site access paths, offline handling |
| You want AI recommendations & automated measurement | Logs, skill standards and experiment infrastructure come first | Recommendation quality comes from clean skill-role-content mapping plus exposure and outcome logs — not from prompts. Skip that order and the recommendations cannot be validated. | Are exposure and outcome logs captured? Can precision/NDCG be computed? |
Content self-sufficiency levels, the five usage patterns and off-season MAU are drawn from https://www.touchclass.com/en/data-report (operating data from 100+ companies · 35 months · 8 industries). External figures: Deloitte 2024, Ministry of Employment and Labor (Korea) 2024.
Global market landscape — which capabilities are being demanded
The following summarizes directions observable in the global market. It is not a recommendation, and it is not a ranking.
Public materials do not reveal the model architecture or performance metrics of any of these platforms.
| Global platform | Direction observable in public materials |
|---|---|
| Cornerstone | Integrates LMS, talent and skills on a skills knowledge graph, paired with labor-market data, workforce planning and a responsible-AI frame. |
| Workday | Handles skills cloud and talent mobility in an HCM context, placing learning on the same axis as HCM data. |
| Docebo | Pursues an AI-first learning ecosystem with AI content creation, AI coaching and a multi-LLM structure, connecting search, recommendation and analytics Q&A on one platform. |
| Degreed | A skills-first LXP: skill normalization, AI-driven skill reviews, and manager, peer, self and project signals used together. |
| Coursera for Business | Combines a career graph, skills tracks, coaching and role-play, and verified assessments. |
| LinkedIn Learning | AI recommendations and AI skill pathways, practice-first AI courses, manager and team insights, linked to internal mobility metrics. |
| Sana Learn | An AI-native suite bundling LMS, LXP, authoring and virtual classroom. |
| Arist | Flow-of-work enablement delivered through work messengers and SMS — no portal visit assumed. |
| TechWolf | Skills intelligence infrastructure that infers skills from work signals — productizing the skill data layer itself. |
Source: global AI HRD/L&D market research frame. The descriptions above reflect directions observable in each company's public materials and are not an assessment of performance or quality.
Where TouchClass sits in this frame
The four layers above describe what the global market demands — not the TouchClass feature list.
The table below states only what we can currently evidence, and marks partial support as partial.
| Layer | What the global market demands | What TouchClass can currently evidence | Rating |
|---|---|---|---|
| Data & semantics | Skill taxonomy & ontology, skill graph, LRS, vector store | We operate learning history, completion data and a knowledge base for AI retrieval. However, the skill graph — skills, roles, content and learners connected as relationships — is partially supported, and we do not provide full mapping to an open skill taxonomy. | △ Partial |
| AI services | RAG retrieval, recommendation, content & assessment generation, coaching, agents | Released in sequence: AI chatbot (2025.02), Quick Maker (2025.04), Short Class (2025.06), AI authoring tool (2025.07). Generates curricula and learning pages from a URL or a file and translates into 14 languages. See AI features. | ○ Evidenced |
| Experience | Mobile, in-workflow embedding, multi-channel delivery | Designed mobile-first, with learning delivered via app push, messaging, notices and pop-ups. Supports a 14-language interface. | ○ Evidenced |
| Governance | RBAC, audit logs, explainability, exclusion from model training | Holds ISMS-P and ISO/IEC 27001:2022 certification (recertified together in January 2026). Does not use knowledge assets created or provided by customers during AI service usage as AI model training data. See Security and Enterprise security. | ○ Evidenced |
Rating key: ○ verifiable from published evidence · △ partial support. The same rating scheme is applied in the LMS comparison checklist. Certification source: https://www.touchclass.com/en/security
Frequently asked questions
The eight questions we are asked most often when organizations evaluate LMS and LXP.
What is the difference between an LMS and an LXP?
An LMS distributes courses built by the L&D team and aggregates completion and audit evidence; its unit is the course and its headline metric is the completion rate. An LXP is built around learners discovering and creating content and receiving recommendations based on skill gaps; its unit is the learner and the skill. The boundary is collapsing, however. Josh Bersin describes LMS, LXP and microlearning converging into a single category. In practice it is more accurate to classify candidates by whether they are course-catalog-centric or skill-and-data-centric, rather than by product label.
Will adopting an LXP raise learner engagement?
Not by itself. An LXP is a content supply structure, not a feature. Switch one on in an organization where the frontline does not create content and you are left with an empty feed. In 35 months of operating data across more than 100 companies, Level 1 (content fully supplied by the L&D team) was the largest group at about 40% of content self-sufficiency, while Level 4 — where learner-generated content is self-sustaining — accounted for 5% or fewer. Engagement is determined by content supply frequency and the combination of usage patterns, not by platform category.
What is a skill graph?
A data structure that connects skills, roles, content and learners as nodes and relationships. Its purpose is to answer, from one graph, questions such as: which skills does this role require, which skills is this learner missing, and which content closes that gap. It belongs to the data and semantics layer of the next-generation learning architecture and is a precondition for personalized recommendation and internal mobility management. The real design question is whether skill names are defined in-house or mapped to an open taxonomy such as O*NET or Lightcast Open Skills.
Does TouchClass provide a skill graph?
Partially (△). TouchClass operates learning history, completion data and a knowledge base for AI retrieval, but a full skill graph — adopting a skill taxonomy and connecting roles, competencies and content as relationships — is not currently a complete capability. The same rating is published in our LMS comparison checklist. If skill-based internal mobility management is your first-priority requirement, verify this item before shortlisting any candidate, including us.
Should we replace our current LMS with an LXP?
The decision follows your organizational stage, not the product. If statutory compliance evidence is the first priority and there is almost no always-on content, an LMS with completion and audit automation is sufficient. If the frontline has begun creating content and you now need social learning and curation, it is time to introduce LXP elements. Before replacing anything, separate whether low usage is caused by the product or by the content supply design. You can self-diagnose on the LMS/LXP health check.
What should be in place before adopting an LXP?
Data comes first. The execution principle from the research is to build skill-role-content mapping, exposure and outcome logs, and experiment infrastructure first, and only then layer RAG and conversational analytics on top. Recommendation quality comes from cleaner mapping and logs, not from better prompts. It is also safer not to use high-risk data — meeting transcripts, collaboration messages, free-text performance reviews — as recommendation inputs without explicit notice, consent, access control and audit.
Why put standards like xAPI or LTI in the RFP?
Because if learning data lives only in a vendor-proprietary format, years of learning history and skill data vanish the moment you switch platforms. xAPI and the LRS are the format and store for recording learning activity beyond the portal; LTI is the specification for connecting external learning tools; Open Badges 3.0 and CLR 2.0 bundle learning and competency history into records that remain verifiable outside your organization. In what format, under whose ownership, and by what procedure data can be exported is safer treated as a contract clause than as a feature.
Should we pick a global LXP or a Korean LMS?
There is no public evidence that would let anyone declare one superior. Public materials do not disclose model architecture, feature definitions, recommendation metrics or fairness-testing methodology for any of these platforms. Decide by organizational condition instead. If statutory compliance evidence and Korean data-protection requirements come first, verify local certification (ISMS-P) and audit automation. If skill-based internal mobility comes first, verify skill taxonomy and open-standard mapping. The criteria are itemized in the LMS comparison checklist.
Next steps
Once the definitions are clear, the next task is measuring where your organization actually stands.
Which stage are we in?
The LMS/LXP health check assesses your current usage rate and content supply structure.
CompareCompare candidates item by item
The LMS comparison checklist sets out evaluation items and rating criteria by category.
DataSee the raw operating data
Usage patterns and MAU distribution across 100+ companies over 35 months.
