FLYER Work: Enterprise LMS + Private AI

1Summary and recommendation

Corporate learning is a large, growing, and structurally dated market. The LMS software category alone is worth roughly US$15–19 billion in 2026 and growing about 20% a year, inside some US$400 billion of total corporate training spend. Buyers consistently complain about the same three things: opaque pricing, heavy administration, and slow rollouts. Meanwhile, the two vendors that actually combine an LMS with a private AI assistant over company documents (Docebo, generally available Fall 2026, and Workday Learning powered by Sana, generally available July 2026) sell only to 1,000-seat-plus enterprises on multi-year, quote-only contracts. Mid-market LMS vendors have no private AI at all.

Recommendation. Build FLYER Work as an enterprise-grade LMS with private AI included, aimed first at the mid-market and at regions Western vendors under-serve (EU mid-market, APAC and Southeast Asia), and sold on four things competitors cannot easily copy: one transparent all-in price, a 30-day guided go-live, region-pinned data and models, and a compliance pack that treats the EU AI Act as a feature. The product story inside that wrapper is function copilots that train as they answer: a Sales assistant, a Support assistant, a Compliance assistant, an Engineering assistant, a People assistant and a Finance assistant, each with its own knowledge scope and audience, whose unanswered questions become skill gaps, whose skill gaps become AI-drafted micro-courses reviewed by a subject-matter expert, and whose courses are cited back inside future answers.

What this document decides

• Positioning and the segment we attack first (section 3, 4, 11).
• The full feature set, split into table stakes required to pass an international RFP and the differentiators we lead with (section 5).
• The detailed specification of the private AI assistant per function, the core of the product (section 6).
• Seven product screens in the Cloud theme so the team can see the product, not just read about it (section 7).
• Architecture, non-functional requirements, pricing, and a three-phase roadmap (sections 8–12).

2Market research

Research was carried out on 06 September 2026 across analyst publications, vendor announcements, pricing pages, marketplace listings and review sites. Verified facts are cited in section 14. Where a figure is an inference it is marked as such.

2.1 Size and growth

The three research houses disagree by about 30% on absolute size, so the LMS software figure should be used with investors and the US$400B total-spend framing with buyers. No credible Southeast Asia-specific corporate LMS figure exists; the region is small in absolute terms but has almost no local presence from the vendors below.

2.2 Competitors and their AI plays

2.3 What buyers require to sign internationally

• Security. SOC 2 Type II is the US procurement gate and "cannot be substituted by ISO 27001 alone"; buyers want the vendor's own certification, not the cloud host's. ISO 27001 is the EU and APAC equivalent. HIPAA BAA for US healthcare, FedRAMP only for US federal.
• Content standards. SCORM 1.2 and 2004 are a pass/fail filter; xAPI is table stakes; cmi5 for new mobile and offline content; LTI 1.3 for third-party tools.
• Identity and HR data. SAML 2.0 and OIDC single sign-on, SCIM 2.0 provisioning, connectors to Workday, SAP SuccessFactors, BambooHR and ADP.
• Data residency. Docebo offers US, EU, UK, Canada, India and Australia; Cornerstone offers US, UK, France and Germany with one hosting location per customer. Singapore is the de-facto APAC anchor.
• Accessibility and language. WCAG 2.1 AA is the baseline; WCAG 2.2 and the European Accessibility Act (June 2025) now appear in RFPs. Docebo ships 50+ interface languages; AI translation of content is now expected.
• Frontline and offline. No-email login (QR, phone number, employee ID), offline download and sync, modules under ten minutes. Fosway names frontline as under-served.
• Regulated industries. 21 CFR Part 11 for pharma and medtech: immutable audit trail, e-signatures, validation documentation. Extended enterprise portals for partner and customer training.

2.4 What security and L&D buyers demand from private AI

• Zero data retention and no training at the model provider, written into the contract. Docebo and 360Learning both publish "never trains on your data".
• Tenant isolation that is architectural, not policy-based.
• Bring-your-own model and regional endpoints. Only Sana offers model choice today; Docebo does not disclose its LLM. Data processing agreements should name regional processing endpoints.
• Permission-aware retrieval. Most RAG implementations ignore document permissions; buyers now ask for access-control lists enforced at retrieval time and refreshed continuously (Glean, Azure AI Search and Docebo do this).
• Audit logs, citations, guardrails. Every prompt logged, every answer traceable to a source, safety-tested responses.
• EU AI Act, binding for high-risk systems from 2 August 2026. Annex III 3(b) classifies AI that evaluates learning outcomes or steers the learning process in vocational training as high-risk, with obligations on risk management, data governance, logging, human oversight and worker notice, and fines up to €15M or 3% of turnover. It applies extraterritorially to any vendor serving EU workers. A pure question-and-answer assistant is lower risk; AI grading and skills inference need a "high-risk-ready" documentation pack.

2.5 Pricing benchmarks

Inference. There is a five-to-ten-fold price gap between "LMS" ($5–10) and "AI knowledge assistant" ($30–65) per user per month. A bundle at $12–20 that covers both is the clearest anomaly in the data, and SAP's move to include generative AI at no extra cost signals that AI will stop being an add-on within two years.

2.6 What analysts say the next wave is

• Josh Bersin (Feb 2026): only about 5% of organisations reach "dynamic enablement", a platform that houses all company knowledge with content generated on demand and inferred skills; early AI-native adopters cut L&D spend 40–50%; SCORM-era LMSs "face replacement by dynamic content systems".
• Fosway 9-Grid 2026: skills are the number one buyer priority and AI a close second; the standalone learning experience platform is dead; "delivered AI reality still lags market hype"; personalised recommendations and good UX are baseline, not differentiators; frontline and cohort learning are under-served.
• Brandon Hall, eLearning Industry: agents in the flow of work, skills-based learning, and the move from monolith to an interconnected ecosystem.

2.7 White space

1. The mid-market has no LMS-plus-private-AI option. Docebo and Workday/Sana start at 300 to 1,000 seats with multi-year contracts; TalentLMS, LearnUpon and Litmos have no knowledge assistant.
2. Per-function copilots exist only in adjacent tools (Seismic for sales, Glean for general search). Nobody ships pre-configured function assistants wired to learning completions.
3. Transparent pricing. Every enterprise vendor is quote-only with undisclosed AI credits; total cost of ownership opacity is the top buyer complaint.
4. Time to value. Implementation is 20–40% of year-one cost at Docebo; administrative complexity is the number one review complaint for Docebo and Cornerstone.
5. Skills inferred from questions asked. Analysts describe it; no vendor productises it. Cornerstone and Docebo infer skills from HR data, not from what people ask and cannot find.
6. EU AI Act readiness by design is a compliance moat few vendors articulate.
7. Non-English, APAC and frontline (Vietnamese, Thai, Bahasa, Japanese; offline mobile) are explicitly under-served and outside the North American vendors' focus.

3Positioning and USPs

3.1 Three candidate positions

3.2 Unique selling points

3.3 Messaging by audience

What FLYER Work is not. It is not a content library (we integrate Udemy Business, Coursera and LinkedIn Learning rather than compete with them), not a talent suite (no performance reviews, compensation or succession), and not a general enterprise search product (assistants are scoped to functions and to learning outcomes).

4Target customers and personas

4.1 Ideal customer profile

4.2 Personas

5Product scope and feature set

The product is organised into nine modules. Each feature below is tagged as table stakes (needed to pass an international RFP) or differentiator (what we lead with), and with the roadmap phase in which it ships (see section 12).

5.1 Learning core

5.2 Compliance engine

5.3 Private AI assistants

Specified in full in section 6. Summary of scope:

5.4 Knowledge connectors

5.5 AI Course Studio

5.6 Skills and analytics

5.7 Platform, integrations and extended enterprise

6Private AI assistants per function

6.1 Concept

A company does not have one body of knowledge; it has a Sales body, a Support body, a Legal body, an Engineering body. People in one function rarely need, and often must not see, the others. FLYER Work therefore ships one assistant per business function, each an isolated configuration of knowledge scope, audience, tone, tools and policies, and each with its own quality dashboard owned by the function head. A generic "ask anything" assistant is available but off by default.

Each assistant answers only from approved sources within its scope, filtered at query time by the asking person's own permissions in the source systems. It cites every claim. When it cannot find an approved source it says so, logs the question as a knowledge gap, and offers to route it to the topic owner. Those gaps are the raw material for training.

6.2 Anatomy of an assistant

6.3 Retrieval and answer pipeline

6.4 Guardrails and policies

6.5 The learning loop

This is what makes an LMS plus an assistant more than two products in one login.

1. Question. An employee asks the function assistant. The answer is cited, or the assistant says there is no approved source.
2. Gap. Unanswered, low-confidence and repeated questions are clustered into topics and shown to the admin and function owner with counts and source status (no source, stale source, conflicting sources).
3. Draft. The owner clicks "Draft micro-course" or "Request document". Course Studio generates lessons, a scenario and a quiz from the approved sources, every statement traced to a paragraph, or the topic owner is asked for the missing document.
4. Review. A named SME approves, edits or rejects. Nothing publishes without a reviewer.
5. Assign. The course goes to the audience whose questions revealed the gap, and to new joiners in that role by rule.
6. Measure. The assistant cites the course in future answers; question volume on the topic falls; skill scores for the topic rise. The dashboard shows the loop closing.
7. Maintain. When a source document changes, the linked course is flagged and re-drafted for review.

6.6 Default assistant packs

6.7 Quality measurement

6.8 Acceptance criteria for phase 1

• A person who lacks read access to a document in SharePoint never receives text from that document in an answer, verified by an automated permission test suite run on every deploy.
• Every answer that makes a factual claim shows at least one citation resolving to a chunk the asker can open.
• A question with no supporting evidence above threshold returns the "no approved source" response, is logged as a gap, and appears in the admin dashboard within one minute.
• Removing a document from the source or revoking a user's access removes it from answers within 24 hours, verified by test.
• Tenant admin can switch the model provider and region and the change is reflected in the answer provenance panel.
• Course Studio produces a five-lesson micro-course with quiz from three documents in under two minutes, with every statement linked to a source paragraph.

7Product design: seven screens

The screens below are live HTML mockups in the Cloud theme: Inter typeface, hairline borders, dense tables, FLYER purple as the only accent. They are embedded at scale; open any screen full size to read it. The demo tenant is a fictional 4,200-person manufacturer hosted in the EU region.

The L&D director's landing page. Four KPI tiles including AI assistant volume, the compliance programs at risk with owners and overdue counts, AI questions by function, and the knowledge gaps the assistants detected with one-click actions to draft a micro-course or request the missing document.

An account executive's view. Up-next path, due items, assigned training including an AI-recommended micro-course marked as reviewed by Legal, the function assistant with suggested prompts, and the skills profile against role targets.

Six function assistants, each with audience, source count, seven-day volume, grounded rate, unanswered rate and the model it runs on. Right: tenant AI posture with locked policies, and the top unanswered topics with actions.

The Sales Assistant answers a warranty question with numbered citations, states that two restricted documents were excluded by permissions, and offers to turn the answer into a micro-lesson or escalate to Legal. The right panel shows the sources, a recommended course, and how the answer was produced: permission filter, scope, model, retention, logging.

Connectors with the assistants they feed, document counts, last sync and permission status. Right: data handling for the tenant (region, model endpoint, encryption, purge time, never used for training) and document health: stale, conflicting and missing sources.

A micro-course drafted from three approved sources because 37 Sales questions went unanswered. Each lesson shows its grounding and review status; the SME reviewer, source-change alerts, translation on publish and SCORM/xAPI export are visible before publishing.

Everything a security review asks for on one page: certifications and frameworks, tenant-wide AI policy owned by the CISO, access and provisioning, supported standards, and today's audit log including AI questions and provisioning events. An evidence pack exports for auditors.

8Architecture

Multi-tenant software as a service deployed as independent regional cells, with an optional single-tenant tier for large enterprises. The design goal is that a security reviewer can verify tenant isolation, data residency and model routing from the console, not from a whitepaper.

Regional cells

Key technical decisions

• Tenant isolation is enforced at three layers: separate database schemas with row-level security, separate index namespaces, and per-tenant encryption keys. No cross-tenant query path exists in the assistant service.
• Permissions are mirrored, not re-implemented. The connector records the source system's ACL with every document and refreshes group membership continuously; retrieval filters by the asker's mirrored groups before ranking. Documents whose ACL cannot be resolved are excluded rather than exposed.
• Model gateway abstraction from day one, so switching provider or region is configuration. Embedding and re-ranking models run inside the cell so document text never leaves the region for indexing.
• xAPI as the internal event language. Every completion, answer rating and assistant interaction becomes a statement, which makes the built-in LRS, reporting and third-party analytics consistent.
• Reuse from FLYER. Assignment engine, question bank, quiz player, live-session logic and AI grading rubrics from the existing FLYER platforms are candidates for reuse; the corporate product needs a separate tenant model, brand and compliance posture.

9Security, compliance and non-functional requirements

9.1 Certification roadmap

9.2 Service levels and scale

9.3 EU AI Act position

The question-and-answer assistants are transparency-obligation systems: employees are told they are interacting with AI, answers are marked, and logs are kept. Features that evaluate learning outcomes or steer the learning process for vocational training (AI-scored practice, automated assignment from inferred skills) fall under Annex III 3(b) and are built high-risk-ready: human-in-the-loop by default, decisions explainable, logging retained, model cards published, a DPIA template supplied, and a worker information notice template supplied. Individual-level skills inference is off by default and needs data-protection-officer acknowledgement to turn on. This position should be confirmed with external counsel before launch.

10Pricing and packaging

Pricing is published, per active user per month, billed annually, with AI included. An active user is anyone who logs in, completes training or asks an assistant in a calendar month. Fair-use AI allowance is included; overage is priced per question and visible in the console before it is billed.

How it compares

Unit economics guardrail. At 60 questions per user per month and roughly 6,000 tokens per grounded answer, model cost is about $0.30–0.60 per user per month on current frontier pricing, and lower with routing to smaller models for simple questions and caching. That leaves room in the $15 tier even before volume discounts from providers.

11Go-to-market

11.1 Land with one function, expand to the company

The wedge is a 14-day private AI pilot for one function (usually Sales or Support) plus the compliance training that function already owes. It connects one or two sources, stands up one assistant, and shows the knowledge-gap report at the end of week two. The pilot converts into Learn + Private AI for that function, then expands to the company at the LMS renewal date.

11.2 Sequence of markets

11.3 Competitive plays

• Versus Docebo and Cornerstone: price transparency, 12-month terms, 30-day go-live, no implementation fee; show the admin home side by side.
• Versus Workday/Sana: we serve companies below 1,000 seats and companies not on Workday; model choice and region pinning at a published price.
• Versus Glean and Copilot: we are the training system of record with assignments, compliance and SCORM, and we expose our knowledge to Copilot through MCP rather than competing with it. The Copilot rollout is a trigger event, not a threat.
• Versus mid-market LMSs: they have no private AI; we include it.

11.4 Assets needed before first sale

• Public trust center with certification status, sub-processor list, architecture summary and DPA download.
• RFP answer library covering the requirement list in section 2.3.
• Pricing page with calculator; marketplace listings.
• Three design-partner case studies (one per initial region) with before and after on compliance completion, ramp time and question deflection.
• Product name and brand separate from FLYER's K-12 English identity; "FLYER Work" is a placeholder.

12Roadmap and success metrics

Success metrics

13Risks and open questions

Open questions for the review

1. Product name and brand: separate brand, or a FLYER sub-brand such as "FLYER Work"?
2. Build the learning core fresh or extract the assignment, question-bank and grading services from the current FLYER platform?
3. Cloud and default model provider for the EU cell: Azure (simplest Azure OpenAI story for enterprise buyers) or AWS (Bedrock, Singapore presence)?
4. Which three design partners, and do we offer them a founding-customer price in exchange for case studies?
5. Do we ship the "Learn" tier without AI at all, or make AI the only story and simplify to two tiers?
6. Who owns the assistant quality function (a product role, a customer success role, or the customer's SMEs alone)?

14Sources

Research conducted 06 September 2026. Figures from different research houses vary by up to 30%; paywalled analyst grids (Gartner Hype Cycle 2025, Fosway placements) were not accessed directly.

• Precedence Research, Corporate LMS market: https://www.precedenceresearch.com/corporate-learning-management-system-market
• The Business Research Company, Corporate LMS global market report: https://www.thebusinessresearchcompany.com/report/corporate-learning-management-system-lms-global-market-report
• Grand View Research, Corporate LMS market: https://www.grandviewresearch.com/industry-analysis/corporate-learning-management-system-lms-market-report
• Grand View Horizon, APAC corporate e-learning: https://www.grandviewresearch.com/horizon/outlook/corporate-e-learning-market/asia-pacific
• Josh Bersin, How AI transforms $400 billion of corporate learning (Feb 2026): https://joshbersin.com/2026/02/new-research-how-ai-transforms-400-billion-of-corporate-learning/
• Josh Bersin, The enterprise learning tech market transforms around AI: https://joshbersin.com/2026/02/the-enterprise-learning-tech-market-quickly-transforms-around-ai/
• Fosway 9-Grid for Learning Systems 2026 (via Learning News): https://learningnews.com/news/fosway/2026/make-better-learning-systems-decisions-the-9-grid-for-learning-systems-is-out
• Docebo AgentHub launch: https://www.docebo.inc/news/news-details/2026/Docebo-Launches-Docebo-AgentHub-and-Unites-Skills-Intelligence-Enterprise-Knowledge-and-Agentic-AI-in-a-Single-Platform/default.aspx
• Docebo Enterprise Knowledge: https://www.docebo.com/products/enterprise-knowledge/
• Docebo FY2025 results: https://www.docebo.inc/news/news-details/2026/Docebo-Reports-Fourth-Quarter-and-Fiscal-Year-2025-Results/default.aspx
• Docebo pricing on Vendr: https://www.vendr.com/marketplace/docebo · Compono, Docebo true costs: https://www.compono.com/articles/docebo-pricing-2026-true-costs
• Docebo data processing addendum (regions): https://www.docebo.com/company/data-processing-addendum/
• Cornerstone Workforce AI launch: https://www.cornerstoneondemand.com/company/news-room/press-releases/cornerstone-launches-cornerstone-workforce-ai-the-intelligence-platform-for-workforce-readiness-built-to-amplify-human-potential-exponentially-with-ai/
• Cornerstone on AWS Marketplace (Learn+, Elevate+ pricing): https://aws.amazon.com/marketplace/pp/prodview-6kgfjl7vrgfpm
• Cornerstone hosting locations: https://www.cornerstoneondemand.com/legal/hosting-and-subprocessors/
• Workday Learning powered by Sana, general availability: https://newsroom.workday.com/2026-07-22-Workday-Learning,-Powered-by-Sana,-Now-Generally-Available-as-an-AI-Native-Learning-Experience-Built-on-Workdays-Trusted-Data
• Sana Agents pricing: https://sanalabs.com/agent-pricing · Sana pricing analysis: https://www.educate-me.co/blog/sana-labs-pricing
• Absorb Aura launch: https://www.globenewswire.com/news-release/2026/05/19/3297575/0/en/absorb-software-launches-absorb-aura-the-ai-engine-powering-the-agentic-learning-system-that-drives-the-business-forward.html
• SAP SuccessFactors 1H 2026 release, Joule: https://clarity-solutions.de/en/insight/sap-successfactors-1h-2026-release-business-ai-joule-guide/
• 360Learning AI course creation: https://support.360learning.com/hc/en-us/articles/18756214814228-Create-a-course-using-AI · 360Learning AI: https://360learning.com/product/artificial-intelligence/
• TalentLMS pricing: https://www.educate-me.co/blog/talentlms-pricing · LearnUpon pricing: https://www.compono.com/articles/learnupon-pricing-2026-buyers-guide · Litmos pricing: https://www.compono.com/articles/litmos-pricing-buyers-guide-2026
• Skillsoft CAISY growth: https://www.stocktitan.net/news/SKIL/ai-powered-personalized-learning-surges-146-as-employees-build-sjko4n257q24.html · Degreed Maestro: https://degreed.com/experience/artificial-intelligence/
• Content library pricing: https://trainingcost.com/lms-pricing/udemy-business · https://trainingcost.com/lms-pricing/coursera-business · https://trainingcost.com/lms-pricing/linkedin-learning
• Continu AI: https://www.continu.com/features/ai · Arist AI course creation: https://arist.com/product/ai-course-creation · Seismic Aura agents: https://www.seismic.com/newsroom/press-releases/seismic-aura-agents-across-gtm-workflows/
• Microsoft Learning Agent in M365 Copilot: https://m365admin.handsontek.net/viva-learning-copilot-experience-transitions-learning-agent-microsoft-365-copilot/
• Glean pricing: https://coworker.ai/blog/glean-pricing
• Review themes: https://www.g2.com/products/docebo/reviews?qs=pros-and-cons · https://www.g2.com/products/cornerstone-learning/reviews · https://lmschef.com/absorb-lms-review/ · https://www.g2.com/products/360learning/reviews?qs=pros-and-cons · https://peoplemanagingpeople.com/tools/talentlms-review/
• SOC 2 for LMS vendors: https://lmspedia.org/soc-2-type-ii-compliance-for-lms-vendors/
• LMS RFP requirements: https://www.d2l.com/blog/lms-rfp-template/ · E-learning standards: https://www.eleapsoftware.com/glossary/e-learning-standards-a-comprehensive-guide-to-scorm-xapi-cmi5-and-modern-lms-compliance/
• APAC data residency 2026: https://www.functioneight.com/blog/apac-data-residency-requirements-2026/
• Frontline mobile LMS requirements: https://www.opus.so/blog/best-mobile-lms-frontline-teams
• Zero data retention: https://decagon.ai/glossary/what-is-zero-data-retention-ai · Tenant isolation for agentic AI: https://www.parloa.com/knowledge-hub/data-isolation-agentic-ai/
• Permission-aware RAG: https://truto.one/blog/how-to-maintain-document-level-rbac-in-enterprise-rag-pipelines/ · https://www.datarobot.com/blog/acl-hydration/
• EU AI Act Annex III: https://artificialintelligenceact.eu/annex/3/ · https://www.aiactblog.nl/en/annex-iii/onderwijs-beroepsopleiding/learning-outcomes · CSA research note on the high-risk deadline: https://labs.cloudsecurityalliance.org/research/csa-research-note-eu-ai-act-high-risk-compliance-deadline-20/ · Ogletree on US employers: https://ogletree.com/insights-resources/blog-posts/cybersecurity-awareness-month-in-focus-part-iii-the-eu-ai-act-is-here-what-it-means-for-u-s-employers/
• LMS pricing benchmarks: https://softwarefinder.com/resources/lms-pricing · https://www.educate-me.co/blog/lms-pricing · https://www.theworkademy.com/lms-pricing/
• Brandon Hall trends: https://brandonhall.com/the-intelligent-learning-organization-trends-challenges-and-predictions-for-the-year-ahead/ · eLearning Industry 2026 trends: https://elearningindustry.com/top-learning-technology-trends-for-2026

Design language: Cloud theme. Screen mockups are HTML in company-lms/frames/. Demo company, people and figures on the screens are fictional.