Docebo
The most transparent AI stack we have scored in this category, with every model provider named, three vendor guardrail layers switched on and a working MCP server that writes as well as reads. The catch is that the headline agent product was announced in April and still is not shipped, the AI credit meter went live in January, and customer support scores 1.9 out of 10 on TrustRadius against 8.0 for ease of use.
PRICING
| Elevate | Quote only | Mid-market tier. Core learning platform, up to 4 included integrations, standard onboarding. Advanced Analytics and Branded Mobile App are explicitly NOT available at this tier and cannot be bought as add-ons. |
| Enterprise | Quote only | Up to 6 included integrations, up to 10 extended-enterprise branded domains, 2 sandbox environments, Advanced Analytics, Branded Mobile App, Skills Intelligence. Verified buyer quote of $70,000/year for 1,000 users. |
| Add-ons | Quote only | Extended Enterprise, E-Commerce, Gamified Learning, Immersive Learning, AI Virtual Coaching, Docebo Labs by Skillable, Microsoft Teams integration and Salesforce integration are all priced separately and none of them publish a figure. |
| AI credits | Consumption, rate not published | Live since 05/01/2026. Published consumption: AI virtual coaching 110 credits per video scenario, 55 voice, 10 chat; AI video presenter 210 credits per minute; AI-assisted lesson generation 10 credits per lesson. Core AI features (assessment generation, translations, Harmony search and copilot, recommendations) consume nothing. Running out does not disable anything; overage is tracked and invoiced at a pre-agreed rate. |
| Docebo Federal | Quote only | Separate FedRAMP Moderate authorised platform on AWS GovCloud with Okta identity, supporting data up to IL3, FIPS 140-2 and 140-3 validated encryption. |
Docebo publishes no numbers, and the active-user model is the variable that decides whether the deal is good or bad. MAU billing rewards seasonal training patterns and punishes an organisation whose whole workforce logs in every month; YAU counts a learner once per contract year. Exceed the contracted count and Docebo bills the extra users rather than forcing an upgrade, but at about 30% overage a tier move usually becomes the cheaper path. Two further line items matter: implementation, which third-party analysts put at 10-40% of first-year cost beyond the included 8-week Essential onboarding and 100,000-record migration limit, and AI credits, which turned a previously flat AI feature set into a metered one in January 2026 with no published per-credit price. Multi-year deals typically carry 3-5% annual escalation.
checked 2026-08-13 · vendor pricing page
Element scores
Strengths
Docebo does two things almost no enterprise vendor does, and both are worth money to a buyer. First, it names its models. The help centre publishes a per-feature table listing Anthropic and Nova via AWS Bedrock, OpenAI via Microsoft Azure, Gemini and Imagen via Google Vertex AI, and Docebo's own refined open-source models on AWS, alongside which features use generative AI and which do not. Second, it names its guardrails: AWS Bedrock Guardrails, Azure OpenAI Content Filter and Google Vertex AI Content and Safety Filter, running as three parallel layers, with per-feature disable for everything except global search. That transparency sits on top of a compliance stack that is unusually complete for the price band, covering ISO 27001, 27017, 27018, 27701 and 9001, SOC 2 Type II, PCI DSS Level 2, FDA 21 CFR Part 11 and EU Annex 11 with a published GxP guidance document, the EU-US, UK and Swiss Data Privacy Frameworks, and FedRAMP Moderate authorisation for the separate Docebo Federal platform on AWS GovCloud. The MCP server, generally available since July 2026, is one of the better implementations we have scored: OAuth2 across ChatGPT, Claude, Copilot Studio and Gemini Enterprise, learner tools for progress, certifications, search and content fetch, admin tools that create courses, manage bulk enrolments and curate Go1 content, a configurator that lets a Super Admin compose persona-scoped servers from a tool library with per-tool allowlists, and enforcement of Docebo's own user-level permissions on every call so an assistant can never surface something the user could not already open. Write actions can be flagged so the AI client asks for confirmation first. The multimodal layer is genuinely delivered rather than roadmapped, with avatar-led video presenters via Colossyan, chat, voice and video roleplay via Tavus with AWS and Azure speech services, instant feedback on tone and empathy, lesson voiceover in the learner's language, and lessons converted into two-host interactive podcasts. And the business behind it is healthy: $255.1M ARR at 30/06/2026, up 9.5%, average contract value up 27% to $74,800, adjusted EBITDA at 16.4% of revenue, a 93% multi-year renewal rate, and a $70M buyback announced in August 2026.
Honest dings
The headline product is not shipped. AgentHub, announced on 21/04/2026 as the centrepiece of what Docebo called its most significant release ever, is still described in the help centre and on the product page as planned for Fall 2026, and so is Enterprise Knowledge, the 20-plus-connector layer AgentHub runs on. Three of the four elements that would lift Docebo's agentic scores depend on software a buyer cannot use today, which is why the Agents and Multi-agent scores here are low while the marketing reads high. Second, the meter. On 05/01/2026 Docebo moved AI virtual coaching, AI video presenter and AI-assisted lesson generation onto consumption credits at 110, 210 and 10 credits respectively, and publishes none of the per-credit or package prices. When credits run out nothing stops working; usage is simply tracked and invoiced at a pre-agreed rate, which is convenient operationally and a budget hazard without disciplined monitoring. Third, and most consistently reported, support. TrustRadius scores Customer Service at 1.9 out of 10 against roughly 8.0 for ease of use, a gap of that size on the same product from the same reviewers, while Capterra shows 4.2 out of 5. Reviewers describe support as slow, difficult to reach and with issues dragging on for weeks. Any enterprise buyer should negotiate explicit support SLAs and take references from organisations their own size before signing a 3-5 year term. Fourth, the admin experience. Setup is repeatedly described as complex and requiring technical help, legacy admin pages as clunky and tedious, the REST API as challenging and intimidating on pagination, batch operations and rate limits, and manager dashboards as inadequate for tracking overdue training. Reviewers also report that Microsoft Teams sessions lack full completion tracking that the Zoom integration has, which matters if Teams is your virtual classroom. Fifth, the eval gap. Docebo tests its own models internally and says so, but gives customers no way to measure whether an answer was right, which for a platform that will soon answer compliance questions autonomously is a hole worth watching. Finally the lock-in: 250-learner minimum, one-year floor with most customers on 3-5 years, and a 93% multi-year renewal rate that reads as stickiness from the vendor side and as a long commitment from yours.
Sources (19) — every claim traceable
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- help.docebo.com/hc/en-us/articles/22706749501842-Ov… — Official AI transparency reference and the primary evidence for the LLM, Guardrails, Interpretability and Multimodal scores: per-feature table naming Anthropic and Nova via AWS Bedrock, OpenAI via Microsoft Azure, Gemini and Imagen via Google Vertex AI, Docebo-refined open-source models on AWS, Colossyan for AI video presenter, Tavus with AWS Transcribe, AWS Polly and Azure AI Speech for roleplay, AWS OpenSearch for search and Harmony. Also states the three vendor guardrail layers, the customer's ability to disable any AI feature except global search, and internal model assessment plus post-release automated testing (accessed 2026-08-13)
- developer.docebo.com/docs/supported-mcp-tools — Official MCP tool reference (24/07/2026). Source of the Protocols and Function calling scores: learner tools (Get my learning enrollments, Get my learning certifications, Search learning items, Fetch learning item, Ask learning data insights), Content marketplace curation tools gated on an eligible Go1 plan, the note that Fetch learning item reuses Harmony's parsing engine and counts toward Active Users, and Docebo's own server-design guidance against mixing learner and admin tools (accessed 2026-08-13)
- developer.docebo.com/docs/configuring-the-docebo-mc… — Official configuration guide: OAuth2 app registration, per-client redirect URIs for ChatGPT, Claude, Microsoft Copilot Studio and Gemini Enterprise, the MCP configurator with a locked built-in Learner server plus custom servers carrying their own path and tool allowlist, and the server URL scheme (accessed 2026-08-13)
- docebo.com/whatsnew/2026-july — Official July 2026 release notes confirming MCP Server general availability with admin tools for enrolments, courses, learning plans and Go1 curation across Claude, ChatGPT, Copilot and Gemini, and the new persona-scoped configurator (accessed 2026-08-13)
- community.docebo.com/product-q-a-7/q-a-july-2026-wh… — Official Docebo community Q&A (21/07/2026) and the source of two documented limitations behind the Protocols score: admin MCP tools are Super Admin only with manager tools planned for a future release, and a known OAuth limitation around automatic redirect to an organisation's SSO login page. Also confirms write actions can require explicit user confirmation and that the server enforces Docebo user-level permissions (accessed 2026-08-13)
- docebo.com/company/newsroom/docebo-launches-docebo-… — Official Inspire 2026 announcement, 21/04/2026: AgentHub, Enterprise Knowledge with 20+ connectors, Skills Intelligence via the 365Talents acquisition, MCP Server in public beta with July GA, Companion and AI Tutor. Primary evidence that the agentic layer was announced rather than shipped (accessed 2026-08-13)
- docebo.com/products/agenthub — Official AgentHub product page, still marked Coming Fall 2026 as at this audit date, describing 20+ enterprise connectors, 24/7 autonomous operation, per-prompt LLM selection, a reasoning engine and reinforcement learning from user feedback. The gap between this page and what a customer can switch on today is the basis for the Agents score (accessed 2026-08-13)
- help.docebo.com/hc/en-us/articles/31318530800786-AI… — Official AI credit documentation, effective 05/01/2026. Source of the Small models and pricing findings: 110 credits per video coaching scenario, 55 voice, 10 chat, 210 per minute of AI video presenter, 10 per AI-assisted lesson, with assessment generation, translations, Harmony search and copilot and recommendations free. Confirms the platform keeps working on overage and bills at a pre-agreed rate (accessed 2026-08-13)
- docebo.com/pricing — Official pricing page: Elevate and Enterprise tiers only, priced on product tier plus active users under MAU, YAU or RAU models, 250-learner guidance, one-year minimum with most customers on 3-5 years, overage billed rather than forced upgrades with a tier move sensible around 30%, 400+ integrations, 8-week Essential onboarding, and the 30M+ learners figure (accessed 2026-08-13)
- docebo.com/company/compliance-security — Official compliance page and the evidence behind the Guardrails score: ISO 9001 and ISO 27001, 27017, 27018 and 27701 certificates with numbers, annual SOC 2 Type II under AICPA/ISAE 3000, PCI DSS Level 2 service provider attestation, FDA 21 CFR Part 11 and EU Annex 11 support with electronic signatures and audit trails, GDPR, UK, PIPEDA and US state law handling via the DPA, and EU-US, UK Extension and Swiss-US Data Privacy Framework certification (accessed 2026-08-13)
- docebo.com/federal — Official: Docebo Federal holds FedRAMP Moderate authorisation, is hosted on AWS GovCloud with Okta identity, supports data up to IL3, and uses FIPS 140-2 and 140-3 validated encryption (accessed 2026-08-13)
- docebo.inc/news/news-details/2026/Docebo-Reports-Se… — Official Q2 2026 results, 07/08/2026: ARR $255.1M up 9.5%, Average Contract Value $74,800 up 27.0%, revenue $68.65M, adjusted EBITDA $11.23M at 16.4% of revenue, largest OEM customer down to 2.5% of ARR from 8.4%. Used for the users_official and pricing figures (accessed 2026-08-13)
- help.docebo.com/hc/en-us/articles/28459932954898-AI… — Official Shape authoring documentation: summarisation, keyword extraction and semantic image matching on Docebo's own models, AWS Translate for text translation and AWS Polly for audio generation, with generative AI flagged per capability (accessed 2026-08-13)
- community.docebo.com/product-updates/introducing-in… — Official product update (19/01/2026) behind the Multimodal score: lifelike video avatars reacting in real time, two-way adaptive conversation, a scenario builder with 50+ avatars, instant AI feedback on tone, empathy and clarity, and confirmation that from 01/01/2026 AI Virtual Coaching usage draws down AI credits (accessed 2026-08-13)
- getcor.ai/blog/reviews/docebo — Independent review synthesis (28/06/2026) across G2 (740+ reviews, 4.3/5), Capterra (230+, 4.4/5), TrustRadius and Vendr. Source of the support and admin criticisms: Customer Service as low as 1.9/10 on TrustRadius against roughly 8.0 ease of use, support described as slow and difficult to reach with issues taking weeks, complex initial setup, steep admin learning curve, clunky legacy admin pages, limited branding and reporting customisation, and an API described as challenging and intimidating on pagination, batch operations and rate limits (accessed 2026-08-13)
- checkthat.ai/brands/docebo/reviews — Independent review aggregation (30/03/2026) corroborating the support gap across platforms (Capterra 4.2/5 versus TrustRadius 1.9/10 for Customer Service), plus the Microsoft Teams finding that virtual sessions lack full completion tracking that Zoom has, inadequate manager dashboards for overdue training, reports of features seeming abandoned, and the notifications rollout that caused unexpected mass email sends (accessed 2026-08-13)
- costbench.com/software/learning-management/docebo — Independent procurement data: median annual contract around $40,000 from 138 verified purchases, typical band $21,000-86,000, 1,000+ user deployments quoted around $70,000/year, observed $5-10 per active user per month, the Engage entry tier discontinued for new customers in mid-2025, implementation often 20-40% of first-year cost, and 3-5% typical annual escalation (accessed 2026-08-13)
- compono.com/articles/docebo-lms-enterprise-review-2… — Independent review (28/07/2026) used for the lock-in and fit findings: enterprise contracts averaging more than three years with the largest beyond five, a 93% multi-year renewal rate, typical entry around £20,000-30,000+ per year, and the observation that as a standalone LMS Docebo does not connect learning data to broader HR metrics (accessed 2026-08-13)
- peopleopsclub.com/software/docebo — Independent profile used for the two structural limitations worth flagging to regulated buyers: compliance automation is lighter than Cornerstone OnDemand, SAP Litmos or Absorb for multi-level approval chains, conditional certification paths and granular audit trails; and Docebo is a pure-play LMS with no performance management, succession planning or recruiting (accessed 2026-08-13)