What problem does it solve?
Training content is expensive to make and quickly out of date. A single eLearning module takes an instructional designer many hours of scripting, storyboarding, question writing and production, and subject matter experts lose days recording themselves: a senior instructional designer at Zoom describes experts and designers spending an entire day recording to get about 15 minutes of video. When the product, the procedure or the regulation changes, the video has to be reshot, so outdated training stays in circulation.
New systems, compliance topics and multilingual workforces all need material, often in several languages and in accessible formats: Carlsberg, for example, used to hire a second agency to translate each eLearning. Public bodies feel the pressure too. The Veterans Benefits Administration uses an AI assistant to reduce instructor and instructional design burden amid decreased hiring abilities, and to cut classroom time, and the IRS turned to AI voices when return to office mandates meant it could no longer record narration with employees.
How does it work?
- Start from approved sources. The designer uploads the procedure, product documentation or policy the course must teach, and defines the audience and learning objectives.
- Draft the structure. The AI proposes an outline, lesson text and knowledge checks (quizzes and scenarios) mapped to the objectives.
- Produce media. Narration is generated from the script with synthetic voices, and short videos can use AI avatars instead of filmed presenters; images and sounds are generated or selected.
- Localise. Text, narration and subtitles are translated into the languages each market needs, as Carlsberg does for supply chain training.
- Review and publish. A subject matter expert checks accuracy, the designer assembles the module in the authoring tool, and the course is published to the learning platform.
- Update cheaply. When the source changes, the script is edited and the media regenerated instead of reshot.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- Internal tools
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Cycle time reduction | Too few to pool | 90% | 1 | 1 vendor |
| Hours saved | Not pooled | at least 15 hours | 1 | 1 vendor |
| Cost savings | Not pooled | Not pooled: up to USD 1500 | 0plus 1 up to | 1 vendor |
Value drivers: Employee productivity, Speed and cycle time, Lower cost to serve, Inclusion and access.
Indicative value
A learning and development team that builds or updates 200 eLearning modules a year
USD 80,000 to USD 576,000
Course production time released per year
How this is calculated
Formula: modules * hoursPerModule * timeSaved * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Modules built or substantially updated per year modules, modules per year | 200 | 200 | The reference team. Replace with your own volumes. |
| Design and production hours per module today hoursPerModule, hours per module | 40 | 80 | Editorial assumption covering scripting, questions, media and assembly; replace with your own. |
| Share of those hours saved with AI drafting and media generation timeSaved, fraction of hours | 0.2 | 0.4 | Conservative against the vendor reported 90% time savings on video creation at Zoom on this page, because video is only part of a module and expert review time does not shrink. |
| Blended instructional designer and expert hour hourlyCost, USD per hour | 50 | 90 | Editorial assumption. |
What it leaves out: Counts internal production time only. It leaves out translation savings, the value of training that is current instead of outdated, licence costs and the effect on learning outcomes. Synthesia reports at least €30,000 a year in avoided agency fees at Carlsberg, which the model does not include. Synthesia's reported $1,000 to $1,500 a month per employee saving at Zoom is a vendor estimate of internal production efficiency, which overlaps with what this model already counts, so it is not additional value. None of the deployments on this page has published effects on learning outcomes.
Who already uses it?
5 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
U.S. Marshals Service
United States · Government and public sector · 2025
Training content creators at the US Marshals Service, part of the Department of Justice, have piloted TechSmith Camtasia and Audiate since September 2025 to generate on screen presenters and realistic narration from text, and to produce voice audio in a range of voices and tones without recording actors. The goals are to shorten the time to release training material, support Section 508 accessibility and improve the online learning experience. No outcome figures are published.
No outcome disclosed.
Veterans Benefits Administration
United States · Government and public sector · 2025
The Veterans Benefits Administration uses the AI Assistant in Articulate 360 to create interactive eLearning for its claims processors more efficiently. The assistant generates text to voice audio, images, outlines, polls, quizzes and sounds for courses. The inventory entry says engaging eLearning reduces instructor burden and classroom time, that efficient training development reduces instructional design burden amid decreased hiring abilities, and that the time saved in developing training brings cost savings, without a figure. The inventory lists it as deployed; no outcome figures are published.
No outcome disclosed.
Internal Revenue Service
United States · Government and public sector · 2024
IRS training developers used to record course narration with employees and microphone kits; with return to office mandates that was no longer possible. Since May 2024 they enter narration scripts into a web based AI voice tool that returns audio files for import into eLearning authoring applications. Scripts contain no personal or taxpayer information and use fictitious names and addresses. The IRS reports better quality, a wider choice of voices and much faster generation and revision of voiceovers, but publishes no figures.
No outcome disclosed.
Carlsberg Group
Denmark · Manufacturing · 2023
Since 2023, Carlsberg's Integrated Supply Chain Academy has built training in house with Synthesia instead of external video agencies, and use has spread to shop floor onboarding, one point safety lessons, procurement training and change management content. A document, usually a PDF, goes into the tool's AI Assistant, which scaffolds the structure; the video is built in a branded template, generated in English and translated for each market. Synthesia reports that more than 100 employees have built content in under two years, that three agency shoots a year (its conservative baseline, at least €30,000 in fees) no longer sit on Carlsberg's P&L, and that the second supplier once hired to translate each eLearning is no longer needed. These are vendor stated figures, not a measured saving net of licence costs.
- Users served: at least 100, in under two years
"In under two years, more than 100 employees from across the business have built content in Synthesia, with adoption still growing."
Claimed by: vendor
Zoom
United States · Technology and software · 2023
Zoom's instructional designers train more than 1,000 salespeople on selling its products. When training material changed, whole videos had to be recorded again, with subject matter experts spending a day in front of a camera for about 15 minutes of usable footage. The team now produces AI avatar videos with Synthesia and builds them into interactive modules in Rise 360 and Storyline. Synthesia reports 90% time savings on video creation, more than 200 micro videos from one designer in about six months, 15 to 20 hours a month freed for Zoom's subject matter experts who no longer record themselves (the page does not say whether this is per expert or in total), and monthly cost savings of $1,000 to $1,500 per employee previously spent on creating training videos.
- Cycle time reduction: 90%
"90% time savings"
Claimed by: vendor - Hours saved: at least 15 hours, per month
"Time Saved for SMEs: Zoom's subject matter experts no longer need to record themselves, freeing up 15-20 hours each month to work on their actual job."
Claimed by: vendor - Cost savings: up to USD 1500, per month, per employee
"Enhanced Productivity: Thanks to AI video, both IDs and SMEs can work more efficiently. This results in monthly cost savings of $1,000 - $1,500 per employee previously spent on creating training videos."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Approved and current source material for each course
- Learning objectives and assessment standards
- Brand, tone and terminology guidelines, including approved translations
- Written consent for any avatar or voice modelled on a real employee
Systems to integrate
- eLearning authoring tools
- Learning management system
- Document management or knowledge base holding the source material
- Translation memory and terminology tools
Complexity: Low
Authoring tools such as Articulate 360 and video tools such as Synthesia and TechSmith Camtasia already include these features, as the deployments on this page show, and the source material usually exists. The effort goes into review workflows, keeping content tied to approved sources, consent for any real person's likeness or voice, and accessibility.
- 1
Pick content that changes often
Start where reshooting and rewriting hurt most: product training, system training and procedures. Zoom uses it for sales enablement, where videos had to be recorded again whenever the training material changed.
- 2
Ground drafts in approved sources
Generate from the procedure or documentation, not from the model's general knowledge, and keep a link from each lesson to its source so updates can be traced.
- 3
Keep experts in the review, not the recording
Move subject matter experts from recording to reviewing scripts and quizzes. That is where their time is best spent and where errors are caught.
- 4
Set rules for synthetic media
Decide which avatars and voices may be used, get written consent for any real person's likeness, label AI generated media, and check captions and transcripts for accessibility.
- 5
Measure learning, not only production
Track completion, assessment results and learner feedback against earlier versions, so faster production does not come at the cost of learning.
Guardrails
- Every course reviewed and approved by a named subject matter expert before publishing
- Drafts generated from approved sources, with a link to the source version
- AI generated video and audio labelled as such to learners
- No avatar or cloned voice of a real person without written consent
- Captions, transcripts and accessible formats checked before release
KPIs to instrument
- Hours from request to published module, before and after
- Expert review hours per module
- Errors found after publication per module
- Assessment scores and completion rates versus earlier versions
- Age of content (time since last update) across the catalogue
Human in the loop
Instructional designers direct and assemble the course, subject matter experts approve the content, and learning owners sign off assessments, especially any that count towards certification or role decisions.
Common failure modes
- Confident errors in the content
- The model fills gaps with plausible but wrong details. Generate from sources only and require expert sign off.
- More content, not better learning
- Production gets cheaper and the catalogue grows, but nobody checks whether people learn. Track assessment results and retire unused content.
- Consent and likeness problems
- An employee's face or voice is reused after they leave or without clear agreement. Keep consent records and prefer stock avatars.
- Uncanny or disengaging media
- Avatar videos that feel artificial lose learners' attention. Test with learners and mix formats.
What are the risks and rules?
EU AI Act
Depends on design
Generating training content is not listed in Annex III. Providers of tools that generate synthetic audio, image, video or text content must mark the output as AI generated (with an exception for assistive editing that does not substantially alter the source), and deployers must disclose deep fakes, such as an avatar or voice that resembles a real person and would falsely appear authentic (Article 50(2) and (4), with the definition in Article 3(60)). If the same system evaluates learning outcomes or decides access to training that affects a person's work, Annex III point 3 (education and vocational training) and point 4 (employment) must be checked, and those parts can be high risk.
Rules that apply
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Machine readable marking of synthetic audio, image, video and text content by providers, and disclosure of deep fakes (image, audio and video) by deployers.
- Annex III, high risk AI systems referred to in Article 6(2) (European Union, Europe). Point 3 covers AI that evaluates learning outcomes or decides access in education and vocational training; point 4 covers employment decisions.
- Section508.gov (US General Services Administration, North America). Accessibility requirements for information and communication technology at US federal agencies, which cover their training content; cited as a goal in the US Marshals Service pilot. Organizations elsewhere follow their own accessibility rules.
Controls to put in place
- Content ownership and review dates for every course
- Register of avatars and voices in use, with consent records
- Labelling policy for AI generated media
- Accessibility check in the publishing workflow
Frequently asked questions
- How much faster is course production with AI?
- For individual steps, much faster: Synthesia reports 90% time savings on training video creation by Zoom's instructional designers. For a whole module the saving is smaller, because experts still review the content. None of the deployments on this page has published effects on learning outcomes.
- Who uses it in the public sector?
- The Veterans Benefits Administration uses an AI assistant in its authoring tool to build eLearning for claims processors, the IRS generates course narration with AI voices, and the US Marshals Service is piloting AI presenters and narration for training videos.
- Do we need to tell learners that a video uses an AI avatar?
- Under the EU AI Act, providers must mark synthetic media as AI generated and deployers must disclose deep fakes, such as an avatar or voice that resembles a real person. Labelling all AI generated media is the simple policy.
How to cite this page
Blits.ai AI Use Case Library, "AI for creating employee training and eLearning content", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/training-content-generation. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published