Should AI Training Be Online, On-Site or Hybrid?

TL;DR:

• For most Kenyan teams new to AI, a good default is to hold the first session in person, then run shorter follow-ups live online.
• Online works well for people who already work online, are spread across sites and have reliable internet. Research suggests the teaching method matters more than the format.
• Choose fully on-site when people need help at their screens, when logins are likely to cause trouble or when staff have weak internet at home.
• Avoid putting some people in the room and others on a video link in the same session. Remote learners are often left out, and sound is hard to get right.
• Test the connection, plan for power cuts and decide who pays for data before you choose.

Who this guide is for

This guide is for HR managers, learning and development leads and team leaders choosing how to deliver AI training. It helps you pick a format before you pick dates.

If you have already chosen on-site, our guide to the on-site AI training programme covers the room and the logistics. For group sizes and Nairobi traffic, see how Nairobi companies can train staff on AI. Individuals looking for a course can look at online AI courses in Kenya.

What does "hybrid" actually mean?

Hybrid can mean two very different things, and mixing them up causes bad decisions. Ask anyone using the word which one they mean.

In this guide, blended is the arrangement we often recommend. The hybrid room is the one we usually advise against.

What does the research say?

The research suggests that format matters less than what happens in the session. Practice, feedback and enough time count for more. None of the studies below looked at AI training, and most are old, so treat them as a guide.

Web-based and classroom training

A meta-analysis by Sitzmann and others, published in Personnel Psychology in 2006, combined 96 research reports covering 19,331 people on 168 courses. It found:

• Web-based training was slightly better for teaching facts (6% more effective on average).
• Web-based and classroom training were about equal for teaching how to do things.
• When the same teaching methods were used in both, the difference in teaching facts almost disappeared.
• In courses shorter than 80 days, classroom training came out ahead for teaching facts (12 comparisons, 771 people). The authors did not have enough data to test this for practical skills. Most workplace AI programmes would count as short.
• Online learning added to face-to-face teaching did better than classroom teaching alone: 13% more effective for facts and 20% more for skills. Learners, however, were 6% more satisfied with the classroom alone.

Keep the limits in mind. Most of the learners were students: 67% of the reports involved undergraduates and only 15% of the courses involved employees. In the 11 comparisons where people were assigned at random, classroom training did better at teaching facts. Web technology has changed a great deal since 2006.

A US government review

A review for the US Department of Education in 2010, by Means and others, looked at 50 effects from studies between 1996 and 2008. Students online "performed modestly better" than those face-to-face, and blended learning had a larger advantage.

The authors warned that blended groups often got extra learning time and extra elements. So they said the benefit should not be credited to the technology itself. This is a useful warning for anyone reading a sales brochure, including ours.

The hybrid room

A 2020 systematic review by Raes and others, with a focus on higher education, looked at 47 studies of teaching students in a room and students online at the same moment. It reported some cautious optimism, along with real problems:

• Remote learners can feel left out, especially when technology fails.
• People in the room can feel neglected when the teacher is fixing technical problems.
• Remote learners tend to be more passive, and it is harder to get them involved.
• Sound quality is described as the biggest challenge.
• Teachers often report heavy fatigue and more work.

The authors also say most of this research is exploratory, and studies of learning outcomes have only begun to appear.

Video fatigue

Stanford's Jeremy Bailenson published four arguments in 2021 about why long video calls may tire people. The arguments draw on existing research and had not been directly tested on video calls at the time. Our view is that shorter online sessions are a sensible precaution.

Which format should you choose?

Match the format to your people and their work. Use this table first, then check the internet and power sections below.

Why we usually suggest starting in person

In a first session, a trainer can see who is stuck. Many people new to AI will not ask for help on a video call. Their camera is off and you never find out.

The first session is also when the small practical problems appear. Work logins fail, the code for sign-in goes to a phone that is in a bag, and the network blocks a site. These are quick to sort out in a room and slow to sort out online.

This is our judgement from teaching. It fits the finding above that classroom training did better at teaching facts in short courses, and we should say clearly that we have no study on AI courses to prove it. Teams that already work well online can start online without much loss.

Follow-ups suit online well. People have tried the skill, they know their questions, and a 60 to 90 minute call does not cost a morning in traffic. Follow-up after the first session also helps skills last, which we explain in why corporate AI training does not stick.

What do Kenyan internet and power mean for your choice?

Kenya has many more mobile data connections than fixed ones, and power cuts remain common. So plan for staff joining from phones or hotspots, and for interruptions.

Subscriptions are not people, because one person can hold several SIM cards. Still, the gap tells you something important. Do not assume that staff working from home have fibre.

How much internet does a session use?

Video training can use a lot of data, and the needs vary widely. Microsoft publishes bandwidth figures for each person in a Teams meeting. Below is our arithmetic from those figures.

Teams adjusts quality to the connection, so real use will vary. Using the recommended screen-sharing rate, a 90-minute session would use about 1.7 GB for each person. The AI tools people practise on will add to that. Check the equivalent for Google Meet or Zoom if you use those.

Two things follow from this table.

Sharing one office connection can fail. If 20 people each join on their own laptop from one office, you may need 50 Mbps or more for recommended quality. Ask ICT what is actually available at that time of day.

Data costs real money at home. If staff join on their own bundles, the company is passing a training cost to them. Decide who pays before you announce the session. A small data allowance avoids resentment.

A connection test to run one week before

1. Ask each person where they will join from, and on what device.
2. Have them join a test call at the same time of day as the training.
3. Have them open each AI tool you plan to use.
4. Check that the office firewall does not block those sites.
5. Confirm that the sign-in code reaches their phone.
6. Name a phone hotspot or a second connection as backup.
7. Ask about power at each location, and have laptops charged before the start.
8. Give people the materials in advance, so they can continue during an outage.
9. Agree what will happen if the trainer's connection drops.
10. Keep a recording, without real data on screen, for anyone who misses part of the session.

How do you run live online AI training well?

Keep sessions short, and make people do the work. A long talk on a video call is the weakest way to teach a practical skill.

We suggest these rules. They come from our own practice and have not been tested in a study:

• 90 minutes at most, with a break if it runs long.
• A task every 10 to 15 minutes. People open the tool and do something.
• Groups of about 20 or fewer, as with on-site training.
• Two people running it. One teaches and one watches the chat, helps with problems and handles technical issues.
• Breakout rooms of three or four, organised by the kind of work people do.
• Laptops required for practice. Phones work for watching. They are poor for spreadsheets.
• Show of work. Ask people to paste their results in the chat or a shared document, so you can see who is stuck.
• A clear rule for managers. The time is protected, and no calls or messages during the session.
• Cameras on when the connection allows. Do not make it a rule for people on bundles or weak signals.

Self-paced videos can help with the basics, but on their own they rarely change how people work. Use them before a live session or between sessions.

Three sample programmes

Adjust these to your team. They are examples of how a programme could look.

Blended, for staff new to AI

Fully online, for branches or regional staff

Fully on-site, for a single site with weak connections at home

What if you have to run a hybrid room?

Sometimes you cannot avoid it: a few people are at another branch and a date is fixed. If so, set it up with care, or the people online will drop away.

• Give it a technical operator. The Raes review points to a technology operator, present at every session, who sorts out problems in the room and online.
• Keep the in-room group small. One study in the Raes review says experienced instructors can manage without technical support when eight or fewer people are on-site. Larger groups need more.
• Spend money on sound first. Test the microphone from the far end before anyone arrives.
• Have remote people join early to test sound and access.
• Pair each remote person with someone in the room for exercises.
• Ask the remote group first when you ask a question.

A better option is usually to run a separate online session for the people who cannot attend.

Cost and data: what else to think about

Costs differ more than people expect

The trainer's fee is one line. We do not publish prices here, because they depend on scope. Compare the formats on these lines:

See what corporate AI training costs in Kenya and how to budget for AI training.

NITA and online delivery

NITA's Guidelines for Training and Reimbursement date from 2019, and they do not say whether online delivery qualifies. If levy reimbursement matters, get NITA's position in writing before you choose online. The same guidelines set in-house courses at 8 to 25 participants per session unless an exemption is given, with no limit for in-house team-building and sensitisation courses. FuKazee does not yet hold NITA accreditation, so please take that into account in your planning.

Recordings and screen sharing

An online session creates records that a workshop does not. Use masked files, tell people if you are recording, and decide who can see the recording and when it will be deleted. Our guide to the Kenya Data Protection Act and AI explains the duties that apply to personal data.

Three prompts to help you choose and plan

These follow the four Cs we teach: Clarity, Context, Constraints and Customisation. Remove names before pasting anything.

1. Sort your survey results

Below are anonymous answers from staff about where they would join training, what device they have and how good their internet is. Sort them by department. Context: we are choosing between on-site, live online and blended AI training. Constraints: show a table with department, number of people, laptop or phone, and internet rated good, patchy or poor. Do not guess any missing answer. Mark those as [UNKNOWN]. Customisation: finish with three lines on which format suits each department. Answers: [paste with names removed]

2. Turn a day into online sessions

Help me convert a one-day AI workshop into live online sessions. Context: the audience is [team], new to AI, and some use mobile data. Constraints: sessions of no more than 90 minutes, a task every 15 minutes, and no activity needing more than a basic connection. Customisation: give a table with session, skill, task and what to do if someone's connection fails. Workshop outline: [paste]

3. Write the joining note

Write a short joining note for staff attending live online AI training on [date] at [time]. Context: some people have never used an AI tool and some will join on mobile data. Constraints: under 120 words, friendly, plain English. Tell them to use a laptop if possible, keep their phone nearby for sign-in codes, join 15 minutes early, and contact [role] if they have trouble. Customisation: it will be read on a phone.

What goes wrong, and how to avoid it

Questions people ask

Is online AI training as good as in person?

The older research found online and classroom training about equal when the teaching methods were the same, though classroom training did better at teaching facts in short courses. None of that research covered AI. In practice, online works best for people comfortable online with reliable connections.

What is the difference between blended and hybrid training?

In blended training, some sessions are in person and others are online. In a hybrid room, one session has some people present and others on a video link. We usually recommend blended and advise against the hybrid room.

How long should an online AI training session be?

We suggest 90 minutes at most, with a practical task every 10 to 15 minutes. Longer content works better split over several days. This is our practice and has not been proven in a study.

Can staff do AI training on their phones?

They can watch a session and try simple prompts on a phone. Practising with spreadsheets and long documents is much harder. If most of your staff only have phones, consider training them in person with shared laptops.

What internet speed is needed for online training?

Microsoft lists 250 kbps per person as the minimum for screen sharing in a Teams meeting and 2,500 kbps as recommended, both as of the page we checked in November 2026. Multiply by the number of people on the same connection. Test it at the actual time of day.

Should we record online AI training?

Recording helps people who miss a session or lose their connection. Use masked examples, tell participants, and decide who can see the recording and how long you will keep it.

Is online AI training covered by NITA reimbursement?

NITA's 2019 guidelines do not mention online delivery. Ask NITA in writing before you assume a virtual course can be claimed. Registration and approval rules also apply.

Sources

• Sitzmann, Kraiger, Stewart and Wisher, "The comparative effectiveness of web-based and classroom instruction: a meta-analysis" (manuscript of the 2006 Personnel Psychology paper), last checked 10 November 2026
• Means and others, Evaluation of Evidence-Based Practices in Online Learning, US Department of Education, 2010 (full report on SRI's website), last checked 10 November 2026
• Raes, Detienne, Windey and Depaepe, "A systematic literature review on synchronous hybrid learning: gaps identified", Learning Environments Research, 2020, last checked 10 November 2026
• Bailenson, "Nonverbal Overload: A Theoretical Argument for the Causes of Zoom Fatigue", Technology, Mind, and Behavior, 2021, last checked 10 November 2026
• Communications Authority of Kenya, Third Quarter Sector Statistics Report, 2025/2026 (January to March 2026), last checked 10 November 2026
• TechTrendsKE, "Kenya's broadband market grows as fibre, wireless and satellite add more connections" (19 September 2026), last checked 10 November 2026
• EPRA, Energy and Petroleum Statistics Report for the year ended 30 June 2026, last checked 10 November 2026
• Microsoft, "Prepare your organization's network for Teams", last checked 10 November 2026
• NITA, Guidelines for Training and Reimbursement, NITA/IBTA/G/03, February 2019, last checked 10 November 2026

Where FuKazee fits in

FuKazee delivers AI training on-site in Nairobi, by arrangement elsewhere in Kenya, and live online. We will help you choose honestly, and if a shorter or online-only programme would do the job, we will say so.

We assume nobody knows AI, we avoid jargon, and we say plainly that nobody becomes an expert in a few days. We show the free version of every feature, we never promote one tool, and Microsoft 365 and Copilot are part of what we teach. Since 2022 we have trained more than 3,000 people.

See AI training in Nairobi or our wider page on AI training in Kenya. If you are still deciding who should deliver the training, see in-house versus external AI training.

To talk through the right format for your team, email us at lead@fukazee.com