Lecture or hands-on AI workshop: which one to book, and when

Someone in management wants the company to "do something about AI" before the end of the quarter. One person suggests a speaker for the next all-hands meeting, another wants a hands-on day for the team that writes the proposals. Both take people away from work, and the two get compared as if they were the same thing at different lengths. They are not.

The short answer

Book a lecture when the goal is a shared picture: a whole room, a mixed level of experience, one session that sets the same starting point and the same ground rules for everyone. Book a hands-on workshop when the goal is a skill: a small group, real tasks from their own week, and people leaving with something that already works. A lecture changes what people think is possible; a workshop changes what they do on Monday. Many companies get the most from both, in that order: a lecture for everyone, then a workshop for the team with the clearest use, with a follow-up after real use.

What decides it

• The outcome you need. Alignment, curiosity and common rules point to a lecture. A specific task done differently next week points to a workshop. The Kirkpatrick model, a standard way to evaluate training, says to start with Level 4, the results you want, and work back from there.
• Group size. A lecture works for a room of any size. A workshop needs each person at a keyboard with their own task, and a trainer who can reach every screen, so it suits a small group. A large team becomes several sessions.
• How mixed the room is. When some use AI daily and others never tried it, a lecture brings them to one starting point. A workshop works best with people who share a role and repeat similar tasks.
• Preparation and setup. A lecture needs a short call about the audience. A workshop needs work accounts ready, the right connections, a rule on which data may go where, and real, anonymized examples before anyone sits down.
• Time away from work. A lecture costs everyone an hour. A workshop costs fewer people more hours.

Research on teaching points the same way. A PNAS commentary on a meta-analysis of 225 studies reports that failure rates fell from 34% with traditional lecturing to 22% with active learning. Those were university science courses, not workplaces, so read it as a direction, not a forecast. In BCG's AI at Work 2025, a survey of more than 10,600 workers in 11 countries, only 36% of employees felt adequately trained, and those who received five or more hours of training, especially in person and with coaching, were significantly more likely to become regular users.

A worked example

Example, for illustration. A 30-person services company compares three plans. The internal cost is an assumption: 150 NIS an hour of staff time, before any tool seats.

• Plan A, a lecture only. One hour for all 30 people is 30 hours of team time.
• Plan B, workshops for everyone. Three groups of 10, three hours each, is 90 hours of team time, and three rounds of preparation.
• Plan C, both. The one hour for all 30 (30 hours), then three hours for the 8 people who write proposals (24 hours) and a one-hour follow-up for those 8 two weeks later (8 hours). That is 62 hours.

At the assumed rate, the three plans come to 4,500, 13,500 and 9,300 NIS of staff time. Plan C costs about twice Plan A in time and about two thirds of Plan B. Its value depends on one assumption you can check: if the 8 proposal writers each save an assumed 1 hour a week, that is 8 hours a week, and the 32 workshop and follow-up hours are covered in about four weeks. If nobody uses the tools a month later, no plan paid off.

How to decide, step by step

1. Write the result in one sentence. "Everyone knows the rules and what is possible" is a lecture. "The proposal team drafts first versions with AI" is a workshop.
2. Count the people and their roles. A large mixed room points to a lecture. One team that repeats the same tasks points to a workshop.
3. Check whether the tools are ready. No work accounts or data rules yet means a workshop will spend its first hour on logins. Fix that first, or start with a lecture.
4. Ask the speaker or trainer four questions: how they adapt the content to your field, what participants leave with, what happens after the session, and how they handle what may go into which tool.
5. Pick the measure before the date. For a lecture, how many people try one first step within two weeks. For a workshop, who still uses the tools on real work after 30 days.
6. Put the follow-up in the calendar with the session itself.

Mistakes that cost the most

• Booking a lecture to build a skill. People leave inspired and go back to the old routine, because nobody practiced on their own work.
• Running a workshop for 40 people. Nobody gets help at their own screen, and it becomes a long lecture with laptops open.
• A workshop on generic demos. If the examples are not the team's real tasks, the session proves the tool works and changes nothing.
• Judging either by the feedback form. The Kirkpatrick model calls that Level 1, reaction. What matters is Level 3, behavior: who works differently a month later.

When outside help is worth it, and when it is not

If someone on the team already works well with AI and can present to colleagues, an internal session can open the subject at no outside cost, and free courses such as the AI Fluency course in Claude Academy (14 lessons, about 4 hours) can build a common base. Outside help is worth it when the room is large or mixed and needs one clear picture, when nobody has time to prepare real examples, when data rules are unclear, or when an earlier attempt faded.

The lectures page covers talks for companies, teams and events, and the AI training page covers hands-on workshops on the team's own tasks with follow-up sessions.