# Why AI training doesn't stick in large organisations > Corporate AI training often fails after the session. Learn five reasons why it does not stick in large organisations and how to measure real progress. - URL: https://hyperfocusam.com/blog/why-ai-training-doesnt-stick/ - Language: en - Author: Sam Wong (黃力桐), https://hyperfocusam.com/ - 中文版: https://hyperfocusam.com/blog/why-ai-training-doesnt-stick-tc/ (Markdown: https://hyperfocusam.com/blog/why-ai-training-doesnt-stick-tc/index.md) - Published: 2026-03-15; revised 2026-10-16 --- With 11,000+ people trained, I have learned an uncomfortable thing about corporate AI training: the session is rarely where it fails. It fails in the weeks after, when nobody has said what is allowed, nobody owns the follow-up, and people go back to a desk that looks exactly the same as before. A large organisation makes this worse, not better. More people, more departments, more rules, and a training calendar that books one big session and ticks a box. This is how I see it after work with 70+ organisations, and what I now measure instead of the satisfaction score. ## The score is not the outcome When I ran [five online sessions for 1,500+ participants at a major Hong Kong bank](https://hyperfocusam.com/en/cases/hong-kong-bank-ai-training/), the programme scored 9.2 out of 10 in the bank's own participant surveys. I am proud of that number. But a satisfaction score tells you people liked the session. It does not tell you whether anyone opened an AI tool at their desk the next week. The outcomes I look for now are about behaviour: - **Do people come back with questions?** After the [AI for Surgeons course](https://hyperfocusam.com/en/cases/ai-for-surgeons/) at the College of Surgeons of Hong Kong, 9 follow-up questions and suggestions came in through a form, and three Fellows emailed questions directly. Questions mean people are trying. - **Does anyone volunteer to lead?** At [Garden](https://hyperfocusam.com/en/cases/garden-ai-pioneer-programme/), seven of the 19 colleagues in the first cohort volunteered to champion AI in their own departments. - **Does the organisation ask for the next round?** After that first cohort, Garden commissioned three further rounds of training in 2026, then asked for a proposal for the next phase. - **Is there a place to keep asking?** At [HKCT](https://hyperfocusam.com/en/cases/hkct/), the WhatsApp AI question group has more than 200 members. None of those is a score. All of them show whether the training is still alive. ## Learning AI is not a one-off event On 5 October I ran an online Q&A clinic for the surgeons, about six weeks after their two evenings of training. I opened it like this (translated from Cantonese): > "Learning AI, or applying AI, is not a one-off thing. It's not that after one workshop everyone becomes an AI master, or uses AI really well. It's a long-term journey." > — Sam Wong, CSHK Q&A Clinic, 5 October 2026 (01:00–01:21) The programme had three parts: a demonstration evening, a hands-on evening, and the follow-up clinic. The clinic was there because of what I said next: > "The idea we designed around: in the few days, or the week or two, after a workshop, people should be very motivated to try things and learn a bit more. But after a week, two weeks, a month, two months, what then?" > — Sam Wong, CSHK Q&A Clinic, 5 October 2026 (01:21–01:36) That drop is normal. Large organisations just have more ways to let it happen. ## Five reasons it doesn't stick ### 1. Nobody has said what is allowed In big organisations, the first question is not "how do I write a prompt?" It is "am I allowed to use this?" If the answer is unclear, careful people do nothing, which is the rational choice. At Garden, the IT directive allowed only Microsoft Copilot on the web and Perplexity, and the most confidential tier of data was barred from AI tools altogether. At the bank, the AI committee required exercises to use external public data only. Both programmes worked because the training started inside those limits, with a red, amber and green data grading, instead of ignoring them. If your staff do not know which tools they may use, sort that out first. I wrote about [which AI tools a Hong Kong team can actually use at work](https://hyperfocusam.com/blog/which-ai-tools-work-in-hong-kong/). ### 2. The training uses someone else's work A demo of what AI can do is entertaining. It changes nothing on Monday, because nobody's real task was in the room. Training sticks when people practise on their own material: the report they write every week, the email they answer every day. At Garden, each department formed its own cohort, and the exercises followed that department's real working situations, with fictional data. I explained why I keep rooms small enough for this in [why my best AI workshops have thirteen people](https://hyperfocusam.com/blog/why-my-best-ai-workshops-have-thirteen-people/). ### 3. Everyone is trained at once A large organisation likes one big session: it is efficient to book and easy to report. It is good for awareness and a shared vocabulary. It is weak for behaviour change, because nobody can stop and fix their own problem. I run both, and I wrote about [what a 500-person AI webinar can and can't do](https://hyperfocusam.com/blog/what-a-500-person-ai-webinar-can-and-cant-do/). Garden did not start company-wide. It started with HR, then extended department by department. ### 4. Nobody owns the follow-up The workshop ends, the trainer leaves, and no one is responsible for what happens next. The places where I have seen AI use last all have a follow-up that someone owns: a clinic six weeks later, a monthly workshop, a question group where people can ask in the moment they get stuck. HKCT built exactly that: a WhatsApp question group plus a monthly workshop, because the college felt AI changes too quickly to fix a curriculum in advance. ### 5. People get more information, not more relevance After a course, people are flooded with tools, news and buzzwords, and many feel they cannot keep up. Before the clinic we collected the surgeons' questions and grouped them into six concerns: tools, availability, abilities, speed, security and people. Most of them were not about technique. They were about what applies to me. My advice at the end of the clinic was: > "Make every piece of information relevant to you. That is the right way to learn AI, not forcing a lot of information, a lot of news, or tools you don't use onto yourself." > — Sam Wong, CSHK Q&A Clinic, 5 October 2026 (51:00–51:15) ## What a structure that sticks looks like If you want AI use to last in a large organisation, the shape matters more than the content of any single session: - **Settle the rules first.** Which tools, which data, in writing, before training starts. - **Start with one group, not the whole company.** One department or one cohort of curious people, practising on their own work. - **Plan the follow-up before the first session.** A checkpoint a few weeks later, and a channel for questions in between. - **Let champions emerge.** The people who volunteer to help colleagues are your adoption plan. - **Measure behaviour.** Questions asked, champions, tasks people now do with AI, and whether the organisation asks for the next round. If you are choosing a provider, start with [the five questions to ask before you buy AI training](https://hyperfocusam.com/blog/five-questions-corporate-ai-training-hong-kong/). How I run corporate AI training in Hong Kong is on [this page](https://hyperfocusam.com/corporate-ai-training-hong-kong/). ## FAQ ### Why doesn't AI training stick in large organisations? It is usually not the content. Employees are often unsure of what they are allowed to use, exercises do not involve their own work, everyone is trained at once, and no one owns the follow-up after the session. ### How to measure the success of AI training? Look at behaviour instead of satisfaction scores. Check if people are asking follow-up questions, if colleagues are volunteering to champion AI, if tasks are being done differently, and if the organisation asks for more training. ### Is a one-off AI workshop worth it? Yes, for awareness and a shared starting point. For lasting change, plan the follow-up before booking: include a checkpoint a few weeks later and a place for questions in between. ### Should we train the whole company at once? Start with one department or one cohort on their own work. Let those who volunteer to help others become champions, then expand the training.