A Year of AI Training at a Hong Kong Food Company: Six Departments, Twenty Sessions
In July 2025, a Hong Kong food manufacturer -- the Hong Kong food company behind the biscuit tins most of us grew up with -- booked a single AI workshop. Standard format. The kind of session that ends when the projector powers down.
Then they came back.
HR -- January to February 2026
Six sessions, nineteen people. HR tends to go first -- they see the people-side implications earliest, and they usually have the professional-development budget. We built workflows around actual HR processes: recruitment screening, onboarding documentation, internal communications. No grand AI transformation narrative. Just: here is something you currently do by hand, here is how a tool handles most of it, here is where you still need a person making the call.
Sales and Marketing -- March
Three sessions, about thirty participants. Different energy entirely. Sales wanted results they could measure by end of quarter. Marketing wanted creative output they could use that week. The sessions shifted from "how does this tool work" to "can this tool do the thing I need done by Thursday."
Manufacturing and Operations -- April to June
A top-management session had already happened in March. By May, Finance, Accounting, Purchasing, R&D, and Operations had all come through the room. Manufacturing was the most skeptical department -- not because they didn't see the potential, but because their work involves physical systems and production lines, and "AI" sounds abstract when your problems are on a factory floor. Winning over the operations people took longer and required completely different material. The deck that worked for marketing was useless here. Every session needed rebuilding around their actual day-to-day -- purchase order processing, supplier communication, quality-control documentation.
Six departments, roughly twenty sessions, over a hundred staff trained across a year.
What the year taught me
Not every session landed. One early session had too much concept and not enough hands-on time -- the ratio was wrong, and the feedback said so plainly. The next batch got a different session.
When the company ran a formal evaluation in May -- twenty-nine participants surveyed -- the score came back at 4.48 out of 5, eighty-six percent rating 4 or 5 stars. That number is meaningful, but not because it's high. It's meaningful because it came from people who'd been working with AI tools for months -- not riding the excitement of a first demo. These were repeat participants who'd had time to hit the limitations, get frustrated, discover the gap between what the tool promises and what it delivers on a Tuesday afternoon. A 4.48 after months of use is a fundamentally different data point from a 4.48 after an introductory session.
The most unexpected finding was how the later batches benefited from the earlier ones. By the time the finance team sat down, HR staff were already using tools in their daily work. New participants walked in without the usual arms-crossed, show-me posture. They'd already heard about it -- not from a company memo, but from the person next to them at lunch saying "I used it to draft that report last week." Internal word of mouth turned out to be the most powerful adoption driver in the whole engagement, and it only works if you train enough people over enough time for the stories to spread on their own.
Most of my work is one session, one company, one day. A good single workshop has real value. But this engagement taught me that the kind of adoption that changes how a company actually operates requires repetition, patience, and a willingness to rebuild the material from scratch every time a new department walks in with a different set of problems.
One workshop is awareness. Twenty sessions is a workflow.
