Sam Wong.黃力桐Let’s talk training

TRAINING DECISIONS

AI training for finance and accounting teams in Hong Kong

Published:

Short answer: AI training for a finance team should practise the team's own month-end work in the tools the company already allows: summarising reports, drafting commentary, cleaning a spreadsheet before analysis, reading receipts and invoices into a fixed format. It should be strict about two things: which data may go into AI at all, and that every number is checked by a person before it leaves the team.

Sam Wong (黃力桐) is a corporate AI trainer based in Hong Kong. 11,000+ people trained, 70+ organisations, 13 countries.

What should AI training for a finance team cover?

Finance people have usually seen the feature tour already. What they lack is a reliable way to use AI on real files without guessing every prompt. In my sessions the practice covers:

What should a finance team never hand to AI?

Two things. First, confidential data that the company's policy keeps out of AI tools. Financial data is often in the highest tier, so the training must practise with fictional files. Second, the final say on a number. AI can misread a column or invent a total, so a person checks every figure before it goes into a report.

Why does the size of the file matter?

This surprises most finance teams. When I gave Copilot a full sales file with thousands of rows and dozens of columns, the answer came back vague or wrong. The same question on a clean extract, only the columns and the period that mattered, gave a clear answer. Most disappointment with AI in finance comes from feeding it too much, not too little. I wrote this up in plus, minus, multiply, divide: a cheat sheet for teaching Copilot.

What did this look like at Garden?

Garden's department-by-department programme gave finance its own cohort. The company allowed only Microsoft Copilot on the web and Perplexity, some colleagues had paid Copilot licences and some did not, and the most confidential tier of data was barred from AI tools. So every exercise used fictional data built to look like the team's real work, and the content included a Plus, Minus, Multiply, Divide framework for Copilot written for finance:

  1. Plus: give AI the file or context it needs before you ask.
  2. Minus: cut the file down to what the question needs.
  3. Multiply: save a proven set of steps as an agent, such as reading receipts into the same fields every time.
  4. Divide: set custom instructions once, so you stop repeating your preferences.

Across 2026 the programme ran 20 in-person sessions for 7 cohorts, and the company commissioned three further rounds of training after the first cohort.

How is in-house training different from an accounting AI course?

Public accounting and finance AI courses suit one person who wants a broad grounding, and some teach programming. In-house training is for the team as it works today: your approved tools, your licence mix, fictional versions of your own files, and your data policy practised in the room. Nobody needs to learn to code to get value from it.

What should a finance team prepare?

  1. The list of AI tools the company allows, and who has a paid licence.
  2. The data policy, including which tier finance data falls into.
  3. Three to five tasks the team repeats every month, so the trainer can build fictional files in the same shape.

For how I run training across a company, see corporate AI training in Hong Kong (中文).

FAQ

What should AI training for a finance team in Hong Kong cover?

AI training for a finance team in Hong Kong should focus on practising recurring tasks using the company's approved tools. This includes questioning P&L or sales files, drafting monthly commentary, extracting data from receipts and invoices into fixed formats, and condensing reports. Exercises should use fictional files that mimic the team's actual work to ensure practical application without risking data security.

Can finance staff input company financial data into AI tools?

Finance staff can only input company data if the specific data policy allows it for that tool. Because financial data is often classified as highly confidential, many Hong Kong companies prohibit its use in AI tools. Therefore, training should utilise fictional files. Additionally, staff must manually verify every number in an AI output before it is included in any official report.

Why does AI provide vague answers regarding finance spreadsheets?

AI often provides vague or inaccurate answers because the uploaded file is too large or messy. A full export containing thousands of rows, numerous columns, and hidden data can confuse the model. To ensure a clearer, more accurate response, users should trim the file to include only the specific columns and time periods required for the question. Sam Wong calls this the Minus step.

Do finance staff need to learn programming to use AI?

No, finance staff do not need to learn programming to use AI effectively. The primary value lies in using AI integrated within approved tools, such as Microsoft Copilot, to perform tasks like data analysis, drafting commentary, and document processing. While programming courses are useful for building custom models, they are not necessary for a team adopting AI tools for daily work.

What did the AI training for Garden's finance team include?

In 2026, Garden, a Hong Kong food manufacturer, conducted a department-by-department programme with Sam Wong. The finance team participated in a dedicated cohort, using only fictional data to comply with IT rules. They learned the Plus, Minus, Multiply, Divide framework for Microsoft Copilot: adding context, reducing file size, saving repetitive steps as an agent, and setting custom instructions.

Planning AI training for your finance team? Message me on WhatsApp and tell me a little about your team.

Sources