Choose an AI workshop for a finance team

An AI workshop for a finance team should leave people with a safer, faster way to do one recurring job. In Hong Kong, that might be drafting monthly variance commentary, preparing a board pack or checking the first pass of a credit memo. Ask to see the exercise, the data rules and the review step before you buy the session.

I spent 13 years in banking. The hard part of a finance workflow is rarely writing the first paragraph. It is knowing which figure came from which source, who may see the material and who signs off the conclusion. A workshop that skips those steps teaches a demo, not a working method.

What should an AI workshop for a finance team cover?

Start with one task that recurs. Take monthly variance commentary as an example. The analyst has numbers from a controlled spreadsheet, a prior-month narrative and a deadline. A useful workshop lets participants draft an explanation from a cleared sample, trace each claim to the source figures, challenge an unsupported sentence and record who approves the final version.

That is more instructive than a tour of ten tools. Ask the trainer how much time each person spends doing the task, rather than watching a screen. If your team cannot name a repeatable task yet, a short introduction may be the better first purchase. My comparison of AI courses and learning formats in Hong Kong explains where an open class fits.

A good finance workshop teaches the review habit as carefully as the prompt.

Which data will the team use?

Agree the boundary before anyone opens a tool. Can the session use an approved company environment? Will the examples be synthetic or properly cleared? Who confirms that a document may be uploaded? A trainer should be able to run the exercise on realistic material without asking staff to paste customer records into a public account.

The Privacy Commissioner's AI model framework recommends governance, risk assessment, human oversight and employee training when organisations use AI with personal data. It is guidance for organisational practice, not a blanket permission to enter personal data into a tool. For a practical account of the work-data decision, see our guide to using ChatGPT with work data.

Who checks the output?

Make the review visible in the exercise. For variance commentary, one person should compare the draft against the source table, check period and currency, inspect the reason given for each movement and mark any assertion that needs an owner. The reviewer then either approves the text or sends it back. A polished paragraph that reverses a sign is worse than a rough one.

Banks have a further governance context. On 15 October 2026, the HKMA announced its ART framework for responsible AI adoption, with accountability, resilience and trust at its centre. That does not mean every finance team needs a regulatory lecture. It means a bank workshop should fit the institution's approved tools, controls and owners. A corporate finance team outside a bank still needs its own data and sign-off rules.

What should be left behind?

Ask for an artefact the team can use next week: a tested prompt with input boundaries, a short checking checklist, a named owner and a decision about where the output belongs in the existing process. Before booking, ask the provider to show a sample of that deliverable. If the only takeaway is a slide deck, the team may be back where it started by Monday.

There is an honest limit. A single session will not fix poor source data, unclear approval rights or a tool your firm has not cleared. It can expose those gaps and help you choose a modest first workflow. For a different decision, building a prototype inside a bank, I have written about why a bank AI workshop can stall after the demo.

I run corporate AI training in Hong Kong around the work a team actually has to deliver. If you want to start smaller, the free prompt library offers examples to test on material you are allowed to use. The aim is a method your team can repeat, with a person still accountable for the result.