A commentary pack can leave the building with a fluent paragraph nobody checked. The model wrote it in a minute. The reviewer signed it because it sounded like last month. The error shows up when a number in the note does not match the ledger.
If you are buying a day for a finance team, ask which close task the room will run. A general tour of the tools will leave the pack alone. You want one recurring job, on data you are allowed to use, and a check a human still owns before anything goes to a client, an auditor or the board.
What the day is for
The OECD's chapter on AI in finance treats transparency and explainability as part of trustworthy use in the sector. People should be able to understand an outcome and, where it is appropriate, challenge it. The same chapter treats deployment as an organisational change, including reskilling and upskilling the practitioners who will run the system. I use that when a sponsor wants a login and a slide. A finance officer who cannot explain why a draft is usable has been handed a faster way to sound sure.
Our AI training workshop is built for that loop. It runs as a half day or a full day, in person or remote, for people who are not engineers. If someone can use email, they can do the work. For a finance team we map high value, low risk uses in their own jobs, practise on real tasks with dummy or approved data, and teach prompting patterns that produce a draft a reviewer can actually mark. People leave with guardrails for data and accuracy, and with a 90 day plan for one workflow.
Keep the room on one kind of work
A finance officer and a campaign manager do not share a task, and a mixed room slides back to generic prompts. Group people by the work they do. The close, the management pack, the variance note and the control that has to be evidenced each month are different jobs, and the examples have to come from the one you picked. What a single day should leave someone able to do is in what a corporate AI training day should produce. If you are buying across several functions, read what generative AI training for employees should include, and read role-based AI training before you put the whole firm in one room.
Bring a task that already happens. A first draft of the monthly commentary. A reconciliation narrative. A search across a policy pack you are allowed to use. Leave the live client file out of the tool. If the rules for what may be pasted are still being written, run the day on dummy numbers and name the date the rules will be clear.
The check that stays human
The output has to be something a reviewer can challenge. Where did this figure come from. What would make it wrong. Who signs before it moves. A draft that nobody can trace is a liability with better grammar. Put the check on the same page as the prompt, in words the team already uses in review.
Skip the tour of the tool market. Pick the few tools the company has approved and go deep on those. A debate about which model is best eats the hour the team needed for the close. Vendor neutral means we use what your people are allowed to use, and we say so at the start.
What to ask before you pay
- Which close task will the room run, on which data?
- What must a person still check before the output leaves the team?
- Which approved tools will we use, and what stays out of them?
- What will look different in next month's pack?
If the last answer is that people will feel more confident, you have bought a briefing. Ask it on the call. If you want the day, tell me the function, the tools they are allowed to use and the one task that eats the close. We will build the session around that task.
