AI and the year-end close: what it can actually do in December

December in a Swiss SME means the same list every year: chase open invoices, count the stock, tidy the accruals, write the staff letter. It is worth asking plainly where a language model helps with that list and where it does not.

Where it genuinely helps

Correspondence at volume. Payment reminders, year-end letters to clients, the note to suppliers about the holiday closure. These are formulaic, repetitive, and need to sound polite in three languages. A model drafts them in seconds and a person approves them in a minute.

Reading things you did not write. A twelve-page lease, a new supplier’s terms, a client’s revised framework agreement. Asking a model to list the payment terms, notice periods and liability caps gives you a map before you read. It is a reading aid, not a legal opinion.

Explaining your own numbers back to you. Paste an anonymised summary and ask which figures moved most against last year and what questions an auditor might ask. It will not find errors, but it is a good sparring partner for the management letter.

Where it will cost you

Anything that must reconcile. Language models are not calculators. They produce plausible arithmetic, which is the most dangerous kind. Do not let one add up a column that has to match the ledger.

The stock count. Valuation depends on your method, your cut-off and your judgement about obsolete items. A model has none of that context and will confidently supply the generic answer.

Anything with a person’s name in it. Performance letters, bonus decisions, termination drafts. Sensitive data, and the December rush is exactly when a shortcut gets taken.

The one habit worth adopting this month

Separate drafting from deciding. A model may write the first version of anything. It may not be the last reader of anything that leaves the building or enters the accounts.

Keep that line and December gets meaningfully lighter. Blur it and you will spend January finding out where.