Finance is one of the functions most exposed to AI and one of the least forgiving of its errors. A confident wrong number in a board pack is worse than no number, because it gets acted on.
The short version of what AI literacy means here: use these tools to draft, extract and explain, and never as the unchecked source of a figure. That means understanding why a language model miscounts, recomputing anything that matters, keeping customer data and price-sensitive information out of consumer tools, and being able to show an auditor what was checked and by whom.
Why finance is early in the queue
A great deal of finance work is structured, repetitive and text-heavy around numbers: reconciliations, variance commentary, invoice processing, management reporting. That profile is exactly what current tools accelerate, which is why adoption in finance has run ahead of policy in most organisations.
Where it genuinely helps
Drafting variance and management-reporting commentary from figures you supply. Extracting data from invoices, receipts and statements. Writing and debugging spreadsheet formulas. Building forecast scenarios and the narrative around a sensitivity. Summarising contracts for payment terms, penalties and renewal dates. Explaining an accounting treatment in plain language for somebody outside finance.
The common thread is that the numbers come from the ledger and the model works on the words.
Where it fails, and why
| Failure | Why it happens | The control |
|---|---|---|
| Wrong totals and aggregates | Language models predict text and do not calculate reliably | Compute in the spreadsheet or ledger, and tie out every stated figure |
| Invented line items | Extraction fills gaps with plausible values rather than failing | Sample extracted figures against source documents, weighted by materiality |
| Period, currency and unit errors | Context is lost across a long document | State periods and units explicitly, then check them in the output |
| Commentary that does not match the numbers | Narrative generated before the final figures landed | Reconcile every statement in the commentary to the underlying data |
| Confidential data exposure | Customer data or inside information pasted into a consumer tool | Approved enterprise tools only, and redact before prompting |
The second row is the one that catches experienced people. A model asked to extract sixteen invoice lines will return sixteen, because returning fourteen and an apology is not what it was built to do. The absence of an error message is not evidence that nothing was missing.
What it does to the control environment
When a machine performs a step, the control environment changes whether or not anybody updated the documentation. A finance lead should be able to answer five questions.
Where does AI touch a figure that ends up in reporting? Who reviews that step, and how is the review evidenced? Does segregation of duties still hold if AI prepared the work and a person approved it? Who signs when AI drafted the analysis, noting that the answer is always a person? How are AI-assisted workbooks version-controlled?
Auditor expectations in this area are still forming. Teams that can already produce recomputation evidence and name the reviewer will be in a considerably better position than teams assembling it retrospectively.
The skills that make a finance professional ready
Knowing which tasks to hand over and which never to. Prompting with explicit periods, units, materiality and output format. Recomputing and tying out every time it matters. Handling customer data and inside information correctly. Documenting what was checked and by whom. Escalating an anomaly rather than smoothing it over, which is the habit that separates a control from a courtesy.
Proving it
The AI Literacy Certificate is the free baseline. Thirty minutes, an identity check before the attempt, and a public register entry at the end. It covers the fundamentals, the risks and the governance awareness that any finance professional now needs, and it costs nothing at any step.
Applied competence is a different claim and needs a different assessment. The Council’s Finance and Accounting certifications examine the work itself rather than awareness of it, at Fundamentals and Professional grades, and carry post-nominals because they are examined credentials rather than awareness ones.
The order is worth keeping. The baseline is free and immediate; the certification is an examination with a fee and a date. Starting with the first costs nothing if you go on to the second, and tells you something useful if you do not.