The close is repetitive, deadline-driven and rule-heavy, which makes it a natural place for AI. It is also where errors become misstatements, so controls matter from day one.
Where AI helps in the close
- Reconciliations: matching transactions across bank, sub-ledger and ledger; suggesting reconciling items; flagging stale or unusual items.
- Journal-entry review: scoring entries for anomalies (unusual accounts, amounts, users, timing) so reviewers look at the riskiest first.
- Flux and variance commentary: drafting explanations of movements from the numbers and notes you provide.
- Close management: chasing open tasks, summarizing status, drafting the close memo.
- Reporting: first drafts of internal management packs and disclosure checklists.
What preparers told regulators
In 2024 the PCAOB staff talked to audit firms and companies about generative AI. Preparers said they used it to draft internal documents and to help with less complex, repetitive work such as preparing account reconciliations or identifying reconciling items. They also stressed that human supervision and review of AI output remain important, and flagged the "black box" nature of some tools.
Getting it right
- Start with matching and anomaly scoring: measurable, with clear exceptions for people.
- Keep AI inside existing controls: the reviewer of a reconciliation still signs off; AI changes how the work is prepared, not who is accountable.
- Evidence: keep what the AI proposed, what the reviewer changed and who approved.
- Measure: days to close, number of manual reconciling items, post-close adjustments.
- Talk to your auditor early: agree how they will test AI-assisted controls.