Financial reporting: what software and AI can take over, what needs a person, and what the data must look like first

Software can already build unit statements from a closed ledger, run consolidation mechanics and check that every reported figure ties to an account balance. People still own estimates, disclosure judgment, board narrative and sign-off on filings. Neither works until account mappings, intercompany tags and entity structures are clean and owned.

Where software already carries the load

Business unit statements are the easiest win. Once each ledger account maps to a reporting line, the balance sheet and income statement for a unit come straight out of the closed books. No one should be retyping them.

Consolidation mechanics come next. Systems translate foreign currency balances, match intercompany pairs, post eliminations from defined rules and calculate minority interests. When the rules are written down and the counterparty data is there, this step runs the same way every period.

Traceability checks are where automation pays for itself quietly. A tool can tie each statement line, footnote figure and regulatory schedule back to the general ledger and flag anything that does not agree. This used to eat days of reviewer time spent ticking printouts.

Recurring management and cost reports also belong here. Refreshing them, slicing costs by cost center or project, and sending them to the right owners is plumbing.

AI adds something different. It can draft first-pass variance commentary from ledger detail and last period's explanations. It can read a footnote draft against the prior year and spot a number that was never updated. It can reconcile the fixed asset register to the ledger and list the differences worth chasing. Every one of these outputs is a draft. Treat it as a junior analyst's work that needs review.

Where a person has to stay in charge

Estimates drive the statements, and they are judgment calls. The allowance for doubtful receivables, an impairment test and an accrual for a liability whose amount is still uncertain all depend on knowing the business. A model can suggest a figure. Someone accountable has to choose it and defend it to auditors.

Intercompany differences that survive matching need a decision. Is this a timing gap, a pricing dispute or a booking error? The answer often sits in an email between two controllers.

Legal and management consolidation look similar but serve different masters. The legal view follows ownership. The management view follows how leaders run the business, and that shifts with reorganizations. Deciding the management structure and explaining why it differs from the legal one is human work.

Statements for board review involve choices about emphasis. How to describe a weak quarter, which risks to raise and what to leave for questions are not data problems.

Quarterly and annual filings, along with shareholder reports, carry personal certification and legal exposure. Disclosure wording, materiality calls and final sign-off stay with named individuals. Regulatory reports raise a similar issue whenever the rules are ambiguous: someone has to interpret them and stand behind the interpretation.

Manual journal entries need a human approver who understands why each one exists.

What the data has to look like first

Automating a messy close just produces wrong statements faster. Several conditions have to hold before any of the above is safe.

The mapping from ledger accounts to statement lines must exist once, in the system, with a named owner and a change history. If each report has its own mapping in its own spreadsheet, fix that before anything else.

Intercompany transactions need the counterparty recorded at the moment of entry. Adding it later, at month end, is where elimination failures begin.

Subledgers for assets, receivables and payroll should be reconciled to the ledger before the reporting cycle starts. If those reconciliations happen during reporting, every downstream number is provisional.

Both entity hierarchies, legal and management, have to live in the consolidation tool. A management structure that exists only in someone's workbook cannot be automated, audited or inherited.

Manual journals should carry their support and approval inside the system. Cost center and project attributes on transactions must be complete enough that cost reports need no cleanup. When the chart of accounts or the entity structure changes, prior periods need restating on the same basis. Otherwise every comparison breaks.

Questions to ask the people who run it

The documented process and the real one rarely match. These questions tend to surface the gap.

• After the system produces the statements, which numbers get changed, and where are those changes recorded?
• Which spreadsheet would cause the most damage if it disappeared tomorrow?
• When two units disagree on an intercompany balance, who makes the call?
• What goes into the board pack that never passes through the ledger?
• Does any regulatory report draw from a different source than the published statements?
• Is anything copied by hand from one system into another?
• When did someone last reconcile the management consolidation to the legal one, and what did they find?
• What does the final reviewer actually check before signing, and what do they assume someone else checked?
• Which footnotes are rebuilt from scratch each period, and which are just rolled forward?

APQC's Process Classification Framework® (PCF) is an open standard developed by APQC, a nonprofit that promotes benchmarking and best practices worldwide. To download the full PCF or to view definitions and measures, please visit www.apqc.org/pcf.