White paper · Accounting & general ledger

Accounting automation with AI: building a close you can explain

A practical guide to source evidence, recurring work, period judgments, late changes and the controls behind a supported accounting close.

View Markdown

A supported close connects source records, accounting judgments, approved entries, reconciliations and the reports built from them. Automation can carry repeatable preparation. AI can help investigate inconsistent evidence, while responsible people retain accounting judgments and the authority to approve adjustments and close periods.

Begin with the explanation behind the numbers

A financial report is more useful when the team can explain the records and decisions behind it. That explanation becomes difficult when support is scattered across contracts, acceptance records, payroll information, supplier correspondence and spreadsheets. The challenge is not simply producing a report; it is maintaining a defensible path to it.

Start by identifying the decisions that must be made during your close. Some work follows established schedules and rules. Other work depends on evidence that someone must interpret. Give each item an owner, supporting records, a review state and a clear description of what would count as complete.

Proper’s Accounting films explore that distinction through Glen and Alto. They are workflow illustrations, with AI Wow explicitly a future concept. Creative approval of a film does not establish that every illustrated capability is available in the product. Demonstrate the intended configuration before relying on it.

Use recurring work to create a clear review queue

Recurring preparation can include scheduled entries, allocations, reversals and matching tasks where the organization has defined appropriate rules. The operational aim is to reduce repeated setup while retaining enough support to inspect the result. A recurring template still needs an owner when the underlying facts change.

Give unexplained differences a visible route to resolution. A bank or subledger balance that does not match should not disappear inside a completion percentage. Record the difference, the source versions compared, the responsible person and the next investigation step.

A useful close dashboard therefore distinguishes completed work from work awaiting evidence, review or a successful posting. These states help reviewers direct attention. They also prevent a generated file or prepared journal from being mistaken for an accepted accounting result.

Treat period allocation as an evidence question

Glen’s AI Now story concerns an invoice arriving after work has crossed accounting periods. The relevant task is to trace the priced phases, acceptance records and applicable accounting policy, then prepare supported entries for review. The invoice date is one fact in that investigation; it does not establish the entire treatment by itself.

For an illustrative evaluation case, provide a contract with separately priced phases, records showing different acceptance dates and the later invoice. Ask the demonstrator to show which evidence supports the proposed period allocation and which assumptions still need a person’s decision.

Keep the proposed entries, any related reversal and the later invoice connected. The accounting owner should be able to verify that the accepted treatment is recorded once through the intended sequence. This is a workflow test, not advice to apply a particular recognition policy to a real transaction.

Keep the judgment and the action distinct

An assistant may assemble records, propose an explanation and prepare entries. The appropriate reviewer still needs to decide whether the accounting treatment is supported. Define that decision separately from the authority to post, reopen a period or change a report.

The review should identify the exact evidence and proposed result. If new material information arrives, the process should show which decision may have changed and return it to the relevant person. A prior approval is a historical fact; it is not automatically a current approval of revised numbers.

This structure also makes disagreements easier to resolve. Record the unresolved question and who can answer it. Missing confirmation from a supplier, unclear contract language or conflicting acceptance evidence should remain visible until resolved or addressed through the organization’s accounting process.

Verify the path from an entry to a refreshed report

Completion requires evidence from the systems that hold the accounting record. Check the posting result, related reconciliation and report version. If a connector or import rejects a change, the report cannot be described as updated simply because someone requested the correction.

When a report is refreshed, preserve the link to the books and reviews it reflects. That lets the team answer a practical question: which report was supported at the time it was circulated, and what changed afterward? A new file name alone provides little assurance.

For a board pack, also identify the explanations or commentary affected by a change. A revised ledger amount may alter a margin discussion or another management narrative. The responsible reviewer should see that dependency rather than discover the inconsistency during the meeting.

Use late evidence to test the AI promise

Alto’s revised Accounting AI Wow film is an illustrative future concept involving several connected investigations. A customer email says only one phase was accepted while the accounts show revenue for three. Payroll costs were incurred; supplier work still needs confirmation. The concept follows those facts to affected judgments and reporting.

The story then introduces changes on other projects and a supplier record that contradicts earlier confirmation. The proposed assistant keeps the investigations moving, brings changed decisions back to the right people and verifies revised results after approval. The promise is sustained work across dependencies, not simply another alert.

Use that scenario to ask precise product questions. Which records can the available system read? How does it identify a changed dependency? Who receives an unresolved question? What evidence proves a posting and report refresh occurred? Record the answers separately from the future concept.

Evaluate a complete close case

Select an anonymized case representative of your business. Include routine work, a judgment requiring evidence, a rejected handoff and a late change after review. Ask every provider to use the same case and explain where the configured system needs human work.

Keep a written record of the edition, integrations, permissions and supported scope. The strongest outcome is a repeatable process your team can operate and explain, with clear limits and recoverable exceptions.

  • Trace the source records through the proposed treatment and approval.
  • Verify accepted postings and the associated reconciliation.
  • Confirm which report and review versions reflect the result.
  • Introduce new evidence and inspect the affected review path.

Questions & answers

Can AI choose the accounting treatment?

It can assist with assembling and explaining evidence. The responsible accounting professional must retain judgments requiring expertise, policy interpretation or resolution of uncertainty.

Is a generated board pack proof that the close is complete?

No. Check that the pack reflects the accepted books, completed reconciliations and required reviews. Unresolved items should remain explicit.

Does the AI Wow film establish current functionality?

No. It is a future concept and an evaluation scenario. Confirm any available functionality through a demonstration of the exact product configuration.

Start with your workflow.

Bring your records, exceptions and the decisions that matter to your team.

Let’s talk Proper