Client Story

Month-end close and accruals: from email chasing to governed evidence

A multinational finance function closes its books across multiple legal entities and currencies each month. The client is not named here at their request. This case study documents the close and accruals workflow; the forecast side of the same programme is documented separately.

Octopus AI was deployed as the Artificial Financial Intelligence (AFI) layer on top of the existing ERP and close calendar. Nothing was replaced. The ERP remained the system of record throughout.

The challenge

Accruals depended on memory. Each month the team rebuilt a list of commitments from open purchase orders, contracts and recollection of prior periods, then chased owners by email for confirmation of what had actually been delivered.

Two problems followed. Late confirmations pushed judgment calls to the final hours of the close, and the reasoning behind each accrual lived in inboxes rather than beside the entry.

Auditors then asked the question the process could not answer cheaply: on what basis was this amount accrued, who confirmed it, and where is that confirmation now.

What was deployed

A close and accruals AFI capability was deployed against the existing ERP with scoped read-only access and role-based access control. It runs in four stages:

Stage 1
Accrual identification
Open purchase orders, goods receipts, contract terms and prior-period patterns are read from the ERP and compared against what has already been posted. Candidates are ranked by materiality, not by whoever remembered them.
Stage 2
Owner confirmation
The accountable owner receives a request naming the specific commitment, the period and the amount derived from source. Every outbound message is approval-gated by a named human before it sends, and every reply is logged as evidence.
Stage 3
Entry preparation
The journal entry is assembled with account coding from the customer's own chart of accounts, the calculation applied, and the supporting records attached. Nothing posts until an accountant approves it.
Stage 4
Close-cycle recall
Each cycle's confirmations, judgments and reversals are retained. Next period starts from what was decided last period instead of reconstructing it from inboxes.

What changed

Accrual candidates are derived from source records each period instead of reassembled from memory.
Confirmation requests go to the accountable owner with the commitment, period and derived amount stated, and every send is approved by a named human first.
Each prepared entry carries its calculation, its source records and the owner's confirmation as attached evidence.
Judgment stays with the accountant: entries are prepared for approval, never posted autonomously.
Prior-period decisions are retained, so the next close starts from documented reasoning.

A note on numbers

This page deliberately publishes no close-cycle time or cost figures. The only measured results Octopus AI publishes come from the forecast deployment documented in the enterprise forecast case study. Close and accruals outcomes are described here as process and control changes until measured figures are agreed for publication with the client.

Questions finance leaders ask about this deployment

Which finance function does this case study cover?

Month-end close and accruals: identifying accrual candidates from open purchase orders, receipts and prior-period patterns, requesting confirmation from the accountable owner, and preparing the journal entry with supporting evidence attached.

Does the AI post journal entries by itself?

No. Journal entries are prepared, not posted. A named accountant reviews and approves each entry, and the approval is recorded with who approved it, what changed and when.

How are accrual amounts calculated?

As deterministic code against reconciled source data — open purchase orders, goods receipts, contract terms and prior-period actuals. The language model drafts the description and conducts the confirmation dialogue with the owner; it does not produce the amount.

What evidence does an auditor get?

Each prepared accrual carries the source records it was derived from, the calculation applied, the confirmation exchange with the accountable owner, and the named approver with a timestamp. Every figure traces back to the general ledger line it came from.

Did this replace the ERP or close-management system?

No. The ERP remains the system of record and the close calendar remains where it was. Octopus AI runs as a layer on top, reading source data and returning prepared entries and logged commentary.

Are the close-cycle time savings published?

Not yet. Octopus AI publishes measured figures only for the forecast deployment documented in the enterprise forecast case study. The close and accruals results are described here as process and control changes rather than as numbers.