Month end close: which work disappears and which just moves

What automating a month end close actually removes, what it moves to whoever reviews the output, and why a close ends when its last input arrives rather than when the work is done.

By the Foxnut team · Updated 6 August 2026

How a month end close gets automated, and what that leaves behind

The UK government’s Whole of Government Accounts consolidate more than 10,000 public sector bodies. In July 2025 it published the accounts for the year ended 31 March 2024 - more than fifteen months late, with the Comptroller and Auditor General disclaiming the opinion rather than signing off. Of the 407 English local authorities in scope, 167 submitted no data at all. No software problem explains that, and no automation project fixes it either.

That is the whole argument of this page, demonstrated at the largest scale it has ever been shown: a month end close automates in four separable layers, and only two of them ever disappear. The first is collection: pulling bank statements, subledger balances, intercompany positions and payroll and tax files into one place on a schedule. The second is matching: comparing those balances against the ledger and surfacing what does not agree. The third is posting: writing the entries whose rule is deterministic, which is recurring accruals, allocations, depreciation, revaluation and the standing reclassifications. The fourth is coordination: the checklist, the dependency order, the status of each task and the sign-offs. Collection and matching genuinely go away. Posting goes away for the entries that have a rule and stays for the entries that need an estimate. Coordination does not go away at all; it becomes visible, which is a different and often larger benefit. What is left after all four is the reviewing, the exceptions, the judgements and the obligation to evidence every one of them, and that work does not shrink when the preparation around it does. It changes hands, from the person who built a number to a person now approving one they did not build. The question of what a closed month leaves the finance team to run is worth settling before the software question, because the answer decides which of the four layers is worth paying for.

The work that disappears, and the work that only moves

The distinction is not about difficulty. It is about whether the step has a rule that can be written down, or a judgement that has to be defended to somebody afterwards.

Close activity	What automation does to it	What is left for a person
Collecting statements, subledger balances and intercompany positions	Removes it, if the source systems have interfaces. This is the largest single block of hours in most closes and the least interesting	Chasing whatever has no interface, which is usually one entity, one bank or one spreadsheet
Matching transactions against the ledger	Removes the matching of items that agree, which is nearly all of them	Every item that does not agree, in a window that does not move
Recurring accruals, allocations, depreciation, revaluation	Removes the posting where the rule is written down, and makes the rule inspectable	Deciding when a rule has stopped describing the business
Accruals and provisions that need an estimate	Nothing. A number nobody can derive cannot be derived faster	The estimate, its basis and the evidence for both
Variance and flux analysis	Produces the first draft: what moved, by how much, against what	The explanation, which is the part anybody reads
The checklist, dependencies and sign-offs	Makes the critical path visible and stops the status chasing	Deciding what to do when a dependency is late, which is a management call
The audit trail	Changes its shape, from a person’s working paper to a system’s configuration and change log	Being able to show that the configuration was authorised and has not moved
Consolidation and elimination	Removes the arithmetic across entities	Agreeing the intercompany differences that the arithmetic exposes

Read down the right-hand column and the pattern is plain. Every residual is either an exception, a judgement or an act of evidencing. None of the three is a volume problem, which is why a close that has been automated well often costs the same number of finance days and produces a materially better result inside them.

What the decision turns on

Six structural dimensions decide whether a process is worth automating. The month end close reads unusually on three of them: its exceptions are concentrated in the inputs rather than in the transactions, its cost of error steps sharply at one moment rather than accruing, and its regulatory exposure is not occasional but named in the audit standard by process.

Month end close: process profile
Dimension	What it reads on the month end close	Source
Exception variance	The standard case is almost all of the volume and almost none of the elapsed time. Exceptions concentrate in the inputs, not in the entries: a subledger that closed late, an intercompany balance the two sides disagree on, an entity that sent nothing. The extreme public instance is a consolidation of over 10,000 public bodies in which 201 entities that should have been consolidated submitted no data at all, and 280 more were consolidated on unaudited figures. The arithmetic was never the problem.	National Audit Office, Whole of Government Accounts 2023-24
Volume	Not transaction count. The close is a fixed-cadence event rather than a queue, so effort scales with the number of ledgers, entities, reconciling accounts and hand-offs between people. The audit standard characterises the process the same way, requiring the auditor to understand the locations and management participants involved and the types of adjusting and consolidating entries, not the throughput. The payback condition follows: where one person can complete the whole close inside the days available without waiting on anyone else, coordination is the only layer worth automating.	PCAOB AS 2201, paragraph .27
Cost of an error	Two-tier, with a step at publication rather than a gradient. Inside the period an error costs a correcting entry. Once statements are issued the correction becomes a regulated act: across the European Economic Area in 2025, enforcers took action in 41% of ex-post examinations of IFRS financial statements, and 11% of those actions required immediate disclosure to the market by reissuing the statements or publishing a corrective note. Three issuers in the whole EEA were required to reissue.	ESMA, 2025 corporate reporting enforcement report
Reversibility	High before the ledger is locked, and one-directional afterwards. The route back is not an undo but a correction carried into a later period: of 240 enforcement actions taken across the EEA in 2025, 213 were satisfied by a correction in future financial statements. The window that matters is the period lock, not the date the error is found, and automation moves errors earlier in the window rather than making the window longer.	ESMA, 2025 corporate reporting enforcement report
Regulatory exposure	Named in the standard by process, not implied. The period-end financial reporting process is defined to include the procedures used to enter transaction totals into the general ledger, to initiate, authorise, record and process journal entries, and to record recurring and nonrecurring adjustments, and the auditor must evaluate its inputs, outputs and IT involvement. The fraud standard then points testing at exactly this window, on the reasoning that fraudulent entries are often made at the end of a reporting period.	PCAOB AS 2201 paragraphs .26 and .27, AS 2401 paragraphs .58 to .62
Vendor market maturity	Mature at matching and orchestration, and mostly bundled into the accounting package or the ERP rather than bought beside it. No public measure of market concentration in this category was retrieved, so none is quoted here. What is documented is how maturity changes the audit treatment: an automated application control need not be retested every year where IT general controls are effective and neither the control nor the environment has changed, which makes change management and access the object of examination instead.	PCAOB AS 2201, Appendix B paragraphs .B28 to .B33

Two of those rows carry the argument. The exception-variance row says the residual work is generated by inputs that arrive from elsewhere, which is a dependency problem rather than a processing problem. The regulatory row says the residual work is inspected by name: the entries produced at period end are the specific population an auditor is directed to test, so removing the person who typed them does not remove the obligation to explain them.

The reversibility row is the quiet one and it sets the design. A close is a window, not a stream. Everything caught before the lock is cheap and everything caught after it travels into a later period as a correction. That is an argument for automating the checks that run early and often, and against automating anything that would delay the moment an exception first becomes visible.

Why a faster close is not the same as a smaller close

A close ends when its last dependency arrives and is resolved. Automation that removes hours from a step which is not on that path removes real work and shortens nothing that anyone measures.

This page opened with the Whole of Government Accounts because it is the cleanest public proof of the point being made here. The reason the accounts arrived fifteen months late was not processing capacity: of the 407 English local authorities in scope, 167 submitted no data, 224 were consolidated on unaudited figures, and only 16 produced audited accounts. No amount of automation inside the consolidation shortens a close whose binding constraint is data that has not been sent.

The same shape governs a company close at a smaller scale. If the group’s third subsidiary always reports on day six, the close cannot finish on day four however good the matching engine is. This is why the first useful measurement is not effort at all. Record, for the last twelve closes, the timestamp at which each input became available and the timestamp of each sign-off. The gap between the last input and the final sign-off is the only part of the calendar automation can compress. The rest is a dependency negotiation with other teams, and buying software for it is a category error.

What automating the close is not

It is not a headcount argument. The residual work is the reviewing kind, and the reviewing is itself inspected: the audit standard requires an understanding of the nature and extent of oversight by management, the board and the audit committee over the period-end process. A close that produces numbers faster and reviews them less has moved backwards on the dimension the standard actually cares about.

It is not fewer controls. An automated posting is still a journal entry, and journal entries and other adjustments made at period end are the population an auditor is told to focus testing on, precisely because that is where manipulation of the reporting process tends to appear. What changes is the form of the evidence. It stops being a preparer’s working paper and becomes a system’s configuration, its authorisation record and its change log, which is why benchmarking an automated control depends on IT general controls holding and on nothing in the control or the environment having moved. Automation converts an evidencing problem into an access and change-management problem. That is a good trade, and it is a trade rather than a removal.

It is not a data quality strategy either. Whether the underlying ledgers are complete, consistently coded and current is a readiness question with its own answer, and a close automated over ledgers nobody trusts produces a wrong trial balance sooner. Naming that question early is cheaper than discovering it in the second month of a build.

And it is not the reporting layer. Machine-readable filing added a step to the end of the close rather than removing one: issuers subject to the European transparency rules must publish annual financial reports in a structured format with the consolidated IFRS statements marked up in inline XBRL. In 2025 enforcers examined the markups of 613 issuers and took action against 80 of them, an action rate of 13%, most requiring a correction in a future annual financial report. Tagging is work that did not exist twenty years ago, sits after the numbers are final, and belongs to the close.

The verdict

The evidence supports automating a month end close early and thoroughly on collection, matching and coordination, and supports expecting the return in reliability and evidence rather than in days. The profile is what settles it. The exception-variance row says the residual work arrives from other people, so the compressible portion of the calendar is smaller than the compressible portion of the effort. The volume row says the cost scales with entities, accounts and hand-offs rather than with transactions, so the payback condition is a structural one: several ledgers or entities, several people handing work to each other, and a fixed date. Where one person can close the books inside the available days without waiting on anyone, the checklist is worth automating and nothing else is.

The reversibility and regulatory rows decide the shape of what gets built. Because the window closes and errors after it travel into a later period as corrections, the value sits in checks that run early and continuously rather than in a faster run of the same checks at the end. Because the period-end process and the entries it produces are named in the audit standards as objects of examination, every automated posting has to carry its rule, its authorisation and its change history with it. A close automation that cannot show why an entry was made and who approved the rule has replaced a defensible manual process with a faster undefendable one.

The honest expectation to hold, then, is a close that is more reliable, better evidenced and roughly the same length, until the dependency conversation with the teams upstream is also had. Foxnut Studios builds systems of this kind and hands them over, which is a reason to be direct about the case where the answer is no: a company whose close is late because a subsidiary reports late does not have an automation problem, and the correct first purchase is a measurement of its own critical path rather than software.

If your close is slow because the work inside it is slow, not because the inputs arrive late from elsewhere, that measurement is worth doing with us rather than alone. Talk to us about where yours is stuck.

Sources

Foxnut Studios works on briefs like this one from Bengaluru and Paris. If you want the shape of that before you talk to anyone, here is how an AI engagement is scoped and priced.