All posts

· Javier Leguina· Automation in practice

Rebuilding the month-end close, from trial balance to statements

The monthly close is the cleanest worked example in operations because the benchmarks are public. We walk one redesigned close end to end, from where the days actually go to what a finance team does with them back.

The monthly close is the cleanest worked example in operations because the benchmarks are public. APQC, drawing on roughly 2,300 organizations, puts the median monthly close at 6.4 calendar days, with top-quartile finance teams getting it done in 4.8 days or fewer and the bottom quartile taking ten or more. A separate benchmark, drawn from more than 10,000 organizations via Rand Group, reports the same distribution: top performers close in five days or fewer, the median around six, the laggards at ten and up. Period-end management reporting stretches the spread further, running from six days at the top quartile to ten at the median and fifteen at the bottom. The reason the close is worth walking step by step is that the difference between the top and bottom of that distribution is five or more full working days a month, and almost none of those days are spent on accounting judgment.

Where the days actually go

If you sit behind a controller for a close, the work is not evenly distributed across the calendar. It piles into the first stretch, and it piles into two activities that have nothing to do with how the numbers should be interpreted. The first is chasing data: the inventory count that hasn't come over from the warehouse system, the credit-card feed that posts late, the intercompany balance that one entity recorded and the other didn't, the accrual estimate that lives in someone's head until you ask. The second is reconciliation: bank to ledger, subledger to general ledger, the supporting schedule to the trial balance, line by line, account by account, with the controller flagging anything that doesn't tie. By the time the trial balance is clean enough to trust, most of the cycle is spent. Building the statements and writing the variance commentary, the part that actually requires a qualified accountant, takes only the final stretch.

That distribution is the whole opportunity, and it is also why the close is a textbook case for the discipline Michael Hammer argued for thirty years ago, which we've written up in don't automate, obliterate. The reflex in most finance teams is to make the chasing faster: a status spreadsheet, a daily standup, a dashboard of open items. That speeds up a step that should not exist in its current form. The redesign question is not "how do we chase faster" but "why is anyone chasing at all," and the answer is almost always that the data is moving between systems by hand because no one ever built the path for it to move on its own.

What is mechanizable and what isn't

Walk the redesigned close in order. Pulling the trial balance from the accounting system, pulling the bank feed, pulling the subledger detail, matching transactions that obviously correspond, and flagging the ones that don't is mechanizable in full; it is rule-bound, high-volume, and the same every period. Standard recurring accruals, the depreciation run, the routine intercompany eliminations, and the first draft of the consolidated statements are mechanizable. Generating the variance schedule, computing every line's movement against prior period and budget, and surfacing the ones outside threshold is mechanizable.

What is not mechanizable, and what should not be, is the judgment at the exceptions. When a reconciliation won't tie because a customer paid the wrong invoice, when an accrual estimate depends on a contract negotiation still in flight, when a variance is large for a reason the numbers can't explain, a human stays in the loop and makes the call. The agent's job is to do the ninety-something percent that is mechanical and to hand the controller a short, ranked list of the few things that genuinely need a person, with the supporting detail already assembled. That is the same human-in-the-loop posture we've described in what an aligned AI engagement actually looks like: the machine clears the volume, the accountant keeps the authority.

Writing back into the ledger

The hard part, in practice, is rarely reading the data; it's posting the result. Most of these finance teams run on QuickBooks or a legacy ERP whose journal-entry interface was built for a person clicking through a form, not for a program. An agent that can reconcile beautifully but can't post the adjusting entries has only moved the bottleneck. This is why a serious redesign treats the write-back path as a first-class engineering problem rather than an afterthought, a topic we've taken up directly in writing back into a system with no usable API. Getting an entry to land correctly and reversibly inside a system that was never meant to be driven by software is what separates a close that is genuinely faster from a demo that looks faster.

The before and after, in days

Hold the redesign against the benchmark. A team sitting at the bottom of the distribution, closing in ten or more days, spends most of those days on the chasing and reconciling that the redesigned process mechanizes. Moving the mechanical work off the controller's desk doesn't shave a step; it collapses the front half of the cycle, where most of those days are spent. The realistic target is the top quartile the benchmark already documents, the difference between a ten-day close and a five-day one, expressed not as abstract efficiency but as five working days a month that the finance team gets back, every month, to do forecasting, to chase margin questions, to give the operating partner an answer in the same week it's asked rather than the week after the books finally tie.

Why now, and the honest objection

The reason this isn't a someday argument is that the field is already moving. APQC reports that thirty-one percent of organizations actively use AI in record-to-report, with another thirty-nine percent in early adoption, which means a finance team still closing by hand is already behind a third of its peers and most of the rest.

A reasonable counter is that close benchmarks are dominated by company size and ERP maturity, not by automation, and that a mid-market firm on QuickBooks will never close like a Fortune 500 on a modern consolidation platform no matter what you bolt on. There's real truth in that. Some of the spread between top and bottom quartile is structural, and no agent changes the fact that a complex consolidation has more to reconcile. But the days that get reclaimed in the redesign are not the structural ones; they're the chasing and the manual matching, and those exist in equal measure at every size, which is precisely why the median has sat above six days across thousands of organizations for years. The structural floor is real, but most teams are nowhere near it, and the distance between where they are and that floor is the work.

Common questions

How many days does a typical company take to close the books each month?
APQC, drawing on roughly 2,300 organizations, puts the median monthly close at 6.4 calendar days, with top-quartile finance teams closing in 4.8 days or fewer and the bottom quartile taking ten or more. A separate benchmark from Rand Group, drawn from more than 10,000 organizations, reports the same shape: top performers close in five days or fewer, the median around six, and the laggards at ten and up. Period-end management reporting stretches the spread further, running from six days at the top quartile to ten at the median and fifteen at the bottom.
Which parts of the month-end close can be automated and which still need an accountant?
Pulling the trial balance, bank feed, and subledger detail, matching transactions that obviously correspond, and flagging the ones that don't is fully mechanizable, as are standard recurring accruals, the depreciation run, routine intercompany eliminations, the first draft of consolidated statements, and the variance schedule. What is not mechanizable is the judgment at the exceptions, such as a reconciliation that won't tie because a customer paid the wrong invoice or an accrual that depends on a contract still in negotiation. In those cases a human stays in the loop and makes the call, while the agent clears the mechanical volume and hands the controller a short, ranked list of the items that genuinely need a person, with supporting detail already assembled.
If our close is slow mainly because we run QuickBooks instead of a modern ERP, will automating it actually help?
Some of the spread between the top and bottom quartile is structural and tied to company size and ERP maturity, and no agent changes the fact that a complex consolidation has more to reconcile. The days reclaimed in a redesign are not the structural ones, though; they are the chasing and manual matching that exist in equal measure at every size, which is why the median has sat above six days across thousands of organizations for years. The structural floor is real, but most teams are nowhere near it, and the distance between where a team is now and that floor is the work.