A controller sends the pack on day five. Clean close, everything tied out, a good month by any benchmark the firm tracks.
The Metric Everyone Optimised
On day six the questions start. Why is contractor spend up 22%. Which of those renewals actually landed in this period. Is that legal accrual the same matter we discussed in April. Each one takes twenty minutes to an hour, and the answer is never in the pack. It is in a contract, an invoice, an email thread and somebody's memory of a call.
APQC defines monthly close cycle time as the calendar days between running the trial balance and completing the consolidated financial statements, measured across 2,300 organisations, with a median of 6.4 days and a top quartile of 4.8 days or less. It is a good metric. Firms have chased it hard and many have won.
Look at where it stops. Trial balance to consolidated financial statements. The moment the statements exist, the measurement is over and the stopwatch goes back in the drawer.
Everything after that line is real work that no benchmark counts. The variance explanations. The board pack narrative. The three follow-up emails from the client's operations lead. The question in a Thursday call about a number from two months ago. None of it appears in the close cycle time, so none of it appears in the case for automating anything.
Which Is Exactly Where the AI Went
Deloitte's Finance Trends 2026 research, published in October 2025, found that 63% of surveyed finance leaders report they have fully deployed and actively use AI within the finance function. In the same research, only 21% say those investments are delivering clear, measurable value, and just 14% of the group seeing strong returns have fully integrated AI agents into specific areas of finance.
A 42 point gap between having the technology and being able to show what it did.
There are several explanations doing the rounds for that gap, most of them about model quality or change management. Here is a simpler one. Automation follows measurement. Everyone measured the close, so everyone automated the close: the reconciliations, the accruals, the consolidation, the parts that live inside the stopwatch. Those parts got faster. The part the client experiences as speed, which is how long it takes to get an answer to a question about the numbers, was never on the clock, so nothing was aimed at it.
Faster arrival at the same unanswered questions is not nothing. It is also not what the investment case promised.
Day Six Is a Retrieval Problem
Notice what answering a variance question actually requires. Not judgment first. Retrieval first.
The contractor spend question needs the contract, the amendment, the two invoices that straddle the period and the email where scope changed. Only once those four documents are on the desk does the accounting question become answerable, and the accounting part usually takes about ninety seconds. The hour went to the hunt.
That is why the work resists the tools aimed at the close. A close automation tool operates on the general ledger, and the answer to the question is not in the general ledger. It is in the documents behind the entries, scattered across a portal, a shared drive, an inbox and whatever the client sent over WhatsApp in March.
The consequences land on firm economics in a way that no close metric will surface. Question turnaround is what a client calls responsiveness. A firm that answers in an hour and a firm that answers on Tuesday are selling different services at the same price. And an answer assembled from memory carries a risk the pack never did, because when it turns out to be wrong there is no record of how it was reached.
What the Fix Looks Like
The principle first: the close produces numbers, and the month is not finished until the firm can support them. Support means the documents behind the entries are findable by whoever gets asked, in the time the asker is willing to wait.
MetaWurks is built for that half. It ingests a client's contracts, invoices, statements and correspondence and lets an accountant query them in plain English, so the contractor spend question is a question rather than an afternoon. Role based access controls decide who can open which client's file, and audit logs record who opened what and when. Documents ingested into the platform are not used to train models and are not exposed to other users, so the client's file stays inside the firm's control while it is being used.
It does not close the books. It answers the questions that arrive once they are closed, which is the half of the month that never made it onto anyone's dashboard.
The close got faster because it was measured. The rest of the month did not, for the same reason.
Join the Conversation
How long does it currently take your team to answer a client's question about a number in last month's pack, and does anyone in your firm actually track that?