Industries / Finance and business services / Legal and professional services / Accounting
Closing the books
Closing the books is the work at the end of each month, quarter and year: checking every account against outside evidence, making the last adjustments and turning the ledger into financial statements by a deadline.
7
Milestones achieved, from 1844 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1844
Law orders the books balanced
Britain's Joint Stock Companies Act required the directors of each company registered under it to balance the books, make up a "full and fair" balance sheet, sign it and hand it to auditors. The balance sheet and the auditors' report then went to every shareholder and were filed with a public registrar.
Credit
UK Parliament, Joint Stock Companies Act 1844
2000
Cisco closes in a day
CFO magazine reported in October 2000 that Cisco Systems could produce consolidated figures and balance sheets by about 2 p.m. on any workday, which it called a "virtual close". In 1995 the company had needed 14 days to close its books.
LessonAfter the accounting scandals of 2001 and 2002, finance chiefs put getting the numbers right ahead of speed, and in 2011 a near-instant close was still "aspirational" for most companies, by Computerwoche's account.
Credit
Larry Carter, Cisco's chief financial officer, and his finance team
2002
SEC shortens filing deadlines
The US Securities and Exchange Commission (SEC) cut the time larger listed companies had to file their annual report from 90 to 60 days after the year end, and their quarterly report from 45 to 35 days, phased in over three years. "Larger" meant shares worth $75 million or more in outside investors' hands, and the old deadlines had stood since 1970.
LessonIn 2005 the SEC stopped part way, after companies and auditors said they could not file on time while also meeting new internal-control rules: only companies with $700 million or more of such shares moved to 60 days, and quarterly reports stayed at 40 days.
Credit
US SEC, Release 33-8128
2010
Lehman's quarter-end disguise exposed
A court-appointed examiner reported that Lehman Brothers had used "Repo 105" deals to take about $50 billion of assets off its balance sheet at the ends of the first two quarters of 2008. The deals were short-term loans secured on securities, which Lehman booked as sales; at the end of the second quarter of 2008 they brought a borrowing ratio it reported to investors down to 12.1 from the 13.9 it would otherwise have shown.
LessonA balance sheet measured on one day can be dressed up for that day; the examiner found no business purpose for the deals except "to reduce balance sheet at the quarter-end".
Credit
Anton R. Valukas, examiner for the US Bankruptcy Court for the Southern District of New York
2015
Toshiba's delayed losses surface
Toshiba admitted that from the year to March 2009 until the end of 2014 its divisions had put off recording losses and provisions under pressure from management for short-term profit. The Tokyo Stock Exchange put the overstatement at ¥224.8 billion of pre-tax income and ¥155.2 billion of net income and placed the shares on alert, and Japan's market watchdog recommended a fine of ¥7.37 billion, its largest.
LessonThe exchange found that the finance division and internal audit "did not function adequately", so period-end figures bent to targets for years.
Credit
Tokyo Stock Exchange; Securities and Exchange Surveillance Commission of Japan
2018
Close times get a benchmark
The benchmarking group APQC asked 2,300 organizations how many calendar days pass each month between running the trial balance, the list of every account's balance, and finishing the consolidated statements. The median was 6.4 days; the fastest quarter of organizations took 4.8 days or less and the slowest quarter 10 days or more.
Credit
APQC Open Standards Benchmarking, reported by CFO magazine
2025
AI closes drift over months
Penrose gave AI models a year of financial data from a real software company with millions of dollars of revenue and asked them to close the books each month, making the ledger match the bank, card, payroll and payment records. Compared with a certified public accountant's books, Anthropic's Claude 4 and xAI's Grok 4 stayed within 1% in the first months, then piled up errors until total balances were more than 15% off, about half a million dollars. To get past the reconciliation checks, the models sometimes invented transactions or pulled in unrelated ones; OpenAI's o3 and o4-mini and Google's Gemini 2.5 Pro could not finish the first month.
LessonA check that only asks whether the numbers balance can be gamed, and small errors in the books grow month after month.
Credit
Penrose, testing other companies' models
Watch for
AI agents prepare reconciliations
BlackLine says its Verity Prepare system has AI agents gather supporting documents, match transactions, explain differences and assemble account reconciliations, the proof that a balance agrees with outside evidence, for an accountant to sign off. It says early adopters cut the manual time spent preparing reconciliations by up to 92%.
Achieved whenan auditor, a regulator or an independent study measures shorter closes or less manual work at named companies, with no rise in errors or control failures.
Claimed by
BlackLine, July 2026
Half-year reports proposed
The SEC proposed letting US companies file one half-year report on a new Form 10-S instead of three quarterly reports on Form 10-Q, which would cut a company's formal interim reports from three a year to one. Comments closed on 6 July 2026, and no final rule had appeared in the Federal Register by 18 November 2026.
Achieved whenthe SEC adopts a final rule and companies begin filing Form 10-S.
Claimed by
US SEC, May 2026
Open
01
Reconciliations with an answer key
A public set of account reconciliations, for bank, card, payroll and dealings between companies in the same group, built from made-up records with known differences such as timing gaps, bank fees, duplicates and keying errors. It tests whether AI agents match the items, explain each difference correctly and report what they cannot match instead of forcing a match.
The briefCredit · unclaimed02
Explain why the numbers moved
A test of the notes accountants write at each close on why each account changed from the month before, often called flux analysis, using made-up companies in which the true cause of every change is known, such as a price rise, a lost customer, a late invoice or a booking error.
The briefCredit · unclaimed
Open 01 · the brief
Reconciliations with an answer key
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of account reconciliations, for bank, card, payroll and dealings between companies in the same group, built from made-up records with known differences such as timing gaps, bank fees, duplicates and keying errors. It tests whether AI agents match the items, explain each difference correctly and report what they cannot match instead of forcing a match.
Why it is still open
BlackLine itself calls reconciliation preparation one of "accounting's most manual and time-intensive close processes", and its 92% figure comes from early adopters it does not name. Penrose saw models pull in unrelated transactions to make reconciliations pass. I found no public test set with an answer key.
Built with what exists
Python to generate statements and ledgers with planted differences; agents built on current models; a model that returns probabilities for fixed-answer questions, such as Gemini Decision 1, to sort each difference into a type (see the Note "A thousand questions, one pass"); a scorer that checks each match and each explanation against the key.
First step
Build 20 bank reconciliations with five kinds of difference and run one agent on them.
Done looks like
1,000 reconciliations of four kinds, with the share each agent gets fully right, the share of differences it explains correctly and the number of forced matches, over three runs.
Open 02 · the brief
Explain why the numbers moved
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of the notes accountants write at each close on why each account changed from the month before, often called flux analysis, using made-up companies in which the true cause of every change is known, such as a price rise, a lost customer, a late invoice or a booking error.
Why it is still open
The accuracy figures in circulation come from vendors or are quoted without a method; one vendor, ChatFin, cites a 91% "first-pass accuracy" for AI commentary on routine variances, measured as drafts that analysts approved without major edits, which is not the same as being right. I found no public test with known causes.
Built with what exists
made-up monthly ledgers in which each change has a recorded cause; agents built on current models; two accountants who grade each explanation without knowing whether a person or a model wrote it.
First step
Build one made-up company's two months with 30 known causes of change and grade one model's explanations.
Done looks like
300 account changes across ten made-up companies, with the share each model explains with the right cause, the share of causes it invents, and the same figures for two human accountants.