The close is an evidence problem wearing an accounting costume
Module · Record to Report
Every close is run blind. The same checklist, a different jam each month, found on day four when nothing can be done about it.
Prophesee Accounting predicts where the close will run late before day one, scores every journal before it is reviewed, and carries each figure to its source so the explanation is written before it is asked for.
The close is not slow because accounting is hard. It is slow because the evidence is scattered.
Balances arrive in one system, support in another and the explanation in an inbox. The team is not reconciling the accounting, it is assembling the proof, by hand, every month, for figures it already believes. The spread between the fastest and the slowest close is almost entirely a spread in how that proof is gathered.
Sources: KPMG, Trends in Material Weaknesses, published 2026, covering FY2025 SEC filings (238 companies in the year, 740 unique companies across 2021 to 2025) · APQC, General Accounting Open Standards Benchmarking (n=2,300 organisations). Fieldwork dates from 2017 and is quoted here as the most widely used close benchmark, with that age disclosed · Financial Reporting Council, Annual Review of Corporate Reporting 2024/25, published 30 Sep 2025.
From assembling the proof to being handed it
ForesightThe checklist is the same every month and the jams are different every month. The bottleneck is found on day four, when there is no time left to do anything about it.
Models trained on the history of your own close predict which accounts, entities and tasks will run late, with the drivers behind each one. The controller sees the shape of the close before it starts and moves the people, rather than discovering the jam midway through.
PulseA journal posted at eleven at night on the last working day looks exactly like every other row in the log, and it is reviewed on a sample once the period has closed.
Every manual journal scored on pattern, timing, preparer, account and value before it reaches review, with postings outside the window or out of sequence raised as they happen. The reviewer works a ranked queue rather than a random selection.
HorizonThe calendar is redrawn each year on judgement. Nobody can show what moving a task, adding a person or changing a cut-off would actually do to the close.
Model the intervention: a revised cut-off, a rebalanced task list, an entity moved to a different sequence. Watch the projected close date update, then track the actual close against that plan so the improvement is attributable rather than assumed.
NexusThe auditor asks how a figure was arrived at, and three people spend two days reassembling something that was obvious to whoever built it four weeks ago.
Ledgers, subledgers, journals, support documents and written explanations resolved into one graph. Ask for any figure and get the source system, the transformation, the approver and the document behind it, with the explanation already drafted.
Turning close challenges into decisions
Forty tabs that one person understands, rebuilt every month, holding the balance sheet hostage until it agrees.
Pulse NexusA journal posted at eleven at night on the last working day looks exactly like every other row in the log.
PulseThe auditor asks how the accrual was reached, and three people spend two days rebuilding an answer that existed a month ago.
NexusBalances are looked at once the period has closed, on a sample, when there is no time left to do anything but explain them.
PulseThe same list every month, and a different jam every month, found on day four when nothing can be done about it.
Foresight13 AI applications that could be relevant
A sample of what becomes possible on the decision layer, not a fixed list: each application draws on the same data foundation and audit trail, and new ones are configured on the engines, not built from scratch.
The accounts and entities most likely to hold up the close, ranked before day one.
Movement against prior period decomposed and drafted before the review, not during it.
Expected accruals estimated from history and open commitments, with a band on each.
Every manual journal scored on pattern, timing, preparer and account before it is reviewed.
Postings outside the window, out of sequence or above threshold, raised as they happen.
Materiality, ageing and posting thresholds written in plain English and tested continuously.
Accounts moving away from their own pattern, raised during the period rather than after it.
Model a revised cut-off or a rebalanced task list and watch the projected close date move.
The actual close tracked against the modelled plan, so the improvement is attributable.
Every figure traced to its source system, transformation and approver in one click.
Balances and support matched probabilistically, with only the genuine residue put in front of a person.
The evidence for a question assembled and drafted, with the source behind each element.
Counterpart entries paired across entities and currencies, with the difference explained.
A day when the controller reviews instead of assembles
Today: Day one begins with the same checklist that failed to predict last month's jam.
ForesightFour entities and eleven accounts flagged as likely to run late, with the driver behind each. She moves two people on day zero.
Today: A manual journal at eleven at night, indistinguishable from every other row.
PulseThree postings out of sequence and one above threshold, scored and routed with the preparer, the account and the history attached.
Today: Redraw the close calendar on judgement, because nothing can model it.
HorizonA revised cut-off tested against a rebalanced task list. The projected close date moves by a day and a half.
Today: The auditor asks how the accrual was reached. Two days of reassembly.
NexusThe auditor's question answered with the source system, the transformation and the approver, in the time it takes to ask it.
Tomorrow's close becomes today's decision.
Know the close before it starts
We agree the metric and the baseline in week one, and measure the result on your data.