Why does my close take eleven days?
Most of a close is not arithmetic. It is a search for things that should have happened and did not: the invoice a supplier never sent, the accrual nobody raised, the bank line that never matched. A conventional ERP cannot help with any of it, because every one of those is an absence, and there is no query that finds a row that does not exist. So the work is done by people who remember what usually happens, at the end of the month, under time pressure.
The shape of the problem
Ask what is wrong with the ledger and you get a report. Ask what is missing from it and you get a person with a spreadsheet who knows that the Dubai office usually bills around the twentieth.
Every exception mechanism in a conventional system is a query over rows. Select the invoices where something is true. There is no predicate that reaches an invoice that was never created, so the entire class of failure that actually delays a close is invisible to the tooling.
What has to exist first
An expectation. Before an event fails to arrive, something has to have recorded that it was expected: this supplier bills monthly and has for two years, this statutory filing falls on this date, this contract obliges a delivery this quarter. Then the absence has a row and the absence becomes queryable.
That is the difference between a close that is a search and a close that is a list. The expectations are generated from what the organisation already does rather than typed by somebody, and the ones that did not arrive are what a close actually works through.
And a score rather than an event
A close that is a binary state can only be open or shut, so the question during the month is unanswerable and the question at month end is a scramble. A close health score is continuously computed, denominated in performance materiality rather than in item counts, and decomposable into the named breaches that produced it.
Denominated in materiality matters. Forty unmatched bank lines worth eleven pounds between them is not a worse position than one unreconciled account holding four hundred thousand, and a score built from counts says it is.
The part nobody enjoys hearing
A score that cannot be raised by suppressing findings is a score people will occasionally dislike. Certifying a period over an open breach demands a written reason and records what was accepted, which is the point.
Six of the eleven scoring domains in this system do not exist yet, and their weight is excluded rather than counted as passing. A score that silently treats absent checks as clean is worse than no score.
What this does not do
Three of six expectation generators are implemented, and every run is started by somebody rather than by a scheduler, so continuous here means re-runnable rather than running. There is no connector to any source system, so the events an expectation waits for arrive when somebody records them.