Run the calendar
Every task with an owner, a due day, a dependency, and a state. The same list each month, which is what makes it improvable rather than heroic.
AI agents · finance
It runs the checklist, prepares everything mechanical, ties the subledgers, and chases whoever is blocking. The chasing is the least glamorous part and reliably the most effective, because a slow close is usually a queue of things waiting on somebody rather than a volume problem.
What it does
Each of these is work a person does today. The agent does them in sequence and stops at the first thing it is not confident about.
Every task with an owner, a due day, a dependency, and a state. The same list each month, which is what makes it improvable rather than heroic.
Reconciliations, accrual and prepaid schedules, depreciation, revenue recognition runs, and intercompany elimination — prepared and held for review.
Subledgers must agree with control accounts before the period advances. A break blocks progression rather than producing a note nobody reads.
It knows which task blocks which, who owns it, and how long it has been sitting — and it follows up without anyone needing to feel awkward.
Period-over-period movement explained by the transactions responsible, written as a first draft for the controller to edit.
Every task carries its supporting records, so the close file an auditor asks for in March already exists in February.
Authority
Status, blockers, ageing of open items, and close-duration analytics. Read-only and always on.
Schedules staged for review, reminders sent to task owners, commentary drafted. This is where the agent does most of its work.
Closing a period asserts the numbers are right, on behalf of a person, to an auditor or a board. It is absent from the permission model at every level.
Nobody notices the month the close went from eight days to eleven, because it did not happen in a month. It happened across four quarters, one added reconciliation and one departed colleague at a time, and by the time it is a problem it feels like the natural cost of being bigger.
The first thing a structured close produces is not speed. It is a measurement — which tasks took how long, which were waiting, which were rework. Most finance teams have never had that, and it usually contradicts what they assumed.
In the closes we have instrumented, the dominant category is waiting — for a feed to populate, for a manager to approve an accrual, for someone to answer a question about a variance, for the last three bills to arrive. Adding people to a queue does not shorten it, which is why hiring rarely fixes a slow close.
Rework is second: something did not tie, and the investigation eats a day. Continuous subledger tie-outs move that discovery to the day the break occurred, when the transaction is recent and someone remembers it.
Customers expect the value to come from automated schedules and reconciliations. Those help. But the task that most consistently moves close duration is the follow-up — knowing that day three’s depreciation run is blocked on a fixed-asset addition that operations has not confirmed, and asking about it on day three rather than day five.
It is also the part people are gladdest to delegate. Nobody enjoys sending the fourth reminder to a colleague, and the agent has no feelings about it.
Questions
Send your checklist and last three close dates and we will show you where the days are going.