A mid-market manufacturer's finance team spent most of each month assembling numbers — the close dragged, variances were explained from memory, and the CFO's transformation agenda kept stalling.
Day 2
exceptions surfaced — instead of day 9
9 areas
scoped to the two that mattered most
WHAT WE DID
Applied AI to close automation, variance detection, and anomaly flagging — grounded in clean data and governance
Redesigned R2R handoffs so nothing waited on individuals
Coached the in-house team so the capability stuck
WHAT CHANGED
The close accelerated and stopped consuming the month
Variance stories arrived with the numbers, not weeks later
Finance hours shifted to judgment calls, not assembly
The situation
A mid-market manufacturer's finance team spent most of each month assembling numbers. The close dragged, variances were explained from memory, and the CFO's transformation agenda kept stalling.
Assembly work expands to fill a close. When the numbers arrive late, the explanation of the numbers arrives later still, and by the time anyone can act on a variance the period it belongs to is two months gone. The team is busy throughout, which is what makes the problem hard to argue with internally.
The transformation agenda stalls for a structural reason rather than a motivational one. It requires the same senior people who are currently assembling, and there is no month in which assembly is optional.
What we did
Applied AI to close automation, variance detection and anomaly flagging — grounded in clean data and governance. The qualifier carries the weight. Automation applied to inconsistent data produces faster wrong answers, and anomaly detection without governance produces alerts nobody trusts enough to action.
Redesigned R2R handoffs so nothing waited on individuals. Most close delay is queueing, not processing. Work sits waiting for the one person who knows how a particular reconciliation is done, and every such dependency is a day.
Coached the in-house team so the capability stuck. A close that only runs while an external team is present has not been fixed; it has been rented.
What changed
Exceptions began surfacing on day 2 instead of day 9, and a scope of nine potential areas was narrowed to the two that mattered most.
The close accelerated and stopped consuming the month. Variance stories arrived with the numbers rather than weeks later. And finance hours shifted toward judgment calls instead of assembly.
The day 2 figure is the one with downstream consequences. An exception found on day 2 can be investigated and corrected within the same close. On day 9 it is a note explaining why the number is what it is.
What this means for finance teams considering AI
The useful applications are unglamorous: detection, flagging, and the removal of handoffs. They work because they attack queueing and repetition, which is where close time actually goes.
The precondition is the part most often skipped. Clean data and defined governance are not preparatory steps before the interesting work; they are what determines whether the output is trustworthy enough to act on. Applied to a messy ledger, anomaly detection produces noise that trains the team to ignore it.
Scope discipline matters as much as the technology. Nine candidate areas narrowed to two is not a reduction in ambition — it is the recognition that a capability which lands properly in two areas will spread, and one spread thinly across nine will not.