this post. one anonymized result from a finance team we built for. a multi-entity professional-services group cut its month-end close from 5 days to under 2. an ai assistant matches the bank-to-ledger lines. a person still signs off before anything posts.
the buyer here is a financial controller at a multi-entity services group. two people ran the close. every month they spent the first 5 working days tying out bank statements against the ledger, entity by entity, before anyone could look at a variance. most of that time was matching lines by hand.
we shipped a reconciliation assistant that auto-matches about 90% of the bank-to-ledger lines and flags the rest for a human. the close now runs in under 2 days. nothing posts without sign-off, and the full audit trail is retained. this post shows what changed for the finance team, not how we wired it. the build maps to the reconciliation use-case at /ai-for/finance and the workflow-automation build at /platforms.
#what actually changed for the two-person close team?
the manual tie-out is gone. a person no longer scrolls two statements side by side looking for the line that doesn't agree. the assistant reads the bank feed and the ledger for each entity, matches what it can, and hands back a short list of exceptions. the two people who used to spend 5 days matching now spend under 2 reviewing the exceptions and reading the variances. that is roughly 3 working days each reclaimed, every month.
how does ai for month-end close actually work?
in this case, an ai assistant reads each entity's bank feed and ledger, auto-matches about 90% of the lines, and flags the rest for a person to review. nothing posts without human sign-off, and a full audit trail is retained. the close dropped from 5 days to under 2. the team spends the reclaimed time on variance analysis, not tie-outs.
#does a human still sign off before anything posts?
yes. every posting is approved by a person. the assistant matches and proposes. it does not post on its own. the roughly 10% of lines it can't confidently match are the ones a human looks at first, because those are where the real judgment lives. the audit trail records what matched automatically, what a person changed, and who approved it. an auditor can trace any line back to the source.
- the assistant auto-matches about 90% of bank-to-ledger lines per entity
- the remaining ~10% are flagged as exceptions for a person to review
- nothing posts without human sign-off; the full audit trail is retained
- close time: 5 days before, under 2 after
- the two-person team reclaimed roughly 3 working days each per month, moved to variance analysis
we are
we ship a result you can measure, 5 days to under 2, with a person approving every posting and a full audit trail retained.
we aren't
we are not selling an autonomous close bot that posts journal entries on its own and asks your controller to trust it.
#why a new metric, and why publish it anonymized?
most of our finance receipts have been about time-to-pay, like the invoice-processing build we wrote up in the /journal. close days are a different metric for a different job. we publish this one because a controller staring at a 5-day close wants to see close days move, not time-to-pay. the client asked not to be named yet, so we are honoring that. the number is real and the build is real.
if you run a close and the first week of every month disappears into tie-outs, that is the shape of problem this fits. we would rather show you one honest receipt than a pitch. tell us what you're building and we will tell you whether close days are the right metric to move first.