The four-day close: what a mid-market group actually changed

A multi-entity group turning over around £400m has taken its close from six working days to four over three quarters. The finance director was refreshingly blunt about where the saving came from. It was not the vendor's flagship autonomous journal feature, which remains switched off. It came from two unremarkable places: intercompany reconciliation matching, and first-draft variance commentary.

Reconciliation matching is the clean use case. The group had roughly eleven thousand intercompany lines a month and three people working them. A matching model now clears the high-confidence majority and routes the residue to a queue with a suggested explanation attached. The team still reviews everything above a threshold. The saving is not the matching itself but the collapse of the back-and-forth between entities, which historically consumed the second and third days of the close.

Variance commentary was the surprise. Draft flux narratives generated from the ledger and prior-period commentary are, by the group's own assessment, roughly seventy per cent usable. Nobody publishes them unedited. But a controller editing a draft finishes in fifteen minutes what previously took an hour of blank-page writing, and the drafts are consistent across entities in a way human commentary never was.

The cost line is worth noting. Licences were a minority of total spend. The bulk went on master data remediation — supplier records, entity mappings, chart of accounts alignment — work the group had deferred for years and could no longer avoid once a model was consuming the data. The FD's summary: the pilot did not fail on the AI, and the second pilot succeeded because they had already paid for the plumbing.

Headcount is unchanged. Two of the three reconciliation staff moved to commercial analysis roles that were previously vacant. Whether that counts as a benefit depends entirely on what the board was promised at the outset, which is why we keep saying the promise matters more than the technology.