~€450M revenue · 180 sites · 40+ legal entities across three countries · 30+ bolt-ons in four years
A healthcare platform built through thirty-plus acquisitions closed its books in Excel, one entity at a time. We built a semantic layer across every entity and put close orchestration agents on top of it — eliminations proposed, exceptions surfaced, variance narrative drafted each morning. The sponsor now sees board-ready numbers on day six.
An ERP sat at the center — but the consolidation finished in Excel. Month-end was a manual relay: intercompany eliminations chased over email, entity trial balances pasted into a master workbook, variance commentary written from scratch every period. There was no exception view, so errors surfaced only at review. The sponsor never saw clean numbers until week three of the following month.
The Calendar
The Close Calendar
The Foundation
Before any automation, we built the map: how every entity relates, how every local chart of accounts translates to the group's, which balances mirror each other across borders. It was built by sitting with the controllers and analysts who own each number — not inferred from exports. Everything the agents do stands on it.
40+ Legal Entities · Three Countries
One Layer
What We Built
The agents watch the ledgers as entities post. Intercompany balances auto-match across the group — management fees, loans, recharges — and the eliminating entries arrive proposed, not chased. What used to be an email thread across three countries is a queue the controller clears before lunch.
Every proposed entry is logged, reversible, and attributable. The controller stays in control — the mechanics just stop being theirs to carry.
Accounts breaching tolerance get flagged the day they breach — with the entity, the amount, and the likely cause attached. Errors used to hide until review in week two. Now there is one exception view for the whole group, and it’s empty more mornings than not.
The close stopped being a search for what’s wrong. The system brings what’s wrong to the people who can fix it.
Each morning of the close, the flux and variance narrative is already drafted — grounded in the actual movements, written in the house style, with the numbers cited. The controller’s job becomes judgment: sharpen the story, kill the noise, sign off.
Writing commentary from scratch every period was the quiet tax on the whole team. It’s gone.
This group buys companies. Every acquisition used to reset the close — new accounts, new mappings, new spreadsheet tabs. Now a new entity gets mapped into the layer once, and the agents treat it like every other entity from its first month-end.
The buy-and-build machine keeps running. Finance no longer sets the pace of integration — it absorbs it.
Close Morning
Entities closed
34 / 42
Eliminations proposed
12
awaiting review
Tolerance flags
5
open
Proposed Eliminations
Entity 07 ↔ Entity 19
Management fee · €1.24M
Entity 03 ↔ Entity 31
Intercompany loan interest · €412K
Entity 12 ↔ Entity 26
Shared-services recharge · €188K
Tolerance Flags
6400 · Agency staffing
Entity 22 · +€212K vs. tolerance ±€75K
7010 · Site rent
Entity 09 · +€96K vs. tolerance ±€40K
4200 · Payor receipts
Entity 34 · −€310K vs. tolerance ±€120K
Flux Narrative · Draft
Drafted 06:58 · awaiting controller edit“Group revenue closed 2.1% above plan, driven by higher visit volume across the southern region and two sites annualizing ahead of schedule. Staffing costs breached tolerance in Entity 22 — agency usage up €212K on unfilled clinical rosters, consistent with the pattern flagged in the prior period…”
One view for the whole group. Everything proposed, nothing posted without a person.
Close speed is a board-control metric, not a back-office one. A sponsor flying blind for three weeks can't act.
The Impact
The close runs in five business days, and the sponsor holds board-ready numbers on day six — two weeks earlier than before. Roughly three full-time roles' worth of mechanical work moved to analysis: variance investigation, site economics, deal support.
Reforecasts run weekly now, up from monthly — because the numbers are trustworthy enough, early enough, to forecast against. Pricing moves, staffing decisions, and capital allocation happen inside the month they're needed, not after it.
And the next acquisition doesn't threaten any of it. New entities onboard into the layer — the close stays at five days no matter how many logos join the group.
We design and deploy AI-native systems for companies moving fast in competitive markets.
Talk to LightCI