The situation
A 4,000-person industrial distributor was closing the books every month through a Rube Goldberg machine: 14 analysts, 37 spreadsheets, and a five-day reconciliation cycle that produced a variance report everyone already disagreed with by the time it landed in the CFO's inbox. The finance leadership wanted to "use AI" — but the real problem was the seam between their ERP, their FP&A tool, and the weekly operating cadence.
What we shipped
We embedded a three-person Vert3x team — a delivery lead, an ML engineer, and a former controller — inside the FP&A organization for a twelve-week engagement.
- A continuously-reconciled operating picture built on top of the existing ERP, eliminating the manual stitch between source systems.
- A forecasting model that produced a rolling four-quarter view updated daily, calibrated on the last eight quarters of their actuals.
- A variance-commentary agent that drafts the "why did this move?" narrative for each business unit — always reviewed by a human analyst before it leaves the building.
- A handoff playbook and in-house runbook that the client's FP&A team now owns.
Results
We stopped arguing about what the numbers were and started arguing about what to do about them. That's the whole point. — VP Finance
- Close cycle compressed by 62% — from five days to under two.
- ~480 analyst-hours reclaimed per month, redeployed to business-partnering work.
- Forecast accuracy improved 18 points at the four-quarter horizon.
- 100% transfer — the running system is now operated entirely by the client's in-house team, with Vert3x on call for quarterly reviews only.
Why it worked
The win wasn't the model. The win was picking one operating loop — monthly close and variance commentary — and refusing to ship anything that didn't collapse it. Embedded operators meant we could make the unglamorous decisions (which ERP fields to trust, which exceptions to route where) in real time, instead of negotiating them through a statement of work.