4-day close (down from 15) · 92% touchless 3-way matching · $320k annual savings · 1 finance team, same headcount, more sleep.
Forge Robotics' starting point — 15-day close, growing pains
Forge Robotics is an industrial automation company that builds warehouse robots. They grew from 40 employees to 280 in eighteen months, and somewhere around employee 180 their finance function quietly snapped. Same five-person team, same NetSuite, ten times the invoice volume, three new entities, and a new revenue model with hardware deferrals that made every reconciliation a small forensic exercise.
By Q4 2025, their month-end close routinely took 15 business days. That meant management reporting landed three weeks into the next month — useful for archaeology, useless for decisions. The CFO, Priya Mehta, had been hired six months earlier with a mandate to professionalize the function. Her honest assessment in our first call: "We're not closing the books. We're catching up to them."
Two things were eating the calendar. First, invoice intake was almost entirely manual — vendors emailed PDFs to a shared inbox, an AP analyst opened each one, keyed line items into NetSuite, and routed for approval. Second, three-way matching (PO + receipt + invoice) was a spreadsheet exercise that broke every time a vendor partial-shipped or split a PO across multiple invoices.
Week one: invoice intake automated
We don't do six-month implementations. Priya's team went live with Atlas on a Tuesday, and by Friday of that same week, the shared AP inbox was being processed automatically. Atlas connected directly to the email mailbox, the vendor master in NetSuite, and the company's document store. Every incoming PDF or scanned image was parsed, normalized against the vendor record, and posted as a draft bill — header, lines, GL coding, tax — within about 90 seconds of arrival.
The first week's accuracy on header-level parsing was 97.4%. Line-level coding was 88% on familiar vendors and lower on new ones, which is exactly what you'd expect. Atlas flagged the rest for human review with the specific fields it wasn't confident about, instead of dumping the entire invoice back on the analyst.
The AP analyst, Marcus, told us in week two that his job had shifted from "data entry plus a little judgment" to "judgment plus a little data entry." That's the shift we're optimizing for.
Week three: 3-way matching went touchless
The bigger unlock came in week three. Atlas connected to Forge's procurement system and warehouse receiving feed and started running three-way matching automatically on every PO-backed invoice. Match tolerances were configurable per vendor class. Partial shipments were tracked across multiple invoices. Quantity-and-price mismatches were surfaced with a side-by-side view of the original PO, the receipt, and the invoice line.
By the end of week three, 92% of PO-backed invoices were matching without any human touch. The remaining 8% were going to the right person with the right context, instead of sitting in a shared queue waiting for someone to figure out who owned them.
What surprised us was how much downstream friction that one workflow removed. Vendors stopped calling about late payments, because invoices were actually moving. Buyers stopped fielding "is this PO closed?" questions, because the system knew. The controller stopped doing manual journal entries for accruals, because Atlas was posting them based on receipts.
Month one: the close shrank to 4 days
The first close after Atlas went live was day 9. The second was day 5. The third — last month — was day 4. We're not claiming a smooth curve; the team worked hard to redesign the close calendar, and there were a few painful days re-mapping accounts. But the underlying mechanic is straightforward: when AP is current in real time, the close stops being a catch-up exercise.
Specifically: by the time the books closed on day 4, three things that used to consume most of week one were already done. Bills were posted (Atlas had been posting them daily). Accruals were calculated (Atlas had been doing this from the receipt feed). Bank reconciliations were 80% complete (Atlas had been auto-matching transactions throughout the month). Day-of-close work was now genuinely close work: review variances, post adjusting entries, run reports.
Priya's quote in our QBR: "I have a finance function again. Not a clerical function with a finance hat on."
What changed for the finance team's day-to-day
The headline number is $320,000 of annualized savings — partly from not backfilling two open roles, partly from late-payment penalties they no longer pay, partly from early-payment discounts they can now capture because invoices are processed in days instead of weeks. That number is real, but it's not the part Priya talks about most.
What she talks about is what the team is doing instead. Marcus, the former AP analyst, runs vendor cost analysis now. The senior accountant is leading the implementation of a new revenue recognition model. The controller is building out FP&A capability that used to be outsourced. The team didn't shrink; the work changed.
If you'd asked Priya in November what she needed, she'd have said "two more headcount." If you ask her now, she'd say "one more senior analyst, and they're going to do interesting work." That's the right kind of hiring problem to have.
Forge is still rolling out further Atlas workflows — vendor onboarding, expense management, intercompany reconciliations. We'll write about those when the numbers settle. But the close is the one that mattered, and it's done.