How a specialty foods distributor closed the books nine days earlier

The month everything was on fire

Folsom & Co. distributes high-end specialty foods — high-margin olive oils, high-maintenance cheese suppliers, high-volume restaurant clients. The company does about $340 million in annual revenue, moving product through nineteen freight providers across five warehouses.

In January 2026, the finance team hit a wall. Year-end close had taken twenty-two days. The CFO wanted it under ten. Margaux Bellanger, who'd been running finance operations for three years, knew the bottleneck wasn't her team's competence — it was the sheer volume of paper they were drowning in.

"We processed about 4,200 invoices a month," Bellanger says. "Every one of those touched at least three people before it got into NetSuite. Someone opened the email. Someone keyed in the data. Someone else approved it. And then someone else reconciled it at month-end."

The math was brutal: three full-time accountants, spending roughly 80% of their time on manual invoice processing. Not analysis. Not forecasting. Just typing numbers from PDFs into fields.

"We had three accountants whose entire job was chasing paper. Now they do analysis. Nobody got fired — they just stopped hating their jobs."

— Margaux Bellanger, Director of Finance Operations

What the old process looked like

The invoice lifecycle at Folsom & Co. before automation was a case study in friction. Here's what happened every time a freight provider sent an invoice:

1. The invoice arrived via email — sometimes as a PDF, sometimes as an image, occasionally as a fax-to-email scan that looked like it had been photographed through a screen door.
2. An AP clerk opened the email, downloaded the attachment, and visually matched it to a purchase order in NetSuite.
3. The clerk manually entered every line item — vendor name, invoice number, date, line amounts, tax, total — into NetSuite's AP module.
4. A second person reviewed the entry for accuracy. On a good day, this took five minutes per invoice. On a bad day — when the invoice had twelve line items and the vendor's format had changed — it took twenty.
5. A manager approved invoices over $5,000.
6. At month-end, a third person reconciled the AP subledger against bank statements.

The error rate was the quiet killer. Bellanger's team estimated that roughly 4% of invoices had data-entry errors — a transposed digit, a missed line item, a tax calculation that didn't match. At 4,200 invoices per month, that's 168 errors. Each one took an average of thirty minutes to find and fix during reconciliation.

"We were spending more time finding our own mistakes than we were finding vendor mistakes," Bellanger says. "That's when you know the process is broken."

The moment she decided to build it herself

Bellanger had been through the enterprise software cycle before. She'd sat through demos from three AP automation vendors in 2024. The proposals came back at $8,000 to $14,000 per month, with six-month implementation timelines and dedicated "customer success managers" whose main job seemed to be scheduling follow-up calls.

"I remember looking at one proposal that was $12,000 a month and thinking: I could hire another accountant for that. And the accountant would start producing on day one, not after a six-month implementation."

But she didn't want another accountant. She wanted the work to stop existing.

That conversation was in late December 2025. By the second week of January, Bellanger had signed up for a Lido trial and started mapping her nineteen freight providers' invoice formats.

Building it: the first two weeks

The build happened in two phases. The first was what Bellanger calls "the ugly spreadsheet phase."

She started with three of her highest-volume freight providers — companies that together accounted for about 40% of the monthly invoice volume. She forwarded a week's worth of their invoices to Lido and set up extraction templates for each format.

"The first template took me about two hours," she says. "I had to figure out where each field was on the invoice, map it to our NetSuite fields, and test it with a batch of real invoices. By the third template, I had it down to about twenty minutes."

The extraction accuracy was immediately better than manual entry. Lido's OCR pulled fields at 99.2% accuracy on the first pass. After Bellanger fine-tuned the templates with a few dozen corrections, it climbed to 99.7%.

The second phase was the integration. Bellanger connected Lido to NetSuite using the API connector, which created AP vouchers automatically from the extracted data. She added a Slack notification for any invoice where the extracted total didn't match the PO amount by more than 2% — those went to a human for review.

"The first template took me about two hours. By the third one, I had it down to twenty minutes. By the tenth, I was just copying and tweaking."

— Margaux Bellanger

The part that almost killed it

Three weeks in, the system hit a wall. Four of the nineteen freight providers sent invoices as image-based PDFs — scanned documents with no selectable text. Lido's standard extraction couldn't read them reliably.

"I almost gave up at that point," Bellanger admits. "Those four providers were about 15% of our volume, and if I couldn't automate them, the team would still need to manually process 600 invoices a month. That's enough work to keep one person busy full-time."

She reached out to Lido's support team, who pointed her to their handwriting and image extraction mode. It handled three of the four providers immediately. The fourth — a small regional carrier that appeared to be printing invoices on a dot-matrix printer from 1994 — required a custom template that took an additional day to build.

What the team does now

Six weeks after Bellanger started building, the system was processing all 4,200 monthly invoices. The AP cycle time dropped from eleven days to eighteen hours. Month-end close went from twenty-two days to thirteen.

But the more interesting change was what happened to the three accountants who'd been doing manual entry.

One moved into vendor analytics — tracking spend patterns, negotiating better terms, identifying duplicate charges. In her first month in the new role, she found $34,000 in duplicate payments that had been missed during manual reconciliation.

Another took over cash flow forecasting, building models that used the automated AP data to predict payment timing more accurately. The third became the "automation owner" — monitoring the system, handling exceptions, and building new workflows for other departments.

"Nobody got laid off," Bellanger says. "Nobody even changed desks. They just stopped doing the thing they hated and started doing the thing they were hired to do."

The numbers, six months in

The ROI calculation is straightforward. Before automation, the fully loaded cost of three accountants spending 80% of their time on AP processing was roughly $216,000 per year. The tooling costs — Lido, plus a few API integrations — run about $2,400 per month, or $28,800 annually.

The net savings in labor redeployment alone is roughly $187,000 per year. But Bellanger argues the real value is in the errors they're not making anymore.

"Every AP error has a cost," she says. "The obvious cost is the thirty minutes to find and fix it. The hidden cost is the vendor relationship damage when you pay late, or the duplicate payment you don't catch until the quarterly audit. We've eliminated almost all of that."

The team now catches vendor-side errors — overcharges, duplicate invoices, contract-price discrepancies — that they never had time to look for before. In the first quarter after automation, they identified $127,000 in vendor billing errors.

What she'd tell you over coffee

Bellanger's advice to anyone considering a similar project is blunt: "Stop researching and start building. I spent six months reading vendor comparison articles and sitting through demos before I realized I could just build the thing myself in two weeks."

She's since expanded the automation to cover expense reports (via Ramp) and vendor onboarding documents. The finance team's next project is automating revenue recognition for their restaurant subscription clients.

"The hardest part wasn't the technology," she says. "It was convincing myself that I was allowed to just... do it. Without a project plan. Without a steering committee. Without IT's blessing. Once I got past that, the actual building was the easy part."