Three engagements. One playbook: automate the predictable, review the edges, ship the artifacts. Invoice ingestion, reconciliation and the monthly close cycle — all cut down by an order of magnitude.
Every study is chart-heavy and stands on its own. Click through for the full breakdown of what changed and how we built it.
40,000 invoices a month, 200+ supplier formats, six clerks. We replaced manual keying with an LLM-in-the-loop pipeline — same team, 20× throughput.
2,400 SKUs · five tender types · 180K daily transactions. We took auto-match from 68% to 99.9% and cut break volume by 82×.
We made the monthly close a nightly job. Cycle time down 91%, reporting TAT down from 15 days to overnight.
A third of invoices arrived unreadable — phone photos with keystone distortion, faded thermal prints, scanner shadows. A vision agent picks a bespoke recipe per page, recovering 88% of what OCR previously rejected.
Plotted on a single axis: the before-and-after productivity multipliers for each workstream. Same team. Same book of business.
"In under a year the operating cadence of the finance function changed. Faster ingestion drove better reconciliation, cleaner recon drove a faster close, and the close drove real-time reporting. Each win compounded the next."
The same shape every time: automate what's predictable, put a human on the edges, ship the artifacts to the client. Nothing fancy — just done properly.
A week of time-and-motion. Where the hours actually go, not where anyone thinks they go.
Rules for the deterministic, models for the fuzzy. Cover the predictable 80% first.
Confidence-gated review. Corrections train the next round. The pipeline gets sharper every week.
The pipeline, the model, the console — all handed over. When we leave, they keep the leverage.
Two-week paid pilot. We map the pain, ship the first slice, and hand over the artifacts. You keep the leverage from day one.