The challenge
A mid-size UK logistics company was processing 400+ supplier invoices per week. The process required three operations staff to manually open PDFs, extract line items, cross-reference against purchase orders, and enter data into their ERP system.
The result: 3 FTEs occupied for most of their week, a 12% error rate, and a 5-day payment cycle that was straining supplier relationships.
What we built
We designed and shipped a three-stage automation pipeline:
- Ingestion — invoices arrive by email; an AI agent reads the PDF, extracts structured data (vendor, line items, totals, VAT), and validates against the PO database
- Review layer — flagged mismatches (price variance, missing PO reference) are surfaced in a simple Slack interface for one-click approval or rejection
- ERP sync — approved invoices are automatically posted to their NetSuite instance via API
No humans touch invoices that match. Exceptions get a one-click workflow, not a multi-step manual process.
Results
| Metric | Before | After | |---|---|---| | Manual processing time | ~24 hrs/week | ~1.5 hrs/week | | Error rate | 12% | <1% | | Payment cycle | 5 days | 2 days | | Staff hours freed | 3 FTEs | — |
Timeline
- Week 1: Discovery, API access, sample invoices analysed
- Week 2: AI extraction pipeline + validation logic built
- Week 3: ERP integration, review UI, team training, go-live
What they said
"We didn't believe it would work until we saw it running. Our ops team now handles exceptions in 10 minutes a day instead of spending all week on invoices."
— Operations Director