Invoice intake cut from 6 hours/day to 20 minutes.
Automated document collection, OCR extraction, and reminder flow for a mid-size accounting firm.
The state we found.
The firm handled roughly 800 client invoices per week. Each one was received by email, downloaded by hand, renamed, filed in OneDrive by client and period, and entered into the accounting system manually.
Two junior accountants spent most of their mornings on this instead of reconciliation work. Clients who didn't send documents on time had to be chased manually — and often weren't, because no one had the time.
The owner could not hire fast enough to keep up with new clients without sacrificing margin.
What we built.
We built a Make-orchestrated pipeline that listens to the shared inbox, extracts invoice data via OpenAI Vision, files PDFs with a consistent naming convention, writes the structured data into the accounting system, and sends automated reminders to clients who miss the monthly deadline.
- 01IMAP listener on shared inbox, filtered by recognized client sender domains
- 02Attachment routing to OneDrive with deterministic naming: {client}_{period}_{doctype}.pdf
- 03OpenAI Vision extraction of invoice number, date, VAT, totals, counterparty
- 04Validation rules with confidence threshold — anything under 95% goes to a human review queue
- 05Direct API write to the accounting system with idempotency keys
- 06Monthly cron job: detect missing documents per client, send tiered reminders on day 3, 7, and 14
€40k+ recovered in 90 days. No manual follow-ups written.
Book the call. Bring the bottleneck.
30 minutes. No deck. We either see the automation in your process — or we tell you it is not worth building.