Sebastian de GoeijShort answer
Automate invoice processing by first centralising delivery, then leaving recognition and coding to AI, and finally setting a confidence threshold per administration above which invoices are booked automatically. Start with suggestions, capture recurring agreements as rules and raise automation step by step, measuring the automation rate per administration.
Key takeaways
- Automation starts with delivery: one inbox per administration.
- Choose recognition without templates or the work shifts to maintenance.
- Let AI code from history; only capture explicit agreements as rules.
- Use a confidence threshold per administration and raise it gradually.
- Measure automation rate, lead time and open client questions.
Why automate invoice processing?
It is the largest block of recurring work in most firms. Indicatively, going from ±6 minutes of manual work to ±1 minute of review per invoice saves about 667 hours on 10,000 invoices a year, and makes processing consistent regardless of who books.
Step 1: centralise delivery
No tool can process invoices scattered across personal mailboxes and shoeboxes. Give every administration one inbox fed by a purchase and sales email address, a client app, Gmail, Outlook, IMAP, cloud storage, WhatsApp, the Exact digital mailbox or SnelStart documents.
Step 2 and 3: recognition and coding
Choose recognition without templates, then let AI decide GL account, VAT code, cost centre and contact from the administration's history. Use administration rules, global rules and plain-language instructions only for explicit agreements.
Step 4 and 5: checks and the confidence threshold
Make sure deviations are flagged (VAT, VIES, IBAN, duplicates, high amounts, unknown suppliers) and set the threshold per administration.
- 1Start every administration on suggestions.
- 2Check after a few weeks which scores hold up in practice.
- 3Move stable administrations to smart confirm.
- 4Turn on fully automatic booking above a suitable threshold.
- 5Keep approvals before or after booking where the client wants them.
Step 6 and 7: close the loop and measure
Connect bank matching and client questions in the same flow, then track the KPIs below.
| KPI | What it tells you | How to steer |
|---|---|---|
| Automation rate | Share booked without manual work | Adjust threshold and rules per administration |
| Correction rate | How often proposals are changed | Capture agreements as rules or instructions |
| Lead time | Arrival to booking | Speed up delivery and review |
| Open client questions | Information still with clients | Automatic follow-up and client portal |
FAQ
Questions about this topic.
Centralise delivery per administration, connect your accounting software and let the AI propose for a few weeks before raising automation.
Indicatively from ±6 minutes to ±1 minute per invoice; about 667 hours on 10,000 invoices a year.
Not with AI recognition without templates.
Yes, threshold and mode are set per administration.
How Autoboeker solves it
