Sebastian de GoeijShort answer
AI learns to book from four sources: the administration's history, your team's corrections, the firm's rules and plain-language instructions. From these it derives a booking proposal with a confidence score per invoice. The more stable the history and consistent the corrections, the higher the confidence and the more can be automated.
Key takeaways
- History in the accounting software is the main learning source.
- Every correction is kept and improves later proposals.
- Rules and instructions capture agreements explicitly.
- New administrations without history start with lower confidence.
Does the AI learn from my accounting software's history?
Yes. Master data and past bookings show which GL account, VAT code and cost centre are usual per supplier.
What happens to my team's corrections?
They are kept in the feedback history, used per administration and firm-wide, and can be applied to similar bookings.
Rules versus instructions
Rules are firm agreements; instructions are plain-language agreements through the AI assistant, with 139 examples in the prompt library.
| Source | What it captures | Scope |
|---|---|---|
| History | How it was booked before | Per administration |
| Corrections | What your team wants differently | Per administration and firm-wide |
| Administration rule | Agreement for one client | One administration |
| Global rule | The firm's way of working | All administrations |
| Instruction | Agreement in plain language | One administration or firm-wide |
What about a new client without history?
The AI uses firm-wide patterns, master data and your own context such as industry. Start such an administration on suggestions.
FAQ
Questions about this topic.
Yes. Corrections are kept and can be applied to similar bookings.
Client data is not used to train public AI models.
How Autoboeker solves it
