AI-Assisted Bookkeeping Tooling with Bexio Integration
02/2026 – 04/2026
MCP-based bookkeeping assistant: CSV import of bank transactions, AI-generated journal entry suggestions, automated receipt number generation and structured document filing in Bexio.
1. Background
Manually posting bank transactions in Bexio was time-consuming: account statements had to be reviewed manually, appropriate journal entries derived from the chart of accounts and documents assigned one by one. The goal was tooling that supports this process in a structured way, without removing control over the actual posting from the user.
2. Stakeholders
3. Challenges
- Bank transactions are provided as CSV and must be interpreted, categorised and mapped to the correct accounts
- Relevant journal entries and chart-of-accounts logic are complex and context-dependent — difficult to map correctly with a generic AI model without customisation
- Receipts (invoices, vouchers) must be reliably assigned to each posting line and archived with a consistent receipt number
- The process must keep control with the user — the AI suggests, the human decides and posts
4. Analysis
The process was structured into two phases: preparation and execution. In the preparation phase, the user uploads the CSV bank statement and associated receipts. The MCP server reads the transactions, pulls existing posting data and account information via the Bexio API, and passes the full context to the AI model. The model knows the relevant journal entries and suggests a complete posting for each transaction — including the contra account, tax category and assignment logic. In the execution phase, the user reviews the suggestions, manually posts in Bexio and uploads the receipt. The tool generates a consistent receipt number and links the document to each affected posting line.
5. Solution Options
6. Evaluation
- ✓Significant time savings when posting: journal entry suggestions substantially reduce the manual research effort per transaction.
- ✓Consistent document filing: uniform receipt numbers and structured assignment eliminate unstructured storage in email inboxes or local folders.
- ✓Control stays with the user: no automatic writes to Bexio — every posting is consciously approved and executed.
- The quality of posting suggestions depends on the completeness of the provided context — ambiguous posting texts or new accounts require manual intervention.
- The Bexio API integration is read-only: the tool does not automate posting, so the user must execute every step in Bexio manually.
- Incorrect posting suggestions can be adopted unnoticed if users do not review them carefully — especially with similar transaction patterns.
- Data privacy: financial data and documents are passed to an AI model — a data protection concept and careful model selection are security-critical.
Concept and initial setup: approx. 3–5 weeks; ongoing maintenance of journal entry context: continuous
7. Recommendation
Decision
GoPriority
High – immediate efficiency gain on recurring, manual posting effort
Next Steps
- Systematically capture the journal entry library and structure it as context for the model.
- Define and document a data protection concept for passing financial data to external models.
- Set up a feedback loop: systematically capture incorrect suggestions and use them to improve the context.
Ready to collaborate?
I am looking for a permanent position as IT Delivery Lead / IT Project Manager.