Alongside the Bärli Tax App, Bode Information Corporation is developing SwiftTaxDesk as the platform for professional tax advisors. While clients submit documents digitally through the app, the advisor workflow remains largely manual. Advisors read every document by hand, copy data into cantonal tax forms, and identify missing files only after reviewing each case individually. As client volume grows, advisor hours grow at the same rate. No automated layer exists between document submission and advisor involvement. This thesis addresses the need for that layer.

Following design science research methodology, an AS-IS analysis identified seven pain points and derived functional requirements for the automated solution. A technology evaluation across 13 criteria shortlisted three AI extraction tools: Azure Document Intelligence, Mistral OCR, and gpt-5.4-mini. A prototype was tested across 22 runs on five real and synthetic Swiss tax documents within the IT infrastructure of Bode Information Corporation. A TO-BE process model was developed for SwiftTaxDesk, defining how automated extraction connects document submission to structured advisor review.

The evaluation produced 19 documented findings. The central result is that training data representativeness is the decisive factor in AI extraction performance. A model trained on a simplified synthetic wage statement achieved 0% effective accuracy on the real official Swiss Form 11. After retraining on the correct official layout, the same model extracted all fields correctly from a genuinely unseen real client document. The failure was in the training data, not the technology, and it is fully resolvable. For bank account tax documents, the Azure Document Intelligence general model achieved 100% accuracy without custom training. Azure Document Intelligence in the Switzerland North region is the only tested tool that satisfies Swiss nDSG data residency requirements and is, therefore, the recommended tool for live client data. Based on these findings, the thesis delivers a six-requirement pipeline specification and a TO-BE process design for SwiftTaxDesk. Bode Information Corporation receives a production-ready roadmap covering tool selection, model training, quality gating, and field validation.