Context and problem

• Employees manually processed high volumes of receipts and invoices
• Documents came in dozens of inconsistent formats
• Staff had low to average computer proficiency
• Work was fragmented across folders and spreadsheets

The goal was not “make a prettier tool” rather to reduce processing time and errors, and simplify daily workflows for the accounting staff..

Constrains

Low to Average Digital Literacy – Most employees were not power users. The system had to minimize abstraction, reduce navigation complexity, and avoid advanced UI patterns.

Client-Defined Financial Structure (Non-Negotiable) – The folder structure (Q/E, Y/E period logic) reflected established accounting workflows and could not be redesigned. The system had to adapt to their mental model.

Operational Priority Over Administrative Features – Daily task execution (processing documents) had to take precedence over secondary features like employee, supplier, and client administration.

Deep Archive Navigation Risk – Large volumes of financial periods created long scroll states and potential orientation loss. Navigation needed recovery shortcuts.

Accuracy Sensitivity (Financial Domain) – Small errors could have compliance or reporting consequences. Visual clarity and confirmation states were critical.

Trust Barrier Toward Automation – Employees were skeptical of machine learning outputs. The interface needed clear feedback states, manual review with editable fields, and visible processing status to build confidence.

High Variability in Document Formats – Receipts and invoices came in inconsistent layouts, requiring a flexible ML parsing system and UX that allowed correction without friction.