After years of internal developing a tool for repeated issues to solve and then finally deciding to acutally make it available to others was a first step. Now finding your first customer is a huge hurdle. Actually implementing the solution for them, especially when tackling a complex problem, is often an even bigger challenge. We've just crossed that bridge.
After intense collaboration with our first customer, we've successfully launched our first comprehensive automation flow for processing and labelling invoices before handing them over to the accounting system. This isn't a simple script; it's an in-depth system designed to solve a genuinely intricate data processing task.
Here's a detailed look at the workflow we've built:
The process starts with User Input, where we defined the input and fields to receive before handing it over to a complex LLM flow. From there, the system branches into several parallel validations and extractions:
Contact Extraction: Pulls relevant contact information directly from the source data.
Data Completeness Check: Uses predefined Data Completeness Guidelines to verify all required fields are present and valid.
Cost Center Guidelines: A set of rules that feed into the data extraction and matching process.
The core processing begins with Extract Invoice Data, which takes the validated input and pulls out the necessary details. This data then flows into two critical matching systems:
Match Cost Centers: Applies the Cost Center Guidelines to accurately categorize expenses.
Iterate over Invoice Line Items: This is where the complexity deepens. For each individual line item on the invoice, the system performs a nested operation:
It references the Chart of Account Matching Guide.
Uses the Chart of Accounts to find the correct accounting classification.
Applies VAT Tax Authority Matching Guidelines to assign the proper code.
Finally, the results from the Data Completeness Check, Match Cost Centers, and the processed line items are all consolidated into a single, structured Output. It's a robust, multi-step flow that transforms raw, messy data into reliable, actionable information.
Building this together with our first customer has been an incredible learning experience, and seeing it live is a major milestone.