
Development of an intelligent matching engine to reduce the volume of manual lettering by 50% to 15%
Context
At the annual closing of accounts, ISTA must reconcile thousands of customer payments (over €40 million in three months) with their invoices. This lettering process is an essential link in managing cash flow and accounting reliability. Until now 50% manual, it strongly mobilized teams and required the hiring and training of two temporary workers each year. The peak in activity weighed heavily on teams, who had to pause other tasks with higher added value. Internal attempts via SAP and Excel had already shown the potential for automation, without allowing business complexity to be addressed on a large scale. To respond to this, ISTA launched an automation project with Theodo, in order to make the process more reliable and reduce human effort on this repetitive task.

Challenges
- Reduce to 15% the cases of manual and semi-automatic lettering, which until now represented 50% of cases.
- Process heterogeneous source data (PDF, Excel, emails, emails, SAP systems, Eloficash) with structures that are sometimes incomplete or not aligned.
- Develop an intelligent matching logic capable of managing complex management cases (partial payment, multi-invoices, absence of transfer notice).
Our approach
A dedicated agile team composed of 2 Theodo AI experts, supported occasionally by a DevOps expert for Azure deployment via Terraform.
Just-in-Time Delivery
A modular architecture with Azure Document Intelligence and integration with existing lettering flows allowed the construction of an application in production in 6 weeks.
Right First Time
In-depth framing workshops with professionals to prioritize treatable cases from the first version.
Product co-construction
With ISTA to define the use cases, critical business rules and the structure of the output files required for SAP import.

