Collections portfolio allocation and monitoring platform

01

Context and challenge

Fashion’s Park used Excel files with macros and complex formulas to allocate overdue and written-off debt portfolios among external collection agencies. Allocation had to account for percentages based on both customer count and total debt amount, as well as specific rules for each portfolio type.

As data volumes grew, the process became unsustainable. Some operations could take at least a week, fail during execution or freeze users’ computers.

02

My responsibility

I took over the evolution of an existing solution with limited documentation. I worked directly with Risk and Collections users to understand the business rules, reconstruct the processes and turn the solution into a more stable, scalable platform.

As the project’s sole developer, I was involved in technical design, development, API migration, data processing, maintenance and the ongoing evolution of the solution.

03

The solution

I developed a web-based administration interface to configure:

• Customer allocation rules.
• Percentages based on customer count and debt amount.
• Separate rules for overdue and written-off debt portfolios.
• Offers targeting specific groups of debtors.
• Parameters used by automated processes.

I also developed:

• Tables, views, stored procedures, jobs and triggers in SQL Server.
• Python and Pandas scripts for processes that were slow in SQL Server.
• APIs initially built with Node.js and later migrated to Flask to meet architectural standards.
• Daily, monthly and quarterly batch processes.
• CSV file reception and validation over SFTP.
• Decision trees to process collection results.
• Emails covering portfolio allocations, alerts, results and logs.
• Views prepared for subsequent analysis in Power BI.

04

Impact

Daily execution

≈3 minAutomated processes

Usually, within the overnight window

<30 minMonthly and quarterly processes

Continuous evolution

In productionNew releases and business rules

Daily processes were automated and ran in approximately three minutes. Monthly and quarterly processes completed within the overnight processing window, usually in under thirty minutes, with information ready at the start of the working day.

The solution was used in production and continued to evolve through new releases and business rules. It replaced a fragile, manual operation with a configurable, monitorable system designed to handle large volumes of information.