Logo Lanfrica

Ntsikelelo-N/finadmin360

Domain:

socioeconomic

Record type:

project
Creator:
Nts
Host:
Transforming a South African finance administrator's manual spreadsheet workflow into an automated, insight-driven data platform — end to end. # FinAdmin360 — Finance Intelligence Platform > Transforming a South African finance administrator's manual spreadsheet workflow > into an automated, insight-driven data platform — built end to end on Azure. --- ## 1. The problem A South African finance administrator was managing supplier invoices, bank reconciliations, VAT submissions, and CIPC returns entirely through Excel spreadsheets. This created four concrete risks: 15+ hours per month lost to manual copy-paste workflows, no early warning system for late-paying suppliers, manual SARS VAT201 preparation with no automated validation on R25M+ of annual input tax, and zero real-time visibility for the CFO into cash flow, overdue balances, or supplier performance. FinAdmin360 replaces that workflow with an automated pipeline that ingests raw invoice and bank data, validates and transforms it through a Medallion architecture, scores open invoices for late-payment risk with a machine learning model, and serves the result through a daily-refreshed Power BI dashboard. ## 2. Results | Outcome | Result | |---|---| | Total spend tracked (24 months) | R193,146,040 across 2,000 invoices | | VAT input tax validated | R25,192,962 — 0 mismatches detected | | Late payment rate identified | 19.9% (398 of 2,000 invoices) | | Late payment value exposure | R46,503,680 | | Currently overdue | 271 invoices, R73.6M outstanding | | dbt schema and data quality tests | PASS=12, ERROR=0 | | ML model (GradientBoosting) | Test AUC 0.610, selected over RandomForest (AUC 0.647) for usable recall on the Late class | | Airflow pipeline | 6 tasks, daily run at 06:00 SAST, completes in under 12 minutes | | Models compared in MLflow | 3 (GradientBoosting, RandomForest, LogisticRegression) | ## 3. Architecture Bronze ingestion lands raw CSV files in ADLS Gen2 via Azure Data Factory. Synapse Serverless SQL exposes the Bronze files as external tables. dbt Core transforms Bronze into validated Silver views and business-ready Gold KPI …