Data Warehouse & Power BI — Acecook Vietnam
A ~100-table SQL Server warehouse holding 300 million records of sales-out data, fed by 80 SSIS ETL flows from SAP HANA, Oracle and Excel, with SSAS models serving Power BI.
What I delivered
- Surveyed, designed and implemented a Data Warehouse for Acecook Vietnam integrating SAP HANA, Oracle and Excel sources into MS SQL Server.
- Built 80 SSIS ETL pipelines orchestrated with SQL Server Agent jobs; ~100 tables, ~300 million records.
- Designed and deployed sales-out data models using SSAS as governed sources for Power BI reporting and analytics.
- Led a team of four across the design and delivery.
Context
Acecook Vietnam is the country's largest instant-noodle manufacturer. Their sales-out data — what actually moved from distributors to retail — lived in three different places: SAP HANA, an Oracle system, and a substantial number of Excel files maintained by regional teams. Every management report was somebody reconciling those three by hand.
The goal was a single warehouse those reports could stand on.
The build
Survey first. Before any modelling, we catalogued every source: what it contained, who owned it, how often it changed, and how much of it could be trusted. The Excel files were the interesting part — they encoded business rules that existed nowhere else, and those rules had to be reverse-engineered before they could be automated.
The warehouse. Roughly 100 tables on MS SQL Server, about 300 million records. Layered: staging that mirrors source, a cleansed layer where the rules live, and a presentation layer modelled dimensionally for analysis rather than for the shape of any source system.
The pipelines. 80 SSIS packages, scheduled by SQL Server Agent. Incremental where the source supported a watermark; full refresh where it did not. Each package logs row counts and duration, so a slow load is visible before it becomes a failed load.
The semantic layer. SSAS models over the presentation layer. This is what stops "sales-out" meaning three different numbers in three different Power BI reports — the measure is defined once, in the model.
What I would do differently
Build the reconciliation into the pipeline from day one. We validated against source during UAT and found discrepancies late. Row-count and sum checks that run automatically after each load would have surfaced those in week two instead of month four.
Push harder on source data ownership early. The Excel files had no owner, which meant every question about a rule took days to answer. Naming an owner per source at kickoff is a five-minute conversation that saves weeks.
Result
Management reporting moved from manual consolidation to a governed warehouse with Power BI on top. For me it was the project that turned "I am an ABAP developer who is learning data" into a second discipline I could lead a team in.
To expand: the dimensional model, the incremental load pattern, and the SSAS measure design.
Want the detail behind this?
Happy to walk through the technical decisions in a conversation.