Strict CSV templates, manual categorisation and PDF-only systems
The firm's supply chain team oversees procurement for restaurant and food service portfolio companies operating thousands of sites across Europe, covering around a billion in spend. Portfolio companies had to upload strictly formatted CSV files, which were then categorised and re-uploaded by hand before reaching Power BI. Distributor data filled the gaps, and one legacy Oracle system could only show orders as PDFs.
A platform that takes data as it comes
We replaced the old solution with a metadata-driven Microsoft Fabric platform. Data lands from any portfolio company in any format, is standardised to one product taxonomy and unit of measure, and is served to Power BI and AI tools. One control table decides what is loaded from whom.
Nothing rejected for format
Bronze keeps every CSV, JSON, extract or PDF exactly as received, and processed files are archived so data never duplicates.
One taxonomy, one unit of measure
Silver maps every product to a standard taxonomy and unit of measure, covering cases, units and each sizes, with volumes, pricing, suppliers and country of origin.
Onboard by configuration
A new portfolio company is added with rows in the control table, not new pipelines. Companies on the same ERP share one parameterised pattern.
Offboard in under half a day
After an exit, a company's sources are switched off and its access removed, with its history kept or deleted as the firm requires.
Each brand sees only its own data
Row-level security in the Direct Lake model keeps portfolio companies separate, while the firm compares spend like for like across all of them.
AI where the data is messy
AI reads PDFs and messy files and suggests product categories, asking the team to confirm when it is not sure.
Where the data lives and how it flows
REST and OData APIs or database replication, wherever a portfolio company's ERP allows it
What changed
The supply chain team now sees procurement spend across the portfolio on one standard basis, without forcing every company into a rigid file template or categorising data by hand.
As the portfolio changes, companies are added and removed through configuration, so the platform keeps pace with acquisitions and exits without being rebuilt.