Case Studies
Real problems. Real solutions. Results you can take to a board meeting.
Demo case study built on fully synthetic data. "Bottega Nord S.r.l." is a fictional client — every document was generated for the demo. The pipeline, and the problems it solves, are entirely real.
Bottega Nord, a specialty food distributor in Lombardy, receives ~800 supplier order confirmations a month — PDF invoices in three different layouts, Excel exports that never match, semicolon CSVs with Italian decimal commas, and orders typed straight into emails. Every Friday, ~6 hours disappeared into manual reconciliation. Duplicate invoices still got paid. Price increases still went unnoticed.
The Pain
- ~800 supplier confirmations/month across 4 incompatible formats
- ~6 hours of manual reconciliation every Friday
- Duplicate invoices still slipped through and got paid
- Silent price increases went unnoticed
The Build
- Five-stage Python pipeline, built with AI-assisted development (Claude Code)
- Ingestion → extraction → normalization → anomaly detection → reporting
- One command, no manual steps
- Fuzzy-matches supplier & product names against a reference catalog
The Numbers
The Kicker
The data is synthetic. The pipeline isn't. If your team runs on supplier paperwork, spreadsheets nobody trusts, or an inbox that eats a day a week — this same architecture drops onto your data. Let's talk →
An inherited Excel model covering 10 EU markets had become unmaintainable — riddled with hardcoded values, broken as soon as a country changed a rate or assumption, and impossible to audit. Finance, Operations, and Country Leads were all working from different versions of the truth. The business needed a single, governed source of analytical truth that could answer utilisation and revenue questions across an entire European network — in real time.
The Problem
- 10-country model with hundreds of hardcoded formula cells
- No single version of truth — teams used conflicting spreadsheets
- Manual refresh cycle: hours of work per market update
- No auditability or change tracking — high risk in regulated environment
What I Built
- Full star-schema data model on Azure Databricks
- Zero hardcoded formulas — all assumptions governed as parameters
- Power BI front-end delivering live cross-country views
- Automated ETL pipelines replacing manual data collection
Outcomes & Impact
Approach
- Discovery: mapped all 10 country models to identify shared vs. market-specific logic
- Architecture: designed star-schema (fact + dimension tables) to handle multi-country complexity cleanly
- Build: developed on Azure Databricks with parameterised assumption tables instead of hardcoded cells
- Reporting layer: Power BI dashboards with cross-country and drill-down views
- Governance: version control, documentation, and stakeholder training across 4+ country teams
SWARCO's Finance and Operations leadership needed visibility they couldn't get from their existing reports — scattered across systems, too slow to generate, and not designed for decision-making. On top of that, strategic investment decisions on EV charging infrastructure required a rigorous analytical framework that could survive board-level scrutiny. I was brought in to build both: the operational reporting layer and the investment analysis that went to the board.
The Problem
- Finance & Ops reporting fragmented across multiple source systems
- No self-service BI — leadership dependent on manual data pulls
- Investment decisions made without a structured analytical framework
- Board needed a clear, defensible view of infrastructure ROI
What I Built
- Power BI dashboards for Finance leadership — P&L, cost tracking, KPIs
- Operations dashboard — utilisation, deployment status, performance trends
- Investment analysis model evaluating EV infrastructure opportunities
- Board-ready outputs: scenario modelling, risk view, recommendation layer
Outcomes & Impact
Approach
- Stakeholder discovery: workshops with Finance and Ops leads to map reporting needs and decision workflows
- Data architecture: connected disparate source systems into a clean Power BI data model
- Dashboard design: built for decision-making — not just data display. KPIs surfaced at the top, drill-through for detail
- Investment model: structured ROI framework with scenario sensitivity analysis and risk flags
- Board delivery: translated analytical findings into clear executive narrative with recommendation layer