Case studies

Numbers measured in real projects. Client names are kept private; when a number comes from a simulation, we say so.

Industrial lakehouse

Context: manufacturing. Challenge: slow reports, scattered data duplicated across layers, and heavy queries running directly on the ERP and other sources. What we did: designed and deployed a lakehouse with the operation's sources and BI models. Then an audit compared what was ingested with what was actually used.

253tables organized

44business areas

631measures in 4 BI models

10unused tables, ready to switch off

For the business: the operation now has its data in one place, organized by area, and knows exactly what it can stop ingesting.

Data platform FinOps

Context: a mid-sized company with a rising data platform bill. Challenge: understand what was making the bill grow. What we did: cost assessment by layer and by usage type, review of billing lines and scenario simulation.

−80%on one cloud protection line, by changing the billing model

−39%cost in the per-layer simulation

~76%of serverless SQL was consumed by BI alone

For the business: cuts prioritized with a known value before touching anything, and the finding that BI was the largest serverless SQL consumer.

Software with an architecture guardrail

Context: manufacturing, an evolving internal application. Challenge: keep the architecture in order while the system grew. What we did: automated architecture tests that block forbidden dependencies between modules, and fixed the violations found.

35 → 0architecture violations

218fields renamed in a single PR, with nothing broken

For the business: large changes made safely, with the architecture protected on every new delivery.