Excel
Too many competing final versions.
Customers / Liantis
One word, four different headcounts.
Liantis merged three companies into one in 2018, and inherited three vocabularies. The word department alone meant four different things across their systems, and meetings routinely opened with people disputing whose headcount was right instead of deciding what to do about it.
- HR & payroll services for entrepreneurs and employers, Belgium
- 100+ years in business
- ~2,000 employees
- 40+ offices
- ~246,000 self-employed customers, ~76,000 employer customers
- 3,200 Power BI reports
- 32 business domains
01The problem
One word, four owners.
After merging RDMB, Profitmo and Zenito in 2018, Liantis inherited three separate vocabularies and application stacks. Definitions drifted by department and by system. The same label could produce four different headcounts depending on which meaning a report used.
Function
The legally required function and role classification
Legal entity
A unit inside the legal company structure
Employee group
The split between blue-collar and white-collar employees
Location
The physical office where someone works
02The false start
Three tools, at once.
Liantis first spread the glossary across Excel, Confluence and SharePoint simultaneously. Each one blocked a different group from getting to the same definition.
Confluence
Good documentation, but licensed to developers only.
SharePoint
A permissions request could take three weeks to clear.
03Why dScribe
One place people could actually use.
Liantis replaced all three tools with dScribe against three practical requirements.
- 01One place both business and technical users could actually reach.
- 02Harmonized definitions across service lines.
- 03Low enough friction that non-technical people would use it without training.
It was available to everyone through a link, included SSO, and did not require a lengthy implementation project.
04The second false start
A place with no rules.
Early success became its own problem. Because dScribe was so easy to use, people wanted to put everything into it: API documentation, event schemas and reference data, with no agreement on what belonged there.
Governance still sat inside IT as a service, not inside the business as a capability. A data steward could tell a business team how to work, but had no mandate behind that instruction.
05The reboot
Governance moves to the business.
About eighteen months in, Liantis ran a strategic reset. Data governance moved from IT to the business, and Bart T. joined as Data Architect & Data Governance Lead specifically to fix it.
Start with the term, then reach the report.
Reports joined the catalog, not just definitions. The Power BI connector auto-scrapes every workspace into dScribe, so Liantis flipped the workflow: search by term first, jump to the report second.
Put one named owner behind every domain.
Liantis adopted domain-driven design across 32 domains and realigned the data warehouse to match. Every definition links to a domain, and every domain has a named owner.
06Databricks and AI
One definition, on both sides of the connector.
Unity Catalog
Schemas, tables and fields enter dScribe through the Databricks connector.
dScribe
Governed definitions read back into Unity Catalog for Liantis's own AI work.
Liantis packages data products per domain, linking each table and field to the governed definition that explains it. The connection works both ways, so the same business meaning is available where people document data and where technical teams build with it.
07Data quality
Measured, so it becomes visible.
Liantis wanted technical checks and contextual accuracy: rules only a domain owner could define. dScribe's data-quality product was co-developed with Liantis as an early design partner.
Three approaches rejected
Soda
A business-friendly interface that priced out a small data team.
Great Expectations
Python on pipelines, usable by engineers only.
Build in Databricks
Full control, with all of the maintenance left to Liantis.
08Four people, four views
Each person sees the decision they need.
Business user
Sees data health directly in dScribe and gets emailed when a rule is violated.
Data steward
Gets an operational dashboard and can export failed rows as CSV.
Management
Sees the trend without the row-level noise.
Business expert
Changes rules directly, without opening an IT ticket.
09In Bart's words
“We needed the product to measure our data quality, because when it's measured, it's visible. And when it's visible, somebody might become embarrassed for the data quality and might actually do something about it.”
See what a governed glossary looks like at your scale.
Twenty minutes, your own definitions, your own disputes settled.