From promise to production: a skills layer in a merged enterprise
Norvane Group, a fictional Belgian manufacturer, bought Keldorf, a German company. Workday on one side, SuccessFactors on the other, Jira, Teams, works councils in two countries. This case covers the implementation end to end and measures every claim it makes.
Take the 2-minute tourRead the caseWeek 2: is the customer ready?
The readiness scan runs before anyone promises a go-live date. It found every planted data problem, with no false alarms. The verdict: not yet. The job architecture needs three weeks of fixes first.
Does a multilingual taxonomy matter?
Recall per language for the three extraction configurations. With canonical English names only, Keldorf's German texts lose most of their skills. That is a fairness problem, not just an accuracy problem.
A = canonical names · B = + aliases in EN/NL/DE/FR · C = + negation, aspiration and ambiguity guards
What should employees be asked to confirm?
Four suggestion policies, measured with simulated employees. The first card decides whether people trust the tool, so the pilot starts with corroborated suggestions (ADR-0009).
| Policy | Right when suggested | Asked / person | Wrong asks / person | Validated / person |
|---|
Do the governance controls hold?
Each control switched off on its own. "Leaks" are opted-out people with evidence or exports, quarantined text used as evidence, and unvalidated skills sent downstream.
The landscape
Pull from every source on our schedule. Push only validated skills, to exactly two places. Governance is applied before inference, not after.