ABOUT THIS EPISODE
One of Clouds on Mars's clients had over 5,000 business rules to validate incoming data. Bad data kept coming in anyway. The rules were checking format. Nobody was checking context.
This episode covers GODI - Governed Operational Data Improvement - a framework built to catch data errors that pass technical validation but are wrong in the context of how a business actually operates.
Marcin Kubica and Aleksandra Gepert walk through a real case: a procurement team entering a 300-day lead time for a component that always arrives in 14 days. The system passed it. Planning triggered emergency orders. Production was at risk. GODI is built to catch exactly this kind of error - early, before it cascades.
You'll take away:
- Why structural validation rules are never complete - and why they become outdated as business changes
- How GODI's four layers work: from deterministic business rules, through historical context, to category comparison and closed feedback loop
- What three levels of user response (accept, correct, escalate) feed back into the ML model and make it sharper over time
- Why organizational readiness matters more than technology in an implementation like this
- What 90% recommendation accuracy and 3% false positive rate looks like in practice
Key topics covered:
- The difference between format-correct data and contextually correct data
- How a single data entry error (300 days vs. 14 days lead time) can trigger emergency shipments and production holds
- GODI's four-layer detection model: business rules, historical context, category comparison, learning loop
- Why escalation cases - where the operator isn't sure - are the strongest learning signal for the model
- Post-mortem culture vs. built-in learning: why lessons from incidents rarely make it back into the system
Data quality tools tend to focus on rules. GODI focuses on judgment. The difference is that rules catch what someone anticipated in advance. GODI catches what didn't fit the pattern - even when no rule was written for it. Over time, with human feedback at every step, the system gets better at distinguishing what's genuinely wrong from what just looks unusual. That's a different kind of reliability than any static rule set can provide.
Watch full episode:https://www.youtube.com/watch?v=NZbJr-jTK-o&list=PL053kgqQv_fspiVbwCStY8rTcjldz4DtO&index=5 Website: https://cloudsonmars.com Follow the host: https://www.linkedin.com/in/marcinkubicaquality/
