The Hidden Cost of Data Debt
summary
Claim: unnoticed data debt silently degrades pipelines and product metrics, sometimes causing catastrophic losses (460 million lost in 45 minutes cited as a caution); concrete failures include orphan dashboards, orphan feature flags, schema changes breaking downstream consumers, training‑serving skew and feature bugs that shift model accuracy from 90% to 80%. Remedies: lineage, ownership, on‑call SLAs, distribution tracking, feature store with versioning, golden datasets, reverse A/B tests; experiments that restart add 50% slower time to market.
AI-written summary · about
chapters (16)
- 0:00Data debt introduction
- 3:16Tech vs data debt
- 6:29Data debt in systems
- 9:43Healthy data principles
- 10:48System-level data debt
- 13:16Data debt and experiments
- 18:38Experimentation best practices
- 19:41Experiments velocity pitfalls
- 22:58ML data debt challenges
- 27:34ML data quality practices
- 29:45Building data-first culture
- 31:56Q&A
- 33:23Q&A
- 35:08Q&A
- 36:55Q&A
- 37:56Q&A
related talks
52mScaling Data Teams: Three Pitfalls and How To Avoid Them
25mWhy Software Engineering Best Practices Fail in Data Engineering
24mA Practical Approach To Unlock Value From Data and Analytics
31mThe Hidden Costs of Data Quality - Tackling Common Data Challenges in ML
38mData Secrets from a Platform Engineer
44mKeep it Clean: Why Bad Data Ruins Projects and How to Fix it
27mThe Cost of a Bad Code and Pipelines
49mData Cleansing & Beyond: How to Address Data Debt for AI