Coming next: The Second Factory →

Back to Library
Data Foundations⚠️ Pitfall

Bronze, Silver, Gold Means Nothing (When Layer Names Don't Help)

The Problem

Your data lake has three layers: Bronze (raw data), Silver (cleaned data), Gold (business-ready data). Sounds sophisticated. Then new analyst joins: "Where's customer data?" Answer: "Check Gold layer." They search. Find 14 tables with "customer" in the name. Which one is right? No idea. They ask 3 people, get 3 different answers. Turns out the "right" customer table is actually in Silver because Gold was never finished. Bronze/Silver/Gold tells you nothing about what's inside. It's like Home Depot organizing by "Bronze aisle," "Silver aisle," "Gold aisle" instead of "Paint," "Tools," "Lumber." Users need to understand what data IS, not what processing tier it's in. Metal names are architecture porn—they make engineers feel organized but leave users lost.

The Principle

Name layers by what they contain, not processing stage. Use names that self-document: Raw (exactly as source systems sent it), Curated_SystemName (cleaned, usable, still system-specific), Integrated_TopicName (cross-system, business-ready). Examples: Raw_Salesforce, Curated_Claims, Integrated_Customer. Every name tells you what's inside. The magic happens when a new analyst can guess where data lives without asking anyone. Think retail: you find paint in the Paint section, tools in Tools, lumber in Lumber. Not in Bronze/Silver/Gold sections. Same with data—name by content, not by processing tier.

🎯

3 Steps to Take Action

  1. 1

    **Audit Your Current Naming** List all your data layer/zone names: - What are they called? - Can a new user guess what's in each layer from the name? Test: Ask 5 analysts: "Where would you look for customer data?" If they can't answer without documentation, your names failed. Common bad patterns: - Bronze/Silver/Gold (processing stage, not content) - Stage1/Stage2/Stage3 (meaningless) - Raw/Cleaned/Final (still doesn't say what's inside)

  2. 2

    **Rename Using Content-Based Pattern** Adopt naming convention: - **Raw_SystemName**: Data exactly as source sent it (Raw_Salesforce, Raw_SAP) - **Curated_Domain**: Cleaned, usable, system-specific (Curated_Claims, Curated_Calls) - **Integrated_Topic**: Cross-system, business-ready (Integrated_Customer, Integrated_Product) Every name answers: "What's in here?" Create migration plan: Start with most-used tables, rename gradually. Update all docs, queries, dashboards to use new names.

  3. 3

    **Self-Documenting Standards** Create naming standards guide: - Layer prefix: Raw/Curated/Integrated - Domain or system name - Optional: granularity (Daily, Summary, etc.) Example: Integrated_Customer_Daily vs Integrated_Customer_Summary Test with new hires: Can they find data without help? Goal: 80%+ of "where's X data?" questions answered by naming alone. Bonus: Every new table must follow naming standard or gets rejected in code review.

💡 Inspired By

Home Depot/library organization analogy from Data Strategy 2023. The problem with abstract layer names that don't communicate content.

This is what Analytics Pitfalls & Principles is about: people helping people.

In Practice

A financial services firm had Bronze/Silver/Gold lake architecture. Beautiful in PowerPoint. Nightmare in practice. New data scientists spent first 2 weeks just learning "which layer has what." Onboarding took 6 weeks because nobody could find data. They renamed everything: Bronze → Raw_SourceSystemName (Raw_Salesforce, Raw_Mainframe, etc.), Silver → Curated_Domain (Curated_Loans, Curated_Customers), Gold → Integrated_BusinessArea (Integrated_Lending, Integrated_Risk). Every name self-documented. New analyst question: "Where's loan data?" Answer obvious: Integrated_Lending. Onboarding dropped from 6 weeks to 1 week. Support tickets about "where's X data" dropped 71%. Bonus: Documentation became less critical because names explained themselves. The architecture didn't change—just the names. But user experience transformed overnight.

More on Data Foundations

From the book

In the book, a new hire takes six months to become useful, mostly because of table names nobody can read. And there are five definitions of an active seller.

This card is one of 57 drawn from The Auction Block by Brian Cariveau — the story of an analytics leader with 90 days to fix what the cards describe.

Each leadership pattern falls into a few pitfalls most often. Find your pattern.The Analytics Leadership Mirror: free, 5 minutes, no account