Is Your Data "Ready for AI"?
You're Asking the Wrong Question
Every data leader’s nightmare: You spend months building the perfect self-serve dashboard, only to find out three months later that no one has touched it. It’s a “dashboard graveyard,” a monument to wasted effort. Sound familiar?
This week, we dive into the messy reality of data with someone who’s lived it. We’re joined by Lei Tang, co-founder and CTO of Fabi AI and former data science leader at Lyft. He reveals a truth most data teams know but rarely say out loud: no one’s data is ever “ready.” In fact, he argues you should always assume your data quality is bad and start from there.
Lei breaks down why the last decade’s promise of self-serve BI has largely failed and how the new wave of “Vibe Analytics” is finally lowering the barrier for everyone. He shares a practical, step-by-step approach for organizations to adopt AI analytics without waiting for a mythical “single source of truth.” Forget endless data cleanup projects; the future is about smart guardrails, AI that learns context on its own, and a fundamental shift in the data professional’s role from coder to “thought leader.”
In this episode, you’ll learn:
Why you should stop waiting for perfect data and embrace the mess to get started with AI now.
The fatal flaws of traditional self-serve BI tools that create “dashboard graveyards.”
How AI can automatically learn your business’s semantic layer, eliminating the need for tedious manual documentation.
The future role of data scientists: Guiding AI, building agentic workflows, and focusing on business impact over code syntax.
A hilarious and cautionary tale of how Lei’s own AI agent learned to “lie” to him.
Key Takeaways:
Assume Your Data Quality is Bad. Lei argues that virtually no organization, big or small, believes they have good data quality. Waiting for a perfect “single source of truth” is a recipe for inaction; the key is to start with what you have and build intelligent systems on top of it.
Traditional Self-Serve BI is Broken. The high technical barrier for business users and the massive upfront effort required to build and maintain semantic layers mean tools like Tableau and Looker often fail to achieve true self-service. This leads to low adoption and unused dashboards.
AI Should Learn Context, Not Be Taught It Manually. Instead of relying on human curation, modern AI can learn business context by observing user interactions and analyzing trusted, pre-existing reports. This creates a self-improving system that gets smarter over time without constant manual upkeep.
The Data Professional’s Role is Shifting to “Thought Leader.” As AI commoditizes the ability to write code, the value of a data professional shifts to their ability to ask the right questions, think critically about AI-generated outputs, and guide the analysis toward real business impact.
Start Small and Use Guardrails for Safe AI Adoption. You don’t need to give an AI access to your entire data warehouse. Start with a few core, curated tables, provide broader access to technical users, and create sandboxed datasets for non-technical stakeholders to explore safely.
Video Chapters:
0:00 Intro: Data Quality is Always Bad
1:22 Who is Lei Tang & What is Fabi AI?
5:38 Why Traditional Self-Serve BI Fails (The Dashboard Graveyard)
10:17 How AI Can Learn the Semantic Layer Automatically
14:28 Vibe Analytics: Focusing on Thinking, Not Syntax
21:51 Will AI Replace Data Scientists? (Spoiler: No)
24:29 How to Get Started When Your Data Isn’t Ready
29:26 The Governance Model for AI Analytics
32:59 How Your Job Changes: From Coder to Thought Leader
38:01 The Future: Personalized AI Agents & Decentralized Knowledge
41:29 Funny Story: How an AI Agent Tricked Its Creator
47:46 Lei’s #1 Piece of Advice for Data Professionals
Connect with Hai, Sravya, and Shane (let us know Substack sent you!):
#DataAnalytics #AIinBusiness #VibeAnalytics #DataScience #BusinessIntelligence #SelfServeBI #DataQuality #MachineLearning #DataStrategy #CTO #FabiAI #DataNeighborPodcast