The Problem With How We've Been Doing This
Last month, I was knee-deep in a Looker Studio dashboard for a client who needed to explain Q3 performance to their board. Thirty charts. Two hundred data points. Zero coherent story. I've been here before—we all have. The data exists. The insights exist. But the bridge between them? That's where most projects die.
NotebookLM showed up at exactly the wrong time, which is to say the right time, and I'm still processing what that means. It's not a replacement for Power BI or Looker Studio. I don't think. But it does something those tools don't: it forces you to articulate why before you visualize what.
What Actually Matters in NotebookLM (Right Now)
- The ability to upload raw datasets and have the system generate narrative context without you narrating first. You don't guide it—it finds patterns you missed.
- Audio synthesis that turns your insights into podcast-quality explanations. I tested this with a financial services client last week and honestly, the tone was eerie—too professional, like it understood the stakes.
- Source citations embedded in every generated insight. This alone eliminates hours of audit trails I used to build manually in Power BI.
- Multi-document synthesis.
- Interactive querying that sits somewhere between natural language and SQL, except it actually works without you knowing either.
Here's What Broke My Workflow
I pride myself on controlling the narrative. In my work with Looker Studio, I decide which metrics matter, which colors matter, which story gets told. With NotebookLM, the AI suggests interpretations that are sometimes dead-on and sometimes so left-field that I have to stop and ask: am I missing something in the data, or is this a hallucination?
The answer is usually both. Not sure this is the right move, but I've started treating the AI's weird suggestions as a second layer of analysis. When NotebookLM flagged a correlation between customer acquisition cost and churn that I'd missed—and it was correct—I realized I'd been building dashboards for months without actually thinking about them.
That's uncomfortable. And valuable. And I genuinely don't know if it'll stick as a permanent workflow change or if I'll revert to my old methods once the novelty fades.
The Specific Thing That Made Me Pay Attention
I uploaded three months of e-commerce transaction data—about 47,000 rows from a mid-sized Bogotá-based retailer—to test how NotebookLM handled messy, real-world stuff. No cleaning. No prep. The system generated five narrative summaries, each taking a different angle on the same data. One focused on seasonal patterns. One on customer lifetime value. One on inventory turnover. One on something called "transaction velocity clustering" that I'd never even considered measuring.
I fed those summaries into Power BI, shaped them into four separate dashboards, and presented them to the client. The one about transaction velocity became their focus. They'd been optimizing for volume. Should've been optimizing for timing. That insight came from NotebookLM's willingness to ask questions I'd trained myself not to ask.
What I'm Still Figuring Out
Does NotebookLM make Looker Studio or Power BI obsolete? No. It doesn't create dashboards. It doesn't handle real-time data pipelines. It can't replace a BI engineer. But if you work in data storytelling—if you spend your time turning spreadsheets into meaningful narratives for stakeholders—this tool is already in your future whether you've tried it yet or not.
I'm not sure what happens next. The integration possibilities are obvious but not yet built. The guardrails around hallucination are better than they were three months ago but still unpredictable. And I keep wondering if I'm just chasing novelty instead of building something that lasts.
But I'm using it anyway. Every week. Alongside Looker, alongside Power BI, alongside the methods that got me here.
