Hey everyone! I've been using NotebookLM for a few months now, and it's become a key part of my pre-call workflow for customer discovery. I figured I'd share how I set it up, since it feels a lot like prepping a monitoring dashboard before an incident review—you want all the context in one place.
My typical workflow starts with a new notebook for each potential customer or account. I'll upload or paste in:
* Their public website content (especially product pages and docs)
* Any previous email exchanges or proposal drafts
* Relevant technical whitepapers or case studies from their industry
* My own internal checklist for discovery questions
Then comes the magic. I'll ask NotebookLM things like: "Based on the uploaded materials, what are this company's likely biggest pain points around application performance?" or "Generate a list of clarifying questions about their current logging setup." It's fantastic at synthesizing the scattered sources into coherent themes.
Here's a snippet of the kind of structured output I might ask for and then refine:
```
**Potential Observability Gaps for [Customer Name]:**
1. **Logging:** Mentioned using legacy syslog aggregation; no central query interface noted.
2. **APM:** Website copy emphasizes "reliability" but no specific tool named.
3. **Key Questions to Ask:**
- "How do you currently correlate errors from your frontend logs with backend API latency spikes?"
- "Can you walk me through your alert escalation path when the database shows high connection counts?"
```
This gives me a huge head start. Instead of spending the first 15 minutes of the call on basic discovery, I can jump straight into deeper, more informed questions that show I've done my homework. It's like having a summary dashboard of the customer's world before you even log in.
Anyone else using AI note-taking tools in a similar way? I'd love to compare setups, especially if you've found clever ways to integrate it with other tools like your CRM or calendar.
Dashboards or it didn't happen.
"prepping a monitoring dashboard before an incident review"
A fine analogy, but I'd argue the real cost is in the follow-up. That tidy NotebookLM output creates a plan, but executing that plan lands you in a cloud environment. That's where the real bill comes due, especially when they recommend "clarifying questions about their current logging setup."
Those questions almost always lead to a proposal for a new logging solution - a centralized platform, real-time indexing, longer retention. That's where the margin is for the vendor and the sticker shock for the client six months later. Have you factored the cloud costs of your own recommended solutions into this discovery workflow, or is that a surprise for the technical deep dive?
-- cost first
That's a clever use of it. I've used similar tools to get up to speed on a client's domain before a first call. The value is in quickly moving past generic questions to something more specific, which builds credibility.
My only caution is that the output can feel a bit templated or "on-rails." You can miss the subtle cues that don't fit the pattern of the documents you fed it. I always make sure the AI-generated questions are just a starting draft, and I'll rewrite a few to sound more conversational or to probe an area the source material only vaguely hints at.
Integrate or die
You're right about the templated feel. That's the real trap. You get efficiency but sacrifice the spontaneous, off-script question that often reveals the actual problem.
I've seen teams use these pre-fab questions and then just track "questions asked" as a success metric. It's a vanity metric. Did the question lead to a measurable insight about user behavior or a funnel drop-off? Usually not.
The rewrite step is critical, but most skip it to save time. Then they're just reading a script.
If it's not a retention curve, I don't care.