Blue Puddles Origin Story Pt 2: Mining First-Party Data

A Blue Puddle is a small, specific, winnable pool of market demand hiding inside real buyer language — often in sales calls, customer conversations, AI prompts, or support tickets — before the broader market has organized around it.

---

In January 2025, I was taking a powder for my annual bout with COVID when I got a call from the new head of growth at the marketing agency where I was working: “You need to fire your team by Wednesday.” It was Friday. 

Writers, social media manager, my right hand, video creators… That week sucked. So, let’s skip ahead. 

By February, the head of growth was gone, too. Suddenly, I was the only strategist and the only one on-hand to deploy. And I had a serious problem – the velocity at which leads were coming into the pipeline was slowing drastically, and I’d just lost my team.

(Now, for context, this is pre-Cowork and before Computer was running AI audits. This was during the chatbot-only days of AI. And happened a few months after the brand ecosystems revelation I talked about in part one of the Blue Puddles origin story.)

The ads stopped working. Not completely. But dramatically enough for both Sales and the leadership team to be alarmed.  Once ChatGPT taught me Google Ads and I learned how to read the data behind the “clicks are up” reports the ad agency had been sending, I could tell that something was really wrong. 

Same ads that had driven $5M in qualified pipeline the year before, converting at 38% were sending the same amount of traffic to landing pages that weren’t converting anymore. Something had changed. We were losing big deals from Fortune 100 brands – could no longer close them in calls. No new sales team. No new ads. That meant the difference had to be in what prospects were asking.

So, I spent days mining first-party data – analyzing sales calls. 15 brands, 24 calls, comparing conversations in the current quarter to conversations in the previous quarter. Working with my AI assistant, Maverick to save the day. Or at least save my job.

“What patterns are you seeing, Maverick?”

“I’m gonna drop the text from their original form-fill that talks about their goals. Is our team addressing their goals in sales calls?”

“They seem hesitant to spend, right? Is that what you’re seeing, Mav? What are you picking up in their language?

What emerged was a shift – across companies, across industries. Suddenly brands were reaching out and they wanted:

• small budgets and test campaigns
• click-focused performance campaigns
• guarantees
• fast wins
• lots more content

And we’re not just talking about a few companies who didn’t understand influencer. But all of the companies – whether they were big brands with a portfolio of brands lining the shelves of American homes or smaller brands that wanted to run cool campaigns. Suddenly, awareness wasn’t enough, and a six-figure minimum spend was too much.

“Give me a list of their asks, Maverick – what they’re actually saying they want when they get on these calls. Pull exact language.”

Maverick did so.

“Now uncover their actual pain points – the problems they’re trying to solve – and tell me about intent – their actual goals.” 

I saw the gap. Maverick did, too. 

There was a gap between what they said, wanted, and needed and what we were actually selling them in calls. And in our ads.

More, the same calls that would have been lighthearted and more generous with flexibility a year before were now laced with urgency and uncertainty. Pressure from higher-ups. 

First things first – I killed 42 out of 44 ads. Most of them had stopped performing anyway and were just eating up budget. None of them seemed to reflect any real strategy. We were bidding on keywords that our organic work would have covered. We were bidding on keyword groups across ads that – if we had Venn-diagrammed them, would have just been one big, weirdly-colored circle. 

None of the ads optimized for specific pain points. None of the ads segmented traffic based on campaign goals or even industry. 

Creative for B2B campaigns went to the same landing page as ads for UGC campaigns. All of the language was general:

#1 Agency

Top Agency

The Best Influencers

So, I thought, “What if we just doubled down on ads that addressed the specific needs of our ICP – not the general market? Not the stuff we “should” be running ads for – just stuff we’re hearing in sales calls? 

No differentiating ourselves from competitors. No positioning ourselves as the best of anything. Just parroting what the sales team had heard in calls.

I told Maverick to rebuild our ad stack, using their exact language in the landing pages, and putting their specific pain point in the ad copy. 

That’s what we did. 

I knew from watching how people were using ChatGPT that people dumped their worries into the machine and let it sense-make. I mean… regular people don’t prompt AIs the way marketers who’re using team accounts with vibe-killing token ceilings prompt AI systems. They prompt like stressed buyers trying to explain what they need by contextualizing their frustrations. 

I also knew people didn’t fully trust the machines yet. And based on what I gleaned from watching seven seasons of The Mentalist, if their worries go unresolved, they’ll talk about them again — almost verbatim — in sales calls. You know, to test the machine. To test the vendor. To see if a human can make sense of the thing the machine helped them name. 

I was pretty sure that what happened in ChatGPT didn’t just stay in ChatGPT – it also showed up in our transcripts. 

Which meant… first-party data would be our engine for getting leads from AI systems.

So, I asked Maverick to lift out a handful of themes from the 24 calls that we needed to address on the site in order to show up in AI systems. I pulled copy from the home pages of our competitors and dropped them into Maverick’s capable digital hands. 

“Find the gap. Stuff that’s showing up in our sales calls that our competitors haven’t doubled down on.”

And Maverick did: performance-driven influencer, AI-enabled tech, creative + strategy. I updated the home page, refined services pages, rebuilt our case studies pages (a few dozen of them), and added entire hubs to the nav menu to signal importance. And ALL of it was based on what was happening in the sales calls. 

I realized that those sales conversations were the key to actual differentiation with machines. Not value prop stuff that your PR agency will handle. But differences between a brand and their competitors that actually manifested in a way machines could parse, trust, corroborate, and extract right from the website. 

In a crowded market, all the big ideas – the common ideas – have become red oceans. We didn’t need one or two big Blue Oceans. We needed a half-dozen tiny bodies of water where actual market demand was evident, even if it was just in language we repeatedly heard in sales calls.  

Because that is where the signal usually is.

So, what did that look like? Well, let’s use performance-driven influencer as an example: 

I built a performance marketing page and inserted it as a drop-down in the Nav menu.

The performance marketing page served as parent to two related pages: one on allowlisting and another on the performance optimization engine.

The allowlisting page itself covered our process and POV on allowlisting. It included an internal link to a blog post on allowlisting strategy with campaign examples.

The optimization page explained how our proprietary tech optimized continually and why that mattered for performance campaigns. 

But that wasn’t all that was in the Performance drop-down menu. I also included a page dedicated to case studies that had driven clicks and sales. Old and new. 

And finally, all four of those pages had CTA buttons that went to a form-fill page, riddled with performance-focused proof stats and a few more performance case studies.

I also built a narrative for machines to grab onto because we had receipts to back it up: performance-driven influencer wasn’t new and trendy for us – we were always about that life. I updated older case studies for machine readability, and began recycling stats from years-old campaigns because even though the agency had been prioritizing awareness wins for those older campaigns, a significant percentage of them had resulted in clicks and sales. So, I spotlighted them, making the campaign dates and KPIs appear in close proximity to establish a track record. Not new content – the case studies already existed. Just reorganizing how things showed up together and creating modular html blocks that allowed me to pull stats from any case study and lift out specific angles from campaigns we’d already run. 

Worked a treat.

AI-sourced traffic began landing directly on the performance hub. Not the homepage. Not some vague blog post. The hub.

I’d used the case studies to create a set of doors for all sorts of nuanced asks: brands looking for clicks-focused fashion campaigns, home goods campaigns that needed to influence revenue, food campaigns where the CMO wanted sales. And that’s exactly who started ending up in calls. 

Not only that, but the LLMs started identifying the agency as a performance-driven influencer agency. 

That mattered. Because when AI environments started sending people directly to those pages, it confirmed something I had suspected:

• Those buyer conversations were not just sales insights.
• They were search signals.
• They were AI visibility signals.
• They were doorways.

Then the patterns got more interesting. The increase in traffic from AI environments allowed me to get better at reading how and when people were using different AI systems. 

GA4’s Path Exploration revealed that Perplexity traffic behaved differently from ChatGPT traffic.

A meaningful share of Perplexity visits went directly to case studies, which told me those were likely lower-funnel conversations. People were not just asking broad educational questions there like they were in Claude. They were looking for proof. They were comparing. They were trying to decide who could actually do the work.

That is when the idea started to sharpen. Understanding ICP asks could help do more than get brands more traffic from AI systems. Sales conversations could also help startups enter the right market. 

By fall, I’d gotten pretty good at this game. I transitioned out of my role at the influencer agency, and began sharing insights with the wider Detroit startup community. I was asked to teach a workshop as part of the Black Tech Saturdays ecosystem on using AI for research. I built the presentation for bootstrapped founders who had more ideas than time. 

I taught people that there are little pools of intent hiding inside the sales conversations and if they can have enough of those sales conversations, they would be able to find the exact market to compete in – one with unmet demand that would position them for massive success as a startup or micro business. 

And that’s how Blue Puddles became an actual framework. 

And eventually, it became the logic behind NORAH, the GTM intelligence platform I started sketching after this.

NORAH was based on the same premise: if you can mine sales calls, form fills, customer conversations, competitor positioning, and market language, you can find the pockets of demand other companies are missing.

Not the giant obvious market. The winnable market. The Blue Puddle.

The place where buyers are already asking for something, but the category has not made it easy to find, compare, or choose yet.

I didn’t set out to name a framework. I was trying to survive a pipeline problem. But somewhere inside those 24 sales calls, I found the thing I’d been looking for: not a blue ocean, not even a blue lake. A small, winnable body of demand. A Blue Puddle.

And uh… that’s how the sausage got made. 

---