← All projects
Specimen No. 01, Podcast SaaS

A great product sitting in the dark doesn't pay the bills.

BrainCap had the product, no pipeline, no sales team. I built the machine that closed the gap, from ICP to booked meeting, fully automated.

n8nHubSpotGPT-4oListen Notes APIHunter.ioApolloCalendly
The harvest
2,400+
qualifying podcasts discovered & scored
Reply rate
8.4%
vs. 3–5% industry average for cold outreach
Time to first meetings
23 in 60d
qualified meetings booked, zero manual prospecting after setup
The soil, starting condition

BrainCap is a post-production platform for podcast creators, transcription, captions, translation, dubbing, all in one place. More than that, it makes every word of every episode findable. The product was proven. The problem was distribution: pre-revenue, no sales team, no marketing budget to burn. Every dollar had to work.

The seed, what we were serving

Podcast creators are everywhere, and nowhere. Finding the right show meant scrolling Spotify manually. Finding an email meant guessing. Writing a pitch meant starting from scratch every time. That doesn't scale, and it doesn't need to. There are over 800,000 active podcasts indexed by RSS. The data exists. The system to turn raw discovery into a booked meeting didn't.

The root system, what was built

Six workflows. One pipeline. Zero manual steps.

Before touching a single tool, we mapped the ICP across six dimensions, who, what, why, what keeps them up, what they want, and crucially, who to skip. That clarity made every downstream decision, the filters, the scoring, the copy, sharper and cheaper.

Discovery Enrichment Scoring Outreach Meeting Booked
n8n · Workflow 1A, Bulk Scrape
n8n Workflow 1A, Bulk Scrape from Listen Notes into HubSpot
WF 1A · Discovery. Reads the entire Listen Notes universe in paginated batches. Offset saved before each batch, crash-safe by design.
n8n · Workflow 1B, Weekly Monitor
n8n Workflow 1B, Weekly monitor and watch list promotion
WF 1B · Weekly Monitor. Every Monday at 9am, scans for new podcasts and promotes Watch List entries that have crossed the threshold.
n8n · Workflow 2, Enrichment Waterfall
n8n Workflow 2, Cost-ordered email enrichment waterfall
WF 2 · Enrichment. A cost-ordered waterfall, RSS feed, website scrape, Hunter, Prospeo, Apollo. Stops the moment it finds an email. Cheapest source always first.
n8n · Workflow 3, Lead Scoring
n8n Workflow 3, Lead scoring model 0–100
WF 3 · Scoring. Every lead scored 0–100 on episode count, publishing frequency, listen score, recency. Only 60+ enters the outreach queue.
n8n · Workflow 4, Personalized Outreach
n8n Workflow 4, GPT-4o personalized outreach
WF 4 · Outreach. Fetches the latest episode, feeds it to GPT-4o, writes a personalised two-sentence opener. Reads like a fan, not a bot.
n8n · Workflow 5, Meeting Handler
n8n Workflow 5, Calendly webhook to CRM and Slack
WF 5 · Meeting Handler. Calendly webhook → CRM update → warm prep email → Slack ping to the team. All within seconds of the booking.
Project walkthrough · 9 min

A guided tour of the full system, in the founder's voice.

The harvest, results

Zero to self-sustaining pipeline.

In the first 60 days the system discovered and scored 2,400+ qualifying podcasts, sent personalised outreach with an 8.4% reply rate against a 3–5% industry baseline, and booked 23 qualified meetings, with zero manual prospecting after setup. The pipeline now feeds itself every Monday at 9am without the founder doing a thing.

Andrew Huberman portrait
Andrew Huberman's Team

Among the clients acquired through the system: the team behind one of the world's most-listened-to podcasts. A bootstrapped startup with no sales team, landing a flagship name. That's what the right system makes possible, not just a result, proof of concept.

What broke. What I fixed.

Three things I'd do differently.

Problem

The scoring let too many weak leads through.

The first version qualified shows that technically hit the thresholds but weren't strong fits. Reply quality was low and it showed in the numbers.

Fix: Recalibrated thresholds, added recency weight. Signal quality improved immediately.

Problem

GPT-4o's first openers read like fan mail.

Early prompt versions were too complimentary, effusive, obviously AI. Creators are sharp; they spotted it immediately.

Fix: Rewrote the system prompt to be curious, specific, shorter. Tone shifted from fawning to human.

Problem

A lot of RSS feeds were broken.

More podcasts than expected had dead RSS URLs. A single-source enrichment would have failed silently on a large chunk of the list.

Fix: The waterfall was built partly because of this. Five sources meant resilience, not redundancy.

"Strategy first. Message second. System third."

Start a conversation