Internal Tooling · Launch Factory
AI Deal Flow Pipeline
An automation that standardized inbound pitch submissions — and removed the advantage of knowing how to pitch well.
The Problem
Launch Factory receives inbound pitch submissions through a website contact form. For a long time, reviewing them meant reading free-form emails, manually pulling out relevant details, and logging everything by hand. It was slow, inconsistent, and easy to let things slip.
But there was a subtler problem underneath the operational one: founders vary wildly in how well they pitch themselves. A polished submission from a mediocre business could outperform a rambling one from something genuinely interesting. The review process had a built-in bias toward founders who knew how to write a pitch — not toward founders building something worth funding.
I wanted to solve both.
How It Works
Webhook Trigger
Contact form submission with a specific subject line fires the pipeline.
JSON Parse + Router
Submission is parsed and routed — non-pitch traffic is filtered out immediately.
GPT Extraction Layer
GPT acts as an internal VC associate, extracting structured deal data from free-form text.
Airtable CRM
Clean, structured deal record lands in the CRM — every field populated, every pitch comparable.
Conditional Email Draft
If key fields are missing, the pipeline drafts a follow-up requesting the information.
The whole thing runs without human intervention. A pitch comes in, gets processed, and lands in Airtable as a clean, structured record — ready to review alongside every other pitch, on equal terms.
The AI Extraction Layer
The most deliberate piece of this was the GPT node. I prompted GPT to act as an internal VC associate with a specific job: read the raw submission and return a structured JSON object with standardized fields — company name, what they do, team background, traction, customers, fundraising stage.
The extraction works regardless of how the founder wrote their pitch. A three-sentence email and a twelve-paragraph deck get processed the same way and return the same schema. Every deal that comes in is immediately comparable to every other deal.
If required fields are missing — team structure, fundraising status, customer traction — the pipeline doesn't just leave them blank. It flags the gap and drafts a follow-up email requesting the information. The founder gets a response; we get what we need to actually evaluate the deal.
A founder who can't pitch well shouldn't get filtered out because of that.
The automation isn't just about saving time. It's about removing a structural bias in the review process. Pitch quality is a skill — and it's not the skill we're evaluating. By standardizing the input, we evaluate the business, not the pitch.
What This Replaced
Manual email review, copy-paste into a spreadsheet, and a lot of judgment calls made on inconsistent information. The new pipeline doesn't just automate that process — it makes the output of that process more reliable.
This wasn't a complex engineering project. It was a Make.com flow with a well-designed prompt at the center. The insight was recognizing that the real leverage was in the extraction layer — not the plumbing around it.