How Nomiki Petrolla turned cold sales pitches into 14 replies from 32 reverse pitches in a week

The problem

She says she has 50,000+ followers across LinkedIn, Instagram, TikTok and Threads, and that “On any given day I get 10–20 cold emails from companies pitching me their product.” She says, “I used to delete them all.” Her view is that some of these senders have products her founders would use and audiences that overlap with hers. She sums up the idea as: “Every cold pitch is a warm lead for a partnership. You just have to flip the script.”

The blueprint (as she describes it)

Scan inbox and spam. The script connects to Gmail over IMAP and pulls every email from the last 24 hours in both Inbox and Spam. Her reason for including spam: “A lot of cold pitches end up in spam, and some of the best partnership opportunities are buried there.”
Classify with Claude. Each email goes to Claude with a classification prompt that asks whether it is a real company pitching a real product that could become a partnership. The filter drops:
- mass marketing from big companies
- private equity acquisition spam
- dev shops offering to build your product
- SEO audit and LinkedIn growth hack outreach
- vague “let’s collaborate” emails
- emails with sketchy domain mismatches or mass-template signs
Score for partnership fit. Claude gives each pitch a relevance score from 1 to 5 based on fit with Theanna’s audience of early-stage founders. A 5 means “every founder needs this”. A 1 means a real pitch that is not a fit.
Write the reverse pitch from a context file. A pitch.md file holds Theanna’s audience, reach, current partners and partnership models. Claude uses it to write a reply that:
- acknowledges the sender’s product
- connects it to her founder audience
- cites her numbers (50k+ followers, 3k newsletter subscribers, 300k quarterly impressions)
- proposes a swap: premium placement in her content, a founder discount from them, a newsletter swap, cross-visibility
- names current partners Customer.io, Wispr Flow and Granola
- closes with “Down to chat about what this could look like?”
Her published build prompt caps replies at 150 words.
Draft as threaded replies. The script creates Gmail drafts threaded on the original email, using In-Reply-To and References headers. Drafts are formatted as HTML emails.
Send a summary. After drafting, the script emails her a report that ranks the pitches by relevance with star ratings.
Schedule it. A cron job runs the script on weekdays at 9am. It runs on her laptop, with no database or cloud infrastructure.
Human review before sending. She opens Gmail, reviews the drafts, sends the ones she likes and deletes the rest. She says review takes 5 minutes.
How she built it. She says the build took about 2 hours with Claude Code. “Most of that was refining the pitch voice and tightening the spam classifier.” The codebase is about 400 lines of TypeScript. The post includes the full prompt she suggests pasting into Claude Code to build a copy.
Planned, not built. She describes next steps: researching senders before drafting, logging contacts in Folk (her partnership CRM) through the Folk API, tracking pipeline stages, and drafting follow-ups for contacts with no reply after 5 days.

Stated results

First run (May 6, 2026): 47 emails scanned, 40 filtered out as non-pitches, 7 pitches identified. She calls this a “15% hit rate”.
From that day: she sent 3 drafts and “Two already replied within 24 hours wanting to set up a call”. She calls this a 43% response rate.
First week (5 business days):
- 47 pitches identified
- 32 reverse pitch drafts sent
- 14 replies (44% response rate)
- 6 partnership calls scheduled
- 2 deals in progress (one newsletter swap, one founder perk partnership)
- 5 minutes a day reviewing drafts
Before and after: before the agent she was responding to “maybe one pitch a week”. She now reports 1 to 2 partnership conversations a day.
Closed deals or revenue: None stated.

Caveats

Vendor and product. The post is on Theanna’s own site in its “Building Theanna” section. It ends with a pitch for the Theanna Agent, a product Theanna sells, and an invitation to join its founder community.
Short track record. Results cover one week. All numbers are self-reported.
Unreconciled figures. On day one she reports 47 emails scanned. For the full week she reports 47 pitches identified. The post does not reconcile these two figures.
Two response rates. The 43% is 2 replies from 3 drafts sent on May 6. The 44% is 14 replies from 32 drafts over the week.
Schedule wording differs. The intro says it “runs every morning at 9am”. The setup section says weekdays at 9am.
Self-assessed accuracy. She says the classifier “correctly identified” the pitches. No error rate or check is given.
Effort claim. She calls the outreach “zero effort to generate”. She separately states 5 minutes of daily review and about 2 hours to build.
Anonymized examples. Company names in her output examples are placeholders such as “[Fintech Startup]”.
Audience figures. The follower, subscriber and impression numbers in the replies are her own claims.
Omitted setup steps. The post includes Gmail and API credential setup steps. They are left out of this teardown.
Handles:
- X: @NomikiPetrolla (verified 2026-10-24: bio links theanna.io)
- LinkedIn: nomiki-petrolla (verified 2026-10-24: linked from theanna.io)
- Threads: @nomikipetrolla (verified 2026-10-24: profile says Building @theannaio)
- Instagram: @nomikipetrolla (verified 2026-10-24: same account as Threads, name match)
- Brand accounts: Instagram @theannaio and TikTok @theanna.io (linked from theanna.io)

Sources

https://theanna.io/building-theanna/how-to-pitch-partners-with-ai-the-reverse-pitch (primary: first-person build write-up on Theanna’s site, “As of May 6, 2026”)

Want agents like these working your distribution? See the Agent Stack Teardown. Get new teardowns weekly: subscribe (double opt-in).