Morrie Schonfeld - How I built a $500k/mo ai app (so you can just copy me)

Co-founder Michael's rule after a food app that didn't do well: don't build a consumer app that can't go viral. They combined their own problem (practicing spoken Chinese without a native partner) with an AI feature that is interesting to watch: holding to speak and having a real conversation with an AI. The 'hold to speak' screen and Pingo's minimal four-colored sound-wave 'character' are what appear in marketing.

For the first ~3 months the two founders made nearly all content (only 1-2 influencers tried). Michael ran a Korean account learning Korean from scratch; Morrie ran a Chinese account. Every video: holding the iPad in a dorm room talking to Pingo, trying different hooks. First viral hook: 'POV, your Chinese teacher on Sundays used to hate you so now you seek validation from an AI', with English and Chinese captions so non-speakers could follow. With founders as the only channel, a user spike could be attributed directly to one video.

Rule from day one: the product (talking to Pingo) is the focus of every video. Early hooks about language learning that didn't show Pingo (people on TikTok discussing languages, speaking a little Chinese then a demo, talking about learning) didn't go viral. Talking-with-Pingo videos worked for views and, being demos, converted well. The same rule applies to all ~100 creators.

Hooks that worked: basic conversation aimed at beginners ('basic Korean quiz', 'learning X in Y language'); cultural hooks per language (K-drama, Squid Game audio on Korean videos, the TikTok ban and RedNote for Chinese, French music on French videos). Then blended in humor, which went even more viral. A content lead who lived in Germany coaches German creators on culture.

Creator mispronounces a word that sounds like something else (often rude) in English, or is just bad at the language; Pingo roasts them instead of saying 'great job'; creator shows emotion, often crying via a Snapchat crying-filter hook. Variations: Pingo is mean without crying; creators add props or activities (knitting, writing, slime, cutting fruit). Run in Korean, German, French, Arabic, Japanese, Russian and more.

Custom scenario in roleplay mode ('I am blank. You are blank. We are doing blank. Go.') has existed since V1. Creators prompt Pingo to be mean, say specific lines or use a tone, sometimes framing it as actors in a play to get around model limits. It wasn't built for marketing; creators repurposed it. Users now ask for a dedicated 'mean Pingo' toggle, which is on the roadmap.

Tried TikTok slideshows (e.g. 'Top five phrases to get fluent in Japanese', last slide 'I learned it on Pingo'), some with strong calls to action. Accounts spiked then crashed. Conversion: strong-CTA slideshows beat weak ones, and videos using Pingo beat both by a wide margin.

VAs (Morrie did it himself for the first few thousand) search TikTok/Instagram by keywords: 'learning German', 'practicing German', 'speaking German'; if the well runs dry, 'German major/minor/class'; native speakers last. Target micro-creators under 10K followers (10-30K ideal, 5-10K also good; some under 2K blew up) with at least one breakout video and real language skill; guess their viewers' country. DM: 'Hey, I love your [language] content. I'm with Pingo. We're a language app with 3 million users. We'd love to make content with you.' On a yes: new account, warm it up, per-video rate plus bonuses, app access; then move to email and a CRM to track follow-ups.

Base pay per video so creators feel their time is valued; the real money is bonuses at 50K, 100K, 200K, 500K and 1M views. Tiers don't stack: a creator gets the highest tier reached (e.g. $100 at 100K, not $100 + $50). Creators post on a new, warmed-up account. An internal tool pays creators.

High-volume recruiting, then an internal clock for how long each creator gets before a keep/cut decision. Top 20-30% produce most views; the bottom produce none ('dead spend'). Recently cut 30-50 creators, leaving ~100. A Discord lets creators share what went viral and teach each other, so the middle follows the top creators' lead. Shifting from 'scale mode' to fewer, more deliberate creators.

Started content coaching about a week before recording. Many flops are good ideas with bad execution: fix lighting, camera angle, audio, framing, and have the creator do something while talking (knitting, drawing, slime, cutting fruit). Point out one small tweak on a recent video. Long-term plan: native coaches per language to feed cultural ideas.

Two budgets: in-house creators (base + fixed bonuses, hard to turn off) and Noise creators on a capped CPM (easy to raise or cut). Noise scales what works and tests quickly. Core in-house creators are the testing cohort. Brief: an example video, why it went viral, key steps, 'recreate as close as possible in your target language'. For a never-tried idea, the team films the example first. Not granular A/B testing.

Russian-language creator content went very viral, but mostly among native Russians (one UK-based creator had 90%+ Russian viewers). Apple, Google Play and web payments are blocked in Russia by sanctions. Now checking creators' language and viewer geography when recruiting, and working on making videos viral among learners rather than native speakers.

Multi-page paywall; first pages constant ('Learn to speak a new language with Pingo. We'll remind you 2 days before your trial ends' + timeline). Test page copy that won: 'Today, you unlock Pingo. In 5 days, you get a reminder and you'll be charged, and try it for free', annual shown with a weekly equivalent. Lost to it: 'Join over 3 million learners' with a real review, ratings and 'No commitment. Cancel anytime'; and a monthly-vs-yearly side-by-side. Intro-offer version: 'one-time offer for 55% off' with the same trial timeline beat the old 'one-time offer' page. Idea borrowed from a Superwall podcast.

Annual $99.99/year with a 7-day trial is the default, shown as a weekly equivalent because the yearly price looks daunting. Monthly plan has no trial and is only visible after tapping 'View All Plans' (shows 12-month and 1-month). Tested showing full annual, annual with monthly equivalent, and weekly equivalent copy.

A/B test of 3-day vs 7-day trials, also varying the reminder (2 days vs 1 day before the trial ends). Segmented US vs international. Next: check usage on days 4-7 of the 7-day trial, since shorter trials save AI voice costs if conversion and ARPU hold. Android users are more sensitive than iOS.

Main paywall offers a 3-day trial; if the user abandons the transaction, a second paywall offers a 7-day trial, cheaper. Messaging is layered so each offer feels better than the last. Next test: 10- or 14-day trial for people who turn down the 7-day, then check post-trial conversion.

Gradually shifting all US billing to Stripe web checkout through Superwall because margins are better than in-app purchase; not possible internationally, so paywalls are segmented US vs international.

Went through YC with strong revenue and CAC/LTV. Investors sought them out: ~100 meetings scheduled, most canceled after the round closed in ~3 days. Venture money covered negative cash flow from app store payout timing and high voice-AI costs. Partner advice: keep pushing the UGC lever that works.