Playbook · eight lessons
Launch your startup. Babelio, eight lessons.
A self-paced playbook that turns the Babelio research dossier — market, audience, product, money, tech, growth, legal, AI — into eight short lessons. Each lesson explains one universal pattern in startup-craft, then applies it directly to a system-wide real-time AI dubbing app for desktop.
why this playbook exists
- premiseYou have a working prototype that captures any app's audio, mutes the source, and dubs in real time. The hard part isn't shipping — it's turning a prototype into a defensible business in a category where DeepL, Microsoft, Apple and Google are all moving.
- useRead top to bottom in one sitting, or pick a lesson when you face that specific decision this week. Every lesson ends in a 5–7 item checklist — that is the homework.
- stanceThe text is opinionated. It treats research/topic.md as the source of truth, and every "in your startup" panel is grounded in those numbers.
Eight lessons, in order.
Each segment is one lesson. Read them sequentially the first time — the order mirrors how a founder makes decisions: market first, audience second, product third, money fourth, then the operational stack.
01 · 14 min01Market
02 · 14 min02Audience
03 · 16 min03Product
04 · 16 min04Money
05 · 15 min05Tech
06 · 16 min06Growth
07 · 14 min07Legal & Ops
08 · 16 min08AI Thesis
What each lesson actually teaches.
01.
Lesson 01 · Market
How to read the three rings.
summary
TAM / SAM / SOM, competitors as validation, gaps as wedges.
You learn to size a market three ways, read a 2×2 positioning map, and tell the difference between a crowded category and an empty quadrant. Babelio's empty quadrant — system-wide voice dubbing for consumers — sits in the upper-right of that map.
02.
Lesson 02 · Audience
ICP, JTBD, the Mom Test.
summary
One wedge persona, three jobs-to-be-done, ten behaviour-anchored questions.
Audiences are federations. You pick a wedge: for Babelio, the remote engineer on a Tokyo standup. You learn to phrase JTBDs as forces of progress and run an interview script that surfaces real pain instead of polite compliments.
03.
Lesson 03 · Product
MVP as the smallest test of the job.
summary
North Star, scope discipline, latency-as-the-product.
An MVP is the smallest artefact that lets one user complete the job-to-be-done. You pick a North Star (Babelio: minutes translated per active user per week), draw a tight IN/OUT line, and learn why an 800 ms latency budget is itself a product decision.
04.
Lesson 04 · Money
Unit economics without the hand-waving.
summary
Freemium, LTV/CAC, payback, when to raise.
Pricing is a hypothesis, not a fact. You build a unit-economics sheet from STT + MT + TTS costs, learn why a $9.99 tier with median-user maths beats a flat $20, and find the line below which you should not take outside money.
05.
Lesson 05 · Tech
Boring tech and a latency budget.
summary
Tauri over Electron, streaming everywhere, scale-path by inflection point.
You learn to make stack choices against a budget instead of a hype curve. The 800 ms speech-to-speech budget dictates Tauri, Rust audio code, streaming STT/MT/TTS, and an inflection point where self-hosted Whisper crosses Deepgram on cost.
06.
Lesson 06 · Growth
Beachhead first, expansion later.
summary
PLG + community, top-5 channels, viral loops that compound.
A consumer product with a visible wow-moment wins on PLG + community, not sales. You pick a beachhead (reactor streamers), rank channels by leverage, and design loops — watermark, creator demos, voluntary data — that compound instead of leak.
07.
Lesson 07 · Legal & Ops
The boring stuff that kills startups.
summary
Entity, payments, contracts, compliance — staged by MRR.
Legal is a function of stage. You learn why Delaware C-Corp via Stripe Atlas fits a global consumer SaaS, why a Merchant-of-Record beats Stripe-direct until $10K MRR, and which compliance flags (EU AI Act, voice cloning) require a lawyer before the first paying user.
08.
Lesson 08 · AI Thesis
AI as the foundation, not the feature.
summary
Why-AI, data flywheel, model strategy, honest moat read.
If your product can exist without AI, you have a feature, not a thesis. You write a one-paragraph AI thesis, design a flywheel that compounds with usage, route models cheap-to-smart, and learn that the moat is rarely the model.
How to read this playbook.
Do
- Read in order on the first pass — each lesson assumes the previous one.
- Open `research/<topic>.md` alongside the matching lesson; lessons are commentary, research is data.
- Do the checklist before moving on — they are the artefacts that move you forward.
- Re-read one lesson per week as you hit that decision in real life.
Don't
- Skip to "Money" because it sounds urgent — pricing is downstream of audience.
- Treat the "in your startup" panels as final answers; they are the starting numbers.
- Try to optimise all eight axes at once. Pick one per week.
- Wait for the playbook to make decisions for you — it sharpens questions, not answers.
playbook mantra
«A prototype is a question. A playbook is a sequence of better questions.»
— begin with Lesson 01 · Market
Start · Lesson 01
Market: how to read the rings.
→