MiuLabs AI Course — Launch & Go-to-Market Plan · 上市與市場推廣計劃

Prepared Oct 17, 2026 (America/Los_Angeles) for Afonso Miu / MiuLabs. Covers the course "Practical AI for E-commerce & Business — 實戰商業AI課程" and the hands-on track "AI Infrastructure for Real Business Problems".

How to read the numbers. Tool prices marked (verified Oct 17, 2026) were checked on the vendor's pricing page today. Market benchmarks marked (sourced: …) come from the Marketing research report (/workspace/miulabs-course-research/report.md, Oct 17, 2026) and name their source and date. Everything labelled Example is an assumption for planning, not a forecast. Anything else says check current pricing. This is not legal or tax advice.

Revised Oct 17, 2026 with Marketing's research. The first draft assumed a $5–$8 cost per lead. Sourced benchmarks are roughly 4–6× higher: Meta education lead ads average $28.22/lead and LinkedIn B2B $202/lead. The ad plan, kill thresholds, unit economics and 90-day scenario below are rebuilt on those figures. Ads now only feed a free workshop or kit, which leads into the cohort, team training and services. Organic content does most of the work.

0. One-page summary — the decisions

1. What exists today (inventory)

Proposed hands-on lab list (Product to confirm; it maps to the brief): L1 Local LLM in 30 minutes (Ollama / LM Studio) · L2 Cloud vs local benchmark on store tasks · L3 Structured outputs: listing generator for products.csv · L4 RAG support bot over store policies + eval set (data/docs, data/eval) · L5 Agents with tool calling / MCP (order lookup) · L6 n8n automations (reviews → sentiment → Slack) · L7 Self-host on a NAS (Docker: Ollama + Open WebUI + n8n) · L8 Expose safely with Cloudflare Tunnel + Access · L9 Observability: logs, traces, cost per run, evals in CI · L10 Capstone: production-ready support stack for the KK store.

2. Positioning and personas

Positioning statement. For e-commerce operators and aspiring AI engineers who are tired of AI demos that don't survive contact with a real store, MiuLabs teaches the AI stack companies actually run, from prompts to a self-hosted RAG bot behind Cloudflare. Unlike generic AI courses, every lesson ships a working deliverable on real store data, in English and 繁體中文.

What we don't sell. Not the generic "start an AI automation agency / get your first AI client" promise. It's crowded at every price, from $59/month Skool groups to sales-call programs (see §4.3), and it pulls in the wrong buyers. No income claims or income guarantees either. MiuLabs sells working infrastructure for stores and the skills to run it.

Tagline options (Marketing to test):

• "Practical AI that ships. 會落地的AI。"
• "From prompts to production — for people who sell online."
• "Build the AI stack your store actually needs."

Message map: P1 → "save 10 hours/week on listings and support". P2 → "build 10 production-grade AI infra projects". P3 → "a 90-day AI rollout plan your team will actually follow". P4 → "實戰商業AI，中英雙語". P5 → "private AI on your own hardware, set up in a week".

3. Offer ladder and pricing (all prices are Examples)

Pricing rules.

1. Founding price for the first cohort only, with a hard deadline (honest scarcity: 25 seats because Afonso reviews every capstone).
2. Cohort buyers get the self-paced course free; self-paced buyers get their price credited if they upgrade to a cohort within 60 days.
3. Regional pricing (Example): consider ~30% lower list prices in TWD / HKD for the zh-Hant edition, using Stripe / Lemon Squeezy local currency. Check purchasing-power fit with 10 real buyers first.
4. Payment plans for the cohort (3 × $420) through Stripe. Higher completion and conversion; slightly more refund risk.
5. Never discount the Core more than 20% outside launch windows. Use bonuses (office hours, kits) instead.

4. Monetization options compared

4.1 Unit economics (Example assumptions, clearly labelled)

Payment fees per sale (fee rates verified Oct 17, 2026; international cards and currency conversion add more, so check current pricing):

Paid-ads funnel, rebuilt on sourced benchmarks. Inputs:

• Meta cost per lead: $28.22, the median for Education & Instruction lead campaigns, with a 10.05% conversion rate (sourced: WordStream/LocaliQ Facebook Ads Benchmarks 2025, Sept 2025; data Apr 2024–Jun 2025).
• LinkedIn B2B cost per lead: $202. Native lead forms run $193 and document ads $142 (sourced: Metadata, Aug 25, 2026; 2025 data).
• Webinar show-up: 57% of registrants (sourced: ON24 2025 Webinar Benchmarks, Feb 2025; B2B-heavy data).
• Free workshop → paid course: 5–10% on average. One free workshop converted only 2% (1,000 leads, 20 students) (sourced: Maven Insights, Feb 15, 2023).
• Free community → paid: about 0.4–0.5% (sourced: Nick Saraev's own figure).
• Refunds: no reliable independent benchmark exists, so this plan uses a 5% refund rate as an Example only.

• Break-even CPL for the cohort (Example, with 5% refunds): about $18 if 2% of leads buy, $46 at 5%, $91 at 10%. That's why the Meta kill line is $45 (§6.4). Above it, the cohort only pays back if conversion beats Maven's average.
• Marketing's estimate (an inference, not a benchmark) is about $280–$1,400 in ad spend per paying student. Cold ads therefore only make sense for offers worth several hundred dollars or more (cohort, team training, services), reached through a free workshop or kit.
• Lesson: ads can't sell the $249 course or a $29 kit profitably. Point them at the free workshop or kit, judge them on revenue per lead (RPL) across the cohort, B2B and service offers, and let organic content carry the self-paced Core.

Cohort economics (Example): 25 seats × $1,190 = $29,750 gross − Stripe fees ≈ $870 − refunds 5% ≈ $1,488 − TA / guest $1,500 − tools $100 = ≈ $25,800 net. That's ≈ $215/hour on about 120 hours (prep + 10 weeks).

Membership (Example): $29/mo with 8% monthly churn → average lifetime ≈ 12.5 months → LTV ≈ $362 (≈ $348 after Stripe fees). Worth it once 300+ people have bought the Core. A free community converts to paid at only about 0.5% (sourced: Saraev's figure), so a 1,000-member free Discord would yield about 5 paid members (Example). Treat the membership as a catch-net for people who won't buy the cohort, not as the main earner.

n8n affiliate (Example): 30% of referrals for 12 months (sourced). If 20 students start a paid n8n plan at an assumed $20/month (Example price; check n8n's current pricing), that's 20 × $20 × 30% × 12 ≈ $1,440 in the first year. Disclose the link in lessons.

First-90-days revenue scenario (Example, not a forecast): 1,000 leads, about 910 organic + about 90 paid (≈ $2,500 ad spend at the $28.22 Meta median) → 20 founding cohort seats at $990 (2% of leads, the low end of Maven's examples) ≈ $19.2k · 20 Core pre-orders at the $199 founding price ≈ $3.8k · 40 tripwires ≈ $1.1k · 1 team workshop at $6,000 ≈ $5.8k · 1 infra setup at $4,500 ≈ $4.4k → ≈ $34.3k net of fees, ≈ $31.8k after ad spend. The cohort plus one B2B or service deal is about 86% of it. Paid ads directly produce under 10% of the leads, which is why organic and the workshop come first.

4.2 Top 3 monetization paths (from Marketing's research)

Avoid (Marketing's inference): the generic "start an AI automation agency" promise. It's crowded at every price, from Kia Ghasem's $59/month group to Liam Ottley's sales-call accelerator. Also avoid income guarantees; even AAA carries an earnings disclaimer.

4.3 Market evidence

Real examples from Marketing's research. All pages were opened Oct 17, 2026, and prices are as shown that day.

Pattern: the big earners grow mainly through organic YouTube, X and LinkedIn. Marketing found no verified creator in this set who names paid ads as their main channel. Maven calls LinkedIn "the most powerful top-of-funnel channel" for its top instructors (sourced: Maven Help Center, Jun 19, 2026).

5. Website plan

5.1 Pages and funnel

Ad / post / video ──► Landing page (persona-specific, EN or 中文)
                         │ email opt-in (free kit or workshop)
                         ▼
                   Thank-you page ── tripwire $29 offer (one-click Payment Link)
                         │
                   5-day email course ──► Workshop (live or replay)
                         ▼
                   Sales page (Core / Core+Labs / Cohort) ──► Checkout (Lemon Squeezy / Stripe)
                         ▼
                   Webhook → Worker → D1 enrolment → magic-link welcome email → /learn
                         ▼
                   Onboarding email + community invite → upsell (cohort credit, kits, services)

Design / motion (your preferred stack): Lenis for smooth scroll, synced to GSAP ScrollTrigger; GSAP (SplitText, ScrollTrigger) for hero and section reveals; React Bits for a few signature components (animated text, spotlight cards, magnetic CTA). Rules: respect prefers-reduced-motion, keep LCP < 2.5 s on mobile (ad traffic is ~80% mobile), lazy-load heavy effects below the fold, no animation on checkout or forms.

5.2 Primary stack: minimum cost on Cloudflare (recommended)

Payment decision.

• Self-paced products and kits (fully automated digital goods) → Lemon Squeezy (5% + 50¢, merchant of record, handles global VAT / GST / sales tax) or Stripe Managed Payments (+3.5% on top of Stripe fees; "online courses and training" is an eligible category; Stripe's docs say products involving human intervention such as live coaching don't qualify).
• Cohort, team training and services (human-delivered) → Stripe Payment Links / Invoicing under your own business; handle tax with your accountant (Stripe Tax for calculation: check current pricing).
• Why MoR for digital goods: a solo founder selling to the EU, UK, Australia, Taiwan and US states otherwise has to register for and file VAT / GST in many places. The ~2–3.5 percentage points of extra fees are cheaper than that work.
• Note: reports say Stripe's Managed Payments was built from its Lemon Squeezy acquisition. Check Lemon Squeezy's roadmap and approval criteria before committing; both require account review.

Minimal data model (D1): users(id, email, locale, created_at, utm_source, utm_medium, utm_campaign, ref) · orders(id, user_id, provider, product, amount, currency, status, created_at) · enrollments(user_id, product, granted_at, expires_at) · progress(user_id, lesson_id, completed_at) · sessions(token_hash, user_id, expires_at) · events(id, user_id, name, ts, props).

Enrolment flow: checkout → provider webhook (verify signature) → Worker upserts user + order + enrolment in D1 → Resend sends the magic-link welcome → Worker fires server-side Purchase to Meta, TikTok and GA4 → adds the contact to the "customers" segment in the list tool.

5.3 Video hosting compared

Cloudflare terms: serving video from R2 or Stream is the sanctioned path. Don't proxy large video from other origins through the free CDN (check current service-specific terms).

5.4 Monthly cost at 0 / 100 / 1,000 students

Example assumptions: ~25 hours (1,500 min) of course video; each active student watches 300 min/month at ~720p (≈ 0.01 GB/min → 3 GB/student/month); mailing list of 0 / 500 / 5,000 contacts. Payment fees are per-sale (see §4.1) and excluded here.

Hosted-platform comparison (verified Oct 17, 2026, monthly billing; annual is cheaper):

Build vs buy recommendation. At 100 students, a hosted platform costs ~$50–$250/month vs ≈ $1–$2 on Cloudflare. At 1,000 students it's still ≈ $50–$500/month vs ≈ $40–$70. Building is also course marketing ("this site is a capstone of the stack I teach"). Build, with guardrails: time-box the LMS MVP to 10 working days (auth, enrolments, lesson pages, video, progress). If it slips past 2 extra weeks, run Cohort 1 on Maven (no monthly fee, 10%) and keep building self-paced on Cloudflare.

6. Social ads plan

6.1 Channel plan

6.2 Funnel by temperature

6.3 Creative angles and sample hooks

Rules: show the real screen and real output in the first 2 seconds, add captions in the ad's language, never make income claims ("make $10k with AI"). Meta and TikTok scrutinise these, and they attract the wrong buyers.

6.4 Test budget and thresholds (budgets are Examples; benchmarks are sourced)

Organic produces the volume and ads are a sanity check. At the $28.22 Meta median, $1,000 buys only about 35 leads, so paid ads can't carry the launch. The 90-day plan assumes over 90% of leads come from organic content, the newsletter, partners and the workshop.

Phase 1, Check (Nov 2 – Nov 22, $1,000 total, Example):

YouTube ads are $0 for now. Organic YouTube is the main channel.

Phase 2, Feed the pre-sale (Dec 3 – Jan 8, up to $1,500, Example): only if Phase 1 meets the keep rules below. Put 70% into retargeting to the December and January workshops and the cohort deadline, and 30% into the best cold ad set. Raise budgets by at most 20–30% every 3 days.

Decision rules. The thresholds are Examples derived from the sourced benchmarks and the break-even math in §4.1:

• Meta / IG. Target CPL ≤ $28 (median $28.22). Kill the channel if CPL > $45 after $500, the cohort break-even at 5% conversion. Kill an ad after about $60 spent (≈ 2× median CPL) with no lead.
• LinkedIn (B2B only). Target ≤ $150 using document ads or lead forms (median $202). Kill if CPL > $250 after $750, or if the first 8 leads produce no sales call. One $6,000 workshop pays for about 29 leads at $202.
• TikTok / X / YouTube / Xiaohongshu. No reliable benchmark, so a channel stays only if its cost per opt-in is ≤ Meta's.
• Scale (move to Phase 2) only if CPL ≤ $28 and workshop show-up ≥ 40% and RPL ≥ 1.5 × CPL once the first workshop's sales are in.
• Workshop → paid. Benchmark 5–10% (Maven). Rework the workshop or offer if fewer than 2% of registrants buy anything within 60 days after 2 workshops.
• Show-up. Benchmark 57% (ON24, B2B-heavy). Target ≥ 40% for a consumer audience, and use reminders and a replay if it falls short.
• Pause everything during Nov 23 – Dec 2, 2026 (BFCM ad auction inflation; operators are busy). Use that window for organic and building.

6.5 Tracking

• UTM convention: utm_source={meta|ig|yt|tiktok|li|x|xhs|newsletter|partner} · utm_medium={paid|organic|email|affiliate} · utm_campaign={2026q4_freekit|2027q1_cohort1|...} · utm_content={hook-id_format_lang} (e.g. listings9min_reel_zh). Product builds a UTM builder sheet; Marketing uses it for every link.
• Capture: the Worker stores first-touch and last-touch UTMs + ref in D1 at opt-in and passes client_reference_id / custom data to checkout so purchases carry attribution.
• Pixels + server events: Meta Pixel + Conversions API, TikTok Pixel + Events API, LinkedIn Insight Tag, GA4. Fire Lead on opt-in and Purchase (value, currency) from the payment webhook, with shared event_id for deduplication.
• Consent: load ad pixels only after consent for EU/UK visitors (cookie banner); Cloudflare Web Analytics is cookieless.
• Dashboard: a single Google Sheet (or D1 → simple /admin page): spend by channel (manual or API), leads, CPL, sales, revenue, RPL, CAC, ROAS, refunds.

6.6 KPIs and targets (Examples, re-baseline after Phase 1)

7. Organic content and community

Organic is the primary acquisition channel. The comparable creators grew mainly through YouTube, X and LinkedIn (§4.3), and paid leads cost about $28 or more. Budget Afonso's time accordingly: the weekly long video and the monthly free workshop matter more than any ad.

Content engine (Afonso records; Marketing repurposes):

Open-source magnet: publish a public miulabs/ai-store-starter GitHub repo (listing generator + RAG bot skeleton on the KK data). Stars, issues and forks bring builders to the paid labs. Keep the polished labs paid.

Community:

• Discord: #start-here, #wins, #help, #labs, #cohort-1 (private), #zh-中文. Weekly "show your build" thread. Cohort spaces close to the public; alumni stay.
• zh-Hant: LINE OpenChat or a private FB group for TW/HK operators; IG + Threads for discovery.
• Rituals: Monday goals, Friday demos, monthly "teardown" of a member's store workflow (with permission) as content.
• Partnerships: co-host workshops with Shopify / e-com agencies, homelab and NAS YouTubers, TW/HK e-commerce associations, n8n community. Offer a 20–30% affiliate commission (Lemon Squeezy has built-in affiliates: check). On the outbound side, join the n8n affiliate program (30% for 12 months) and disclose it wherever n8n is taught.

8. Roles and RACI: Product and Marketing

Charters (handoff-ready):

• Product (AI teammate) owns what we sell and how it's delivered: offer definitions, curriculum completion plan, labs repo, website and LMS build on Cloudflare, checkout and webhooks, analytics instrumentation, student experience, QA, pricing experiments setup. North-star metrics: on-time launch, completion rate, refund rate, LMS uptime, monthly infra cost.
• Marketing (AI teammate) owns who hears about it and why they buy: positioning and messaging, landing-page copy (EN + zh-Hant), lead magnets, email sequences, organic calendar, ads, community growth, partners and affiliates, reporting. North-star metrics: leads, CPL, RPL, workshop show-up, CAC, list growth.
• Afonso is the face and expert: teaches, records, approves pricing / spend / brand, closes B2B and service deals.

RACI (R = Responsible, A = Accountable, C = Consulted, I = Informed):

8.1 Concrete first tasks (start Mon Oct 19, 2026)

Product: first 10 tasks

1. Write docs/offer-spec.md: final tier names, inclusions, prices (from §3), upgrade credit rules, refund rules. Afonso approves by Oct 21.
2. Finalise the hands-on lab outline (L1–L10) with objectives, prerequisites and deliverables in hands-on/labs/README.md. Fill data/docs/ (store policies: shipping, returns, care, final-sale) and data/eval/ (30-question eval set) using the KK store data.
3. Build Lab 1 "Local LLM in 30 minutes" end-to-end (it's the lead magnet). Due Oct 30.
4. Scaffold the repo miulabs-site: Next.js static export (or Astro) + Tailwind + GSAP + Lenis + 3 React Bits components; /en and /zh-hant routes with hreflang; deploy to Workers Static Assets with preview deployments.
5. Create the Worker API + D1 schema (§5.2), /api/subscribe with Turnstile, UTM capture, and Resend double-opt-in email.
6. Set up Lemon Squeezy (and a Stripe account for human-delivered offers); create products and Payment Links in test mode; webhook → D1 enrolment → magic-link email; test the full purchase loop.
7. Implement pixels + CAPI / Events API server-side Lead / Purchase; build the UTM builder sheet; verify events in Meta Events Manager.
8. LMS MVP (10-day timebox): lesson pages from MDX, gated by session; R2 presigned downloads; HLS player with ffmpeg encode script; progress tracking.
9. Port site/index.html to /syllabus; draft legal pages from §10 for review.
10. Weekly cost and uptime report (Cloudflare dashboard + email tool + payments).

Marketing: first 10 tasks

1. Write the messaging doc: positioning statement, 3 taglines, persona message map, objection list with answers, in EN + 繁中. Due Oct 22.
2. Write copy for the home page, /free-kit landing page and thank-you page (EN + zh-Hant) for Product to implement.
3. Write the 5-day email course + 3-email workshop invite sequence + cohort launch sequence (open, mid, 48 h, 24 h, closing).
4. Build a 30-day organic content calendar (YouTube, Shorts/Reels, LinkedIn, X, IG/Threads zh) from the §7 cadence; list the first 8 video topics from the lab list.
5. Produce organic first: 8 YouTube Shorts / Reels scripts from the §6.3 hooks. Then turn the 6 best into ad creatives for the $1,000 Phase-1 test (§6.4). Set naming = utm_content IDs.
6. Set up ad accounts (Meta Business Manager, LinkedIn Campaign Manager; TikTok for boosts only); confirm the Pixel/CAPI events with Product; create audiences (site visitors, video viewers, list lookalike). Load the §6.4 kill and scale thresholds into the KPI sheet.
7. Stand up Discord + zh community with channels, welcome flow and rules.
8. Partner list: 30 targets (e-com agencies, homelab/NAS YouTubers, TW/HK e-com communities, newsletters). Draft outreach template for Afonso to send. Prepare the n8n affiliate application for Afonso to approve, and check the 小紅書专栏 / 知识星球 / WeChat payment and qualification rules for the zh edition.
9. Collect proof: line up 5–10 beta readers for Lab 1 and the free kit to get early testimonials and screenshots.
10. Weekly KPI report (Fridays): leads, CPL, RPL, show-up, sales, top 3 content pieces, next experiments.

9. 30 / 60 / 90-day timeline (Day 0 = Mon Oct 19, 2026)

After day 90: record cohort sessions → edit into the self-paced Core (launch ~Mar–Apr 2027) → open membership once 300+ Core customers → Cohort 2 (zh-Hant-first or bilingual) in Q2 2027.

10. Legal and ops basics (brief; confirm with a lawyer / accountant)

• Entity and banking: sell through a registered business (e.g. a US LLC, or your local equivalent) with a separate bank account; payment providers require business verification.
• Refund policy (Example): self-paced: 14-day refund if < 30% of lessons completed. Cohort: full refund until the end of week 1, 50% until the end of week 2, none after. Services: deposit (50%) non-refundable once work starts; milestone-based. EU / UK consumers have a 14-day withdrawal right for digital content unless they expressly consent to immediate access and acknowledge losing the right: add that checkbox / wording at checkout (your MoR may handle it, so check).
• Privacy policy: what you collect (email, name, progress, payment via processor, analytics / pixels), why, retention, processors (Cloudflare, Resend / Brevo / MailerLite, Lemon Squeezy / Stripe, Meta / Google / TikTok pixels), user rights (GDPR, CCPA/CPRA, Taiwan PDPA, HK PDPO), contact. Consent banner for EU / UK before ad pixels; double opt-in for email; unsubscribe in every email (CAN-SPAM / GDPR).
• Terms: single-user licence, no account sharing or redistribution of videos / materials, code samples under a stated licence (e.g. MIT for the public starter repo), no guarantee of business results, AI-output disclaimer, community conduct rules, service-agreement template for infra work (scope, access, security responsibilities, liability cap, data handling).
• Taxes: MoR (Lemon Squeezy / Stripe Managed Payments) collects and remits sales tax / VAT / GST on automated digital products. You handle tax on cohorts, workshops and services (human-delivered), plus income tax. Keep books from day 1 (export monthly from providers).
• Ads compliance: no income claims, disclose affiliate links, label AI-generated or AI-voiced creative where platforms require, and get written consent for testimonials.
• Third-party licences: GSAP is free under its standard licence (check its no-competing-tool clause); React Bits' Commons Clause bans reselling the components themselves; the KK store data is fictional (keep it that way, with no real customer PII in course material).

11. Risks and mitigations

12. Sources for verified prices (checked Oct 17, 2026)

Stripe pricing (stripe.com/pricing) · Stripe Managed Payments (stripe.com/managed-payments; docs eligibility page) · Cloudflare Workers / D1 / KV / R2 pricing (developers.cloudflare.com/workers/platform/pricing) · Cloudflare Pages limits (developers.cloudflare.com/pages/platform/limits) · Cloudflare Stream pricing (developers.cloudflare.com/stream/pricing) · Bunny Stream (bunny.net/pricing/stream) · Lemon Squeezy (lemonsqueezy.com/pricing) · Resend (resend.com/pricing) · MailerLite (mailerlite.com/pricing) · Brevo (brevo.com/pricing) · Teachable, Thinkific, Podia, Kajabi pricing pages · Maven Help Center ("Getting paid") · GSAP (gsap.com/pricing) · React Bits and Lenis licences (GitHub). Prices change. Re-check before committing.

Market benchmarks and examples (from Marketing's research, /workspace/miulabs-course-research/report.md, opened Oct 17, 2026): WordStream/LocaliQ Facebook Ads Benchmarks 2025 (Sept 2025) · Search Engine Land (Sept 8, 2025) · Metadata, what LinkedIn ads cost (Aug 25, 2026; 2025 data) · ON24 2025 Webinar Benchmarks (Feb 20, 2025) · Maven, effectiveness of one-day workshops (Feb 15, 2023) · Maven marketing guide (Jun 19, 2026) · Maven Lightning Lesson guide (Apr 14, 2026) · Nick Saraev Q&A archive · n8n affiliate program · 界面新闻 on Xiaohongshu 专栏 · The Coaching Times on refund data (vendor data only; not used for planning). Creator examples are linked in §4.3.