Building a Full SaaS Platform in 3 Weeks with Vibe Coding
How a Brussels-based consultancy went from idea to paying customers in 21 days — without hiring a dev team.
The Client & The Problem
A Brussels-based HR consultancy (name withheld for confidentiality) had built a reputation for helping SMEs navigate Belgian labour law. Their secret weapon: a massive internal knowledge base of case law, tax rulings, and template documents, accumulated over 15 years.
The problem: only the senior partners could navigate it. Junior consultants wasted hours searching. Clients couldn't access it at all. And competitors were launching AI-powered "ask a lawyer" tools that felt like the future.
The founder's vision: turn their knowledge base into a SaaS product — a searchable, AI-assisted platform that clients could subscribe to for €49/month. Self-service for simple questions, escalation to a consultant for complex ones.
Traditional quote from a dev agency: €80,000 and 6 months. The founder called us instead.
What We Built
A multi-tenant SaaS platform with:
- AI-powered search: Users ask questions in plain FR/NL/EN ("Can I fire someone during probation in Belgium?") and the platform retrieves relevant documents, summarises them, and cites sources
- Document library: Filterable archive of templates, guides, case summaries — organised by topic and searchable
- User management: Company admins can invite team members, assign roles, and track usage
- Subscription & billing: Integrated with Stripe for monthly subscriptions (€49/user/month)
- Admin dashboard: For the consultancy to see usage stats, popular queries, and gaps in the knowledge base
- Escalation flow: If the AI can't answer confidently, it offers a "book a 15-min consult" CTA that goes straight into the consultancy's calendar
The Vibe Coding Process
Week 1: Scope & Prototype
Day 1–2: Discovery call and full-day scoping workshop. We walked through the entire user journey on a whiteboard, identified must-haves vs nice-to-haves, and agreed on a fixed scope.
Day 3–5: Built a working prototype. Users could search, get AI answers, and browse documents. No billing yet, no user management, but the core value prop was visible and testable.
Outcome: Founder showed the prototype to three existing clients. All three said "I'd pay for this." Green light to continue.
Week 2: Build & Iterate
Day 6–8: Added user authentication, company accounts, role-based access. Integrated Stripe for subscriptions.
Day 9–10: Built the admin dashboard so the consultancy could see what users were searching for and track unanswered queries (insight: gaps in their knowledge base became visible immediately).
Outcome: The founder could invite real beta users and charge them. First paying customer signed up on Day 11.
Week 3: Polish & Launch
Day 11–15: Refinement based on beta feedback. Added filters, improved search relevance, fixed edge cases. Built the "escalate to human" flow.
Day 16–18: Security hardening, GDPR compliance check, added terms of service and privacy policy.
Day 19–21: Soft launch to existing clients. Onboarding flow, welcome emails, help docs.
Outcome: Platform live. 12 paying companies (50+ individual users) by end of week 3.
The Tech Stack
We chose speed and simplicity over perfection:
- Frontend: React + TypeScript + Tailwind CSS
- Backend: Node.js + PostgreSQL hosted on Render
- AI: Claude 3.5 Sonnet (Anthropic) for answering queries, with RAG (retrieval-augmented generation) over the client's knowledge base
- Search: Embeddings-based semantic search (pgvector in Postgres)
- Auth: Supabase Auth
- Payments: Stripe Billing
- Hosting: EU region (Frankfurt) for GDPR compliance
What Made This Possible in 3 Weeks?
1. Vibe Coding (AI-Assisted Development)
We used Claude Code, Cursor, and GitHub Copilot to generate 70–80% of the boilerplate code (auth flows, database schemas, API endpoints). The human developer focused on architecture, business logic, and quality control. Speed multiplier: ~5x.
2. Clear, Fixed Scope
We said no to feature creep. The scoping workshop locked in the MVP. Everything else went on a "Phase 2" list. No scope, no speed.
3. Build-Test-Iterate Cycles
We showed working software to the client every 2–3 days. Feedback was immediate. No "big reveal at the end" — we course-corrected in real time.
4. Leverage Existing Tools
We didn't build authentication from scratch (Supabase). We didn't build a billing system (Stripe). We didn't host our own infrastructure (Render). Standing on the shoulders of giants.
5. The Client Had the Content
They already had 15 years of knowledge base material. We didn't have to create content — just structure it and make it searchable. If they'd started from zero, this would have taken months.
Results After 3 Months
- 120 paying users across 28 companies
- €5,880/month recurring revenue (120 × €49)
- 4,200+ searches performed — 78% answered by AI without human escalation
- 22% escalation rate — users who clicked "book a consult" (new revenue stream for the consultancy)
- NPS score: 72 (considered "excellent" for B2B SaaS)
- Churn: 4% (low for a new product)
What It Cost
| Scoping workshop (1 day) | €950 |
| Development (3 weeks, vibe coding) | €9,500 |
| Testing & launch support | €1,200 |
| Training & handover | €600 |
| Total project cost | €12,250 |
| Monthly running costs (hosting, APIs, Stripe fees) | ~€350/month |
ROI: At €5,880/month revenue and €350/month costs, the platform generates ~€5,500/month net. It paid for itself in 2.2 months. After one year, it's on track to generate €66k in profit.
What the Founder Said
"I thought I'd need six months and a full dev team. Fly AI showed me working software in week one. By week three we had paying customers. The speed completely changed my business model — instead of a big bet, this was a series of small, validated steps. And now we have a product that sells itself."
Lessons Learned
✅ What worked brilliantly
- Launching with beta users in week 2 meant we got real feedback before "launch day"
- The admin dashboard was originally "nice to have" — it became critical for understanding user behaviour
- Charging from day one validated that people would actually pay
- The "escalate to human" flow turned the SaaS into a lead-gen tool for high-value consulting work
⚠️ What we underestimated
- The knowledge base needed more structure than the client thought — we spent extra time cleaning and tagging documents
- Search relevance tuning took longer than expected (embeddings are powerful but need careful prompt engineering)
- Customer onboarding: we built the product fast, but didn't plan the "how do we teach users to use it" part until week 3
Could You Build a SaaS This Fast?
This approach works well when:
- You have a clear value prop that you can describe in one sentence
- The core feature is something AI does well (search, summarisation, classification, Q&A)
- You're willing to launch imperfect and iterate based on real usage
- You have domain expertise or content to build on (not starting from zero)
- You can commit to a tight, fixed scope (say no to "wouldn't it be cool if…")
What This Wasn't
Let's be clear: this was an MVP, not a finished product. After launch, the client added:
- Mobile app (built separately, 6 weeks later)
- Advanced analytics and reporting
- White-label option for larger enterprise clients
- Integrations with Slack and Microsoft Teams
Those features took another 3 months. But the core revenue-generating product — the thing that proved the business model — shipped in 3 weeks. That's the power of vibe coding: speed to validation.
Have a SaaS idea you want to test fast?
Book a free 30-minute discovery call. We'll help you figure out the smallest version of your idea that would prove whether people will pay for it — and we'll tell you honestly how long it would take to build.
Book a discovery call
