The Umrah booking system we shipped in 17 days

Madinah Travels handles Umrah bookings for a few thousand pilgrims a year out of Kuala Lumpur. When they called us, they had a Google Sheet, three WhatsApp inboxes, and a Ramadan-deadline stress response. They had six weeks until the rush. We had to ship.

The spec, in one paragraph

A booking platform that could take a pilgrim through package selection, room assignment, visa document upload, payment, and post-booking itinerary — in English and Bahasa — with a WhatsApp AI assistant that could answer 80% of pre-booking questions without a human.

What we shipped in 17 days

• Next.js on Vercel — booking flow, admin dashboard, payment (Stripe + local rail)
• WhatsApp AI assistant running on Claude with a curated package knowledge base
• Bahasa + English UI and conversation flows
• Automated itinerary generation post-booking
• Admin-side operator tooling to reassign rooms and manage visa docs

What mattered

Deciding what to cut. Visa automation was on the wishlist. We cut it. Shipping a working booking flow with manual visa review was better than a delayed launch with automated visa that had edge cases.

A one-page plan. Before we wrote code, we wrote a one-page spec. The founder signed it. When scope creep showed up — and it did, three times — we pointed at the page.

Weekly review. Every Thursday we shipped. Every Friday they reviewed. By week three we'd caught every major issue before it touched a customer.

What we'd do differently

Build the admin dashboard first. We built customer-facing flows first and rushed the operator tooling at the end. Next time, admin first — because operators use software hours a day, customers use it for five minutes.

Ship the AI assistant last, not in parallel. Running a product build and an AI build in parallel doubled the testing surface. If you have 17 days, ship the product. Ship the AI next sprint.

Invest in the seat recovery flow. When a pilgrim drops out, you need to re-sell their seat within hours. We shipped a manual flow. It should have been automated from day one.

Outcome

The platform launched 5 days before the Ramadan rush. Mobile conversion went from 1.1% (the old site) to 3.8% in the first month. The AI assistant handles 74% of WhatsApp inquiries without human review.

Not bad for 17 days — and a sign of what's possible when you cut the right things.