3wks
idea to full platform launch
AI-Personalised Flight Deal Platform Shipped in 3 Weeks for First Class Flyer
First Class Flyer
3wks
Build and launch
5,000+
Members in month one
4×
Alert open rate vs. newsletter
6×
Subscriber growth in 12 months
Context
The business context
First Class Flyer started as a passion project - a curated newsletter helping frequent flyers find premium cabin deals on points and miles. It built a loyal following fast. But as the subscriber list grew, the manual newsletter model broke down. Every subscriber got the same content, regardless of whether they flew out of London or Mumbai, preferred business class or first, or used Avios versus Chase points. The founder knew the product needed to become a platform. The challenge was getting there without a six-month build that would lose momentum.
The problem
5 specific problems that needed solving
Manual newsletter sent the same deals to all subscribers - a London subscriber receiving Dubai-to-Singapore deals had no use for them
Monitoring fare databases and award space manually was consuming 15+ hours per week of the founder's time
No scalable alerting mechanism: as the subscriber list grew, the manual process became increasingly unsustainable
Zero member data on preferences - no way to personalise even if the infrastructure existed
Previous agency had quoted 6 months and £120k for a 'proper platform' - the founder needed a faster path to market
Our approach
Ship something real in three weeks, then iterate.
The founder's biggest risk wasn't technical - it was losing momentum. A six-month build with nothing to show subscribers would have cost email open rates, subscriber retention, and the founder's own conviction. We ran a rapid scoping session on day one to identify the smallest version of the platform that would be genuinely better than the newsletter - not just technically, but from a member experience perspective. We agreed: the core product was personalised alerts, not a rich web app. Build the preference capture and alert engine first. Build the web feed after subscribers were already using and trusting the alerts.
Onboarding-first design: members could set detailed preferences in 3 minutes - the experience that mattered most in week one
Push before pull: alert notifications shipped before the web deal feed, because pull behaviour takes longer to establish than push engagement
AI preference matching rather than rigid filters - a member who prefers business class but sets 'first class if under 80k Avios' gets that match, not just an exact filter hit
Infrastructure built for iteration: Supabase + Vercel meant the founder could add new alert types without engineering support
What we built
A personalised deal intelligence platform in three weeks
The platform is a full-stack Next.js application with a Supabase backend, OpenAI-powered preference matching, and a Resend-based alert delivery system. Members complete a preference onboarding flow capturing origin airports, destination preferences, cabin class, flexibility windows, and points programme accounts. An automated monitoring pipeline checks fare databases and loyalty award availability every 4 hours, runs each new deal through the preference matching model, and triggers personalised email and push alerts for matching members. The web deal feed shows each member a personalised list of currently available deals ranked by match score.
Preference onboarding flow
A 6-step onboarding flow captures origin airports, destination preferences (specific routes or destination types), cabin class preferences, travel flexibility windows, points programme accounts, and minimum points-per-pound thresholds. Takes under 3 minutes to complete. Stored in Supabase with row-level security per member.
Deal monitoring pipeline
Scheduled Vercel cron jobs run every 4 hours, querying fare databases and award availability APIs for new deals. Each new deal is structured into a canonical schema before preference matching - ensuring consistent data quality regardless of source.
AI preference matching
GPT-4o evaluates each new deal against each member's preference profile, scoring the match on route relevance, cabin class, points programme compatibility, and value. Members get alerts only when the match score exceeds their configured threshold - preventing alert fatigue.
Email and push alert delivery
Resend handles transactional email delivery with dynamic personalisation - each alert email shows the specific match reason ('This matches your London → Bangkok business class preference using your Avios balance'). Web push notifications deliver for mobile members who prefer instant alerts.
Personalised deal feed
A real-time web feed shows each logged-in member their current personalised deal list, sorted by match score and value. Members can save deals, share with other members, and adjust preferences in-line without returning to the onboarding flow.
Impact
What changed in production
Three weeks after launch, the platform was serving 5,000 members with personalised alerts that were materially more relevant than anything the newsletter had delivered.
Launched in 3 weeks. 5,000+ members onboarded month one. Deal alert open rate 4× higher than the manual newsletter. Subscriber base grew 6× in 12 months.
3wks
Build and launch
5,000+
Members in month one
4×
Alert open rate vs. newsletter
6×
Subscriber growth in 12 months
“Thinkscoop shipped something in three weeks that I'd been trying to spec for six months. It still runs flawlessly a year later. My subscribers love it and our subscriber base has grown 6× since launch.”
Founder
Founder - First Class Flyer
Learnings
What we took away from this project
Preference capture is a product in itself
The onboarding flow design took more iteration than any technical component. Too many preference fields and members abandon. Too few and the personalisation isn't meaningful. We landed on six preference dimensions after testing with ten members before launch - enough to make alerts genuinely relevant without the cognitive load of a 20-step setup. The insight: personalisation quality is bottlenecked by onboarding completion rate, not matching algorithm sophistication.
Alert fatigue is a silent churn driver
Our first version sent alerts whenever a deal exceeded a base match score. Within two weeks, some members had received 30+ alerts and were starting to unsubscribe. We added a configurable daily alert cap (default: 3 per day) and a 'snooze' feature for members who'd recently travelled. Unsubscribe rate dropped 60% after this change. The lesson: in alert-based products, sending fewer, higher-quality alerts beats volume every time.
The founder's continued involvement was the platform's best feature
First Class Flyer had a genuine community built on trust in the founder's curation. We made sure the platform amplified that, not replaced it. The deal feed includes founder commentary on the best current deals - a human editorial layer on top of AI matching. This combination of AI personalisation and human curation is what drove the 4× open rate improvement over the fully manual newsletter.
3wks
idea to full platform launch
At a glance
Tech stack
Capabilities
Build something similar?
We've solved this category of problem before. Let's scope yours.
Start a conversation View related serviceMore work
