The problem
Companies that buy leads from TikTok, Meta, Google and partner networks pay for everything that arrives: bots, duplicates, fake sign-ups and people who never meant to buy. A big share of the budget disappears before anyone in sales sees a lead.
The usual fix is a monthly spreadsheet that shows which source was bad, long after the money is gone. BackNova was built to make that decision in real time, on every single lead.
What we built
- A real-time decision endpoint: each click or lead is scored HIGH, MEDIUM or LOW in milliseconds and routed accordingly
- Anti-fraud signals: device fingerprinting, behaviour, velocity and anomaly checks, plus a risk score for every traffic source
- Shadow mode: score and log everything without blocking anything, so a new client can calibrate safely before switching filtering on
- Conversion feedback: the system learns which leads actually turned into customers and adjusts scores
- Analytics dashboard: sources, geography, approve rate, ROI and money saved, with backup routing, failover and a kill switch
- Watchdog alerts in Telegram when a source suddenly spikes or quality drops
- Full SaaS account layer: sign-up with email verification, password reset, team invites, admin panel, plan limits
- Stripe checkout, billing portal and webhooks, plus a bilingual marketing site with docs and a blog
How it's built
One edge API on Cloudflare Workers handles about 157 routes. Scores are computed close to the visitor, so routing adds almost no delay. D1 stores accounts and history; several KV stores keep rate limits, fingerprints and a fraud cache hot.
Scheduled jobs run the watchdog, rescore traffic in batches and send alerts. Dozens of database migrations kept the schema evolving without downtime.
Why it matters for you
This is the deep end of what we build: not a landing page, but a B2B platform with real-time logic, billing, teams and monitoring. A 7-day sprint gives you the first working slice of a product like this, on the same foundation.