These six areas capture the shape of the work across all four initiatives — from having no structured signal at all, to shipping segmentation, return-cycle tracking, and intent-based mechanics that teams act on today.
When I joined FundedNext, the CFD product line had just two models on offer: Evaluation and Express. Shortly after, we started work on a New Model initiative within that product line. I worked directly alongside the Head of Business on model feasibility — stress-testing the variables that would define it: profit targets, drawdown limits, payout structure, pricing, and the trade-offs between them.
The original ambition was a fully dynamic system that adjusted itself per trader. That proved impractical to ship. Instead, we tested roughly 900 candidate variations of the model's parameters for feasibility and narrowed them down to two: Stellar 2-Step and Stellar 1-Step. Both have gone through many rounds of changes since first launch.
Three separate corners of FundedNext, three separate teams' worth of assumptions — and the same failure mode underneath each one: an aggregate number or a binary label was standing in for behavior that was actually far more varied underneath.
An 87% retention figure treated a week-one stacker, a six-month-dormant trader, and a monthly regular as identical — because the system had no way to tell them apart.
A drop-off was marked "churned" and written off, even though roughly a third of those customers were going to come back on their own — the model just wasn't built to expect it.
The FAQ page did exactly one job — answer questions — with no way to recognize or reward the readers who worked through it carefully enough to show real intent.
The aggregate number is a summary, not a description. All three projects find what it's flattening out, then act on that instead.
Classify every purchase by where the trader actually is, not a single "retained" flag.
Track every drop-off through its own cycle instead of writing it off at day 31.
Treat "read to the end" as a real signal — no analytics event captured it before.
The retention report read 87%. But it treated a week-one stacker, a six-month-dormant trader, and a monthly regular as identical — the system couldn't tell them apart.
"Ahmed bought 5 challenge accounts in his first week. Maria came back after 6 months of silence. Chen buys a new account every month like clockwork. How can I send them all the same email?"
Marketing Lead · Prop Trading PlatformNot "did they buy again?" but "where are they right now?" — each purchase classified by timing and trading activity between purchases.
Already funded and trading profitably. The platform's best outcome — not a churn risk.
Completely disengaged, then something brought them back. Every one is a proof point for what triggered the return.
| Action | Segment | Outcome | $ Impact |
|---|---|---|---|
| Switched new week-one stackers from a generic welcome to bundle-specific messaging on managing multiple accounts. | Acquisition Burst | +23% bundle uptake | +$4.1M / yr |
| Launched a VIP tier — early access, premium support — for the most loyal 15–30 day return cycle. | Monthly Retention | +8% retention rate | +$2.3M / yr |
| Stopped "we miss you" emails to traders who were still actively trading on funded accounts — they weren't lost. | Late Retention | −40% complaint rate | $310K saved |
| Built a separate offer structure for Gulf traders, whose average order value ran 38% above the platform mean. | Gulf Region | +22% Gulf revenue | +$1.6M / yr |
| Repositioned the cheaper challenge tier as the entry point for first-time buyers, with a natural path up. | First Purchase Cohort | +15% 1-Phase conversion | +$1.9M / yr |
A customer goes silent for 45 days — old model: "churned," moved to suppression. But she came back two months later and started trading again. She wasn't gone, she was on a break.
Most churn models assume the longer someone's gone, the less likely they return. The data said the opposite:
Returning once makes a customer more likely to return again — compounding loyalty, not volatility.
| Action | Mechanism | Est. Impact |
|---|---|---|
| Engagement prompts at Day 14, 21, and 30 | Reach at-risk customers before the 31-day drop threshold | $2.8M/yr retained |
| Targeted win-back campaigns at Day 35, 60, 90 | Timing-based outreach segmented by drop bucket | $3.1M/yr recovered |
| Multi-return loyalty program for 2+ cycle customers | Recognize and reward Boomerang customers explicitly | $1.6M/yr |
| Remove offer restrictions for returning customers | 52–54% withdrawal stability held across all cycles | $0.5M/yr |
| Super Boomerang VIP identification and treatment | Dedicated program for 3+ cycle customers | $0.3M/yr |
Most visitors skim a FAQ and leave. The ones who read it question-by-question are showing genuine intent — a signal no analytics event captures. FNHunt embedded a hidden discount code inside the answers: not bannered, just placed where only a careful reader would find it. Shipped with its own launch campaign — "The FAQ holds more than answers... now it holds rewards."
Both frameworks started as one-person analyses. The gap between "the segmentation is correct" and "the campaign team can run it daily" was wider than expected — a narrow single-segment pilot first would have proven the workflow faster than shipping the full category set upfront.
The aggregate number is never the whole story. The fastest way to find the real one is to go looking for the behavior the average is hiding.
The outcome numbers are visible above. Here is what went into deciding each one.