A Commission Engine for 10,000 Direct-Sales Members

The challenge

The client runs a direct-sales network where every member's income depends on two things working correctly every single time: a commission calculation that follows a binary/matrix structure without drifting, and a payout system members can trust with real money. Off-the-shelf network-commerce software runs generic payout logic and treats fraud prevention as an afterthought — exactly the gap that costs a network real money once members work out how to game rank thresholds or self-refer their way into bonuses that were never meant to trigger. We designed the fraud model before writing a single payout rule, not after the first incident.

This case study is anonymized. Direct-sales and commission platforms carry strict confidentiality agreements, and the underlying business is sensitive to how it’s described publicly. We’re not naming the client — the engineering problem below is real.

The payout logic

Every member sits in a binary or matrix position under an upline, and every qualifying purchase or recruitment triggers a cascade of commission calculations up that structure — capped by rank, throttled by qualification rules, and required to settle correctly down to the cent before a payout batch runs. We built the calculation engine as its own service layer, decoupled from the storefront and the admin panel, so a catalog change or a promotion never has a side effect on how commissions get computed.

E-wallet and withdrawals

Every member holds a wallet balance instead of an immediate bank payout — commissions land in the wallet, then members request a withdrawal against a minimum threshold and a published fee. Requests above a configurable amount queue for review instead of auto-approving, and the wallet ledger is append-only: every credit and debit is its own row, never an update, so any balance can be reconstructed and audited straight from the transaction history.

Rank engine and fraud controls

Rank progression checks qualifying volume, team depth, and time-in-rank against rules the client’s compensation plan defines, recalculated on a schedule rather than by hand. On top of that sits the fraud layer: self-referral detection that flags accounts sharing a device fingerprint or payment method with their own upline, velocity limits on withdrawal requests, and an admin review queue for any account that trips a rule instead of a silent auto-approval.

Outcome

The platform has processed real-time commission calculations for a member base in the five figures, and every wallet balance still reconciles automatically against its append-only ledger on every payout cycle. We’ve stayed on as the platform’s engineering team on a retainer basis, extending the fraud rules as new abuse patterns show up.