Anti-fraud system in TPP Ops: scoring and automatic quarantine

Loyalty programs with physical gifts combine aggressive promotions, point wallets, and open checkout for the player. This mix attracts abuse: reused coupons, zero-cost orders with real value in the cart, mule accounts, or anomalous purchase speed. TPP Ops responds with a second line of defense: a scoring engine that evaluates each order at the moment of creation and decides whether to proceed, enter quarantine, or block it.

It does not replace pricing controls at checkout — that remains the first barrier. What it does is detect suspicious patterns after the order exists, but before the product is shipped. For operations and support teams, it means fewer surprises in logistics and a clear flow when something smells fishy.

Three risk bands

The pipeline assigns a legitimacy score from 0 to 100, where 100 is a clean order. From 80 onwards, the order goes into the normal flow. Between 50 and 79, it remains in quarantine for manual review: a fraud incident is created, and the team decides whether to release it or confirm it. Below 50, the system only blocks automatically — and restricts the customer — when there are hard signs of economic fraud; soft rules, even if they accumulate, keep the order under review without penalizing a possibly legitimate buyer.

• Clear (≥ 80): standard flow, no retention.

• Review (50–79): quarantine + incident for support.

• Block (< 50): only automatic with hard signal; customer restricted until human review.

What signals does the engine evaluate

Each rule derives from real fraud vectors in gambling operators: promo abuse, wallets, and identity. The catalog grows through independent files — adding a rule does not require touching the review interface.

• Economic (hard): zero-cost order with value in cart, points without wallet debit, coupon with exhausted uses.

• Discounts and speed (soft): excessive discount, multiple orders from the same email in a short window, cart at maximum with promo.

• Address and identity (soft): same address across multiple accounts, disposable emails, recipient different from the account holder, customer on watchlist.

• Network and device (soft): IP collision between accounts, shared device fingerprint — always as corroboration, never as the sole cause for blocking.

Quarantine and operational review

A held order does not advance in logistics: while it is in review or blocked, any state change towards shipping is rejected. Support sees the score, the rules that triggered it, and the associated evidence. They can release a false positive — the order resumes its course — or confirm fraud — order canceled and customer banned —. Everything is traced with notes and reviewing user.

Restricted or banned customers cannot make further purchases: rejection occurs before creating a new order, both in the store checkout and in the public API. The system fails closed: if scoring fails within the transaction, the order is not confirmed.

Brand configuration

Each brand can activate or deactivate the fraud module and opt-out of specific soft rules. Hard rules — likely economic fraud — always remain active when the switch is on. This allows adapting sensitivity to different markets without losing protection on the most severe vectors.

TPP Ops' anti-fraud is designed for teams that already operate loyalty with physical delivery and need visibility before the warehouse, not just alerts afterward. If you manage promos, points, and public checkout, it is the layer that closes the cycle between promotion and fulfillment.

Where it applies

Scoring runs at the store checkout — every order that enters through the public surface of the storefront — and in the v1 API for order reservations. In both cases, the evaluation occurs within the same transaction: either the order is born with its risk verdict or it does not. Network signals (IP, user-agent, device fingerprint) are captured server-side from the headers, without relying on scripts in the buyer's browser.

For operations, the result is a single place to look at suspicious orders, with linked fraud area incidents and a history of signals per order. Less parallel Excel, more decision-making with context.