Risk Engine

Every COD order scored
before it ships.

Cash on delivery fraud costs merchants in Iraq and the Levant $1,000—$24,000 every month. Fake orders, no-shows, customers who receive a package and claim they didn't. Rezva scores every order before the driver leaves — and flags the ones that shouldn't ship.

How It Works

Scored in seconds. Decided before it ships.

The customer never knows this is happening.

01
The order is confirmed.

After the customer confirms through the Order Engine, the Risk Engine triggers automatically. The customer sees nothing. The analysis begins.

02
23 signals run instantly.

Blacklisted phones and addresses. No-show history. Velocity abuse. Unusual order values. High-risk product categories. Conversation behavior. Payment receipt authenticity. Each signal adds to the score.

03
Behavioral analysis runs next.

Customer history, order patterns, delivery outcomes, and cross-merchant signals are analyzed. A final score from 0 to 100 is produced — with a reason.

04
A decision is made.

0—30: ships automatically. 31—60: merchant reviews via Telegram. 61—85: held, merchant alerted immediately. 86+: rejected. Customer gets a polite message. No confrontation. No explanation needed.

What It Catches

23 signals. Before the driver leaves.

Covering every fraud pattern common in Iraq and the Levant's COD market — including the ones that look like normal orders.

  • Blacklisted phone numbers
  • Invalid or undeliverable addresses
  • Repeat no-show customers
  • High order velocity (3+ orders in 24h)
  • Velocity spikes (5+ orders in 48h)
  • Order value 3— customer average
  • Prior disputes on record
  • Frequent address changes
  • New phone + high-value COD
  • High return rate above 50%
  • Trusted account sudden behavior change
  • Cross-merchant serial abusers
  • Network velocity — 5+ signals in 48h
  • Prepaid SIM + first order + high value
  • Odd-hour new customer orders
  • Customers who receive and deny delivery
  • High-resale category orders (electronics, phones)
  • Coordinated fraud ring patterns
  • Conversation pressure signals
  • Forged payment receipts (Vision AI)
  • Recycled payment screenshots
  • Cross-merchant hash matches
The Score

A number, a reason, and a decision.

0 — 30
Ships.

Auto-approved. The driver gets the package. No merchant action needed.

31 — 60
Review.

Merchant gets a Telegram card with the score, the flags, and a reason. One tap to approve or reject.

61 — 85
Hold.

Order held automatically. Merchant gets an immediate alert. Must confirm manually before anything ships.

86 — 100
Rejected.

Customer receives a polite decline via WhatsApp. Merchant is notified. Nothing ships. No confrontation.

Thresholds adjust automatically over time based on your store's actual delivery outcomes.

The Network

Every merchant makes every other merchant smarter.

When a fraud attempt is caught and reported on one store, the signal reaches every other Rezva merchant in real time — anonymously, with no personal data shared. A bad actor who hits Merchant A is flagged at Merchant B within seconds.

No individual merchant can build this alone. A bad actor changes their phone number and starts fresh with you. The network means they don't start fresh — they start flagged.

The network compounds. The more merchants join, the stronger protection becomes for everyone.

Payment Protection

It checks the receipt before you do.

For merchants who accept online transfers and bank payments, Rezva analyzes every payment screenshot before it reaches you. Forged receipts, recycled screenshots, and amount mismatches are flagged automatically.

What it checks

Whether the receipt format matches the actual layout of Zain Cash, FIB, OMT, or the claimed bank. Wrong fonts, wrong colors, and inconsistent styling signal editing.

Amount verification

The amount visible in the image is extracted and compared to the order total. A discrepancy of more than 5% triggers a flag before the merchant sees the screenshot.

Duplicate detection

Every receipt is fingerprinted. The same screenshot submitted twice — to the same store or a different one — is caught and flagged as a duplicate automatically.

From Day One

15 of 23 signals work
without any network.

The cross-merchant network amplifies the engine over time. But the most common fraud patterns in Iraq and the Levant — the ones that cost merchants money every week — are caught on day one.

No customer names or order details are ever shared. Only anonymized fraud signals — so every merchant benefits without exposing anyone's data.

Caught on day one

First-order high-value COD. Prepaid SIM with expensive item. Odd-hour new customer. Invalid or vague address. High-resale product category. Conversation pressure patterns. Forged payment receipts.

These seven patterns account for the majority of COD fraud any individual merchant faces.

Builds with your store

After the first few deliveries, the engine knows your customers. It tracks no-show patterns, return rates, address changes, order value history, and delivery outcomes.

Every confirmed outcome — delivered or not — teaches the engine more about what fraud looks like at your specific store.

Proven weekly

Every Sunday, a calibration report arrives by Telegram. Fraud attempts blocked. Good orders passed. Estimated savings in USD. False positive rate. Accuracy score.

You see exactly what the engine prevented — and whether it's making the right calls.

See it catch a fraud attempt live.

Every COD order scored before it ships.
From day one.

Start Free Trial See Live Demo