Every incoming order is scored in real time on mismatch signals — billing versus shipping, velocity, device, history. High-risk orders wait behind a one-tap review card while the other 96% flow straight to fulfillment. Fraud losses dropped 72%.
Typical results
96%
Orders sail through untouched
2 min
To clear a flagged order
0.25%
Chargeback rate, from 0.9%
Figures reflect typical results for this kind of automation.
The challenge
Limited-edition sneakers attract fraudsters, and the shop was losing thousands of shekels a month to stolen cards and chargebacks. The defense was manual: every order above a threshold — 15% of them — waited a day for someone to eyeball it, slowing honest hype buyers while sophisticated fraud slipped through anyway. Chargebacks hit 0.9% of orders.
Before and after
Chargeback rate
−72%Orders held for manual review
−73%Decision time on flagged orders
−99%The automation
We built a scoring layer that grades every order the moment it lands: billing-versus-shipping mismatch, address and device velocity, disposable email domains, order value against the customer's history. Low-risk orders release to fulfillment instantly. High scorers are held with a one-tap review card on WhatsApp showing the exact signals that fired, so a manager approves or cancels in about two minutes — and every decision feeds back to sharpen the rules.
How it works
- 01
Order scored
Each incoming order is graded instantly on mismatch, velocity, device and history signals.
- 02
Good orders released
96% score low and flow straight to fulfillment with zero added delay.
- 03
Risky orders held
High scorers pause before the warehouse ever sees them.
- 04
One-tap review
A WhatsApp card shows the signals that fired; the manager approves or cancels in a tap.
- 05
Rules learn
Every decision feeds back into the scoring so the net keeps tightening.
Starts from
- Incoming orders
- History & device signals
Delivers
- Instant release to fulfillment
- One-tap review cards
- Decision feedback loop
The impact
Fraud losses typically fall by about 72%, and the chargeback rate drops from 0.9% to around 0.25%. Only about 4% of orders pause for review, down from 15%, flagged orders are decided in two minutes instead of a day, and honest sneakerheads ship same-day.
Questions about this automation
Which signals are used to score an order?
Common signals include a billing and shipping address mismatch, several orders from the same device or address in a short time, disposable email domains, order value far above the customer's history, and first-time buyers ordering high-demand items. Each signal adds to a score, and the release and hold thresholds are set with the store.
Does screening slow down normal orders?
Scoring runs within seconds of checkout, and low-risk orders go straight to fulfillment with no added delay. Only orders above the hold threshold pause, and a manager receives a WhatsApp card listing the signals that fired, with buttons to approve or cancel. Approved orders are released to the warehouse immediately.
How do the rules improve over time?
Every review decision and every chargeback is recorded against the signals that fired on that order. A periodic report shows which signals catch real fraud and which mostly flag honest buyers, so thresholds and weights can be adjusted. This is a multi-step build, usually rolled out in stages, starting in monitor-only mode.