Most jewelry operators think chargebacks on large orders are bad luck. A stolen card, a clever thief, a cost of doing business. We think that is wrong. A large-order chargeback is usually preceded by a readable sequence of behavior, and the checkout score is the least informative part of it.
This matters because the math on a single loss is brutal. The goal of this piece is to show where the signals sit, what they cost to act on, and how to build a short review routine a small team can actually run.
What most operators believe
The common view has three parts. First, fraud is a card problem, so the processor's risk score should catch it. Second, a verified address and a matching CVV mean the order is safe. Third, friction kills conversion, so on high-ticket orders you should stay out of the buyer's way.
Each belief is partly true and mostly incomplete. Processor scores are built across every merchant category. Fine jewelry is unusual: high value, easy to resell, small enough to ship anywhere, and often bought in a hurry as a gift. A generic model sees an unfamiliar buyer spending $4,000 and has little basis to call it wrong. It also cannot see anything your shop knows, such as the conversation that happened before the order.
What we believe instead
Fraud on a large order is a process, not a moment. Someone has to choose the item, get it to an address they control, and do it before the real cardholder notices. That process leaves traces across at least four stages: the browse, the checkout, the fulfillment request, and the aftermath. A single stage proves little. Two or three stacking together is a pattern.
Notice also that there are two kinds of chargeback. One is true fraud, where the cardholder never made the purchase. The other is friendly fraud, where the real buyer disputes a legitimate order, often claiming non-delivery or that the piece was not as described. The signals differ, and your defenses differ too.
The signals, in the order they appear
Before checkout
- Narrow, price-first browsing. The visitor goes straight to your highest-priced items and sorts by price, with little interest in materials, sizing or story.
- No questions on a piece that normally prompts them. Real buyers of a $5,000 engagement ring ask about stone origin, resizing and timelines. Silence is not proof of fraud, but it is a data point.
- Pressure on speed. Messages asking for same-day dispatch, or asking you to skip normal steps.
At checkout
- Billing and shipping mismatch, especially a shipping address that is a freight forwarder, hotel or newly vacant property.
- Several card attempts in a short window, with different cards or repeated declines before one passes.
- Email patterns. A freshly created address with a random string, or a name that does not match the cardholder name.
- Location mismatch between the IP, the billing country and the shipping country.
- An order far above your usual basket. If your median order is $380 and this is $6,200 from a first-time buyer, treat it as an outlier by default.
After the order
- Requests to change the shipping address after payment.
- Requests to ship to a different name or to split the shipment.
- Offers to pay the difference by another method, such as a wire or a second card.
- Unusual urgency to cancel and re-place the order with different details.
The post-order stage is the most underused. Thieves often adjust details after the card clears because they have learned that sellers rarely recheck.
A worked example with stated assumptions
Take a hypothetical shop doing $2.4M a year. Assume the average order is $420, and about 3% of orders, or roughly 170 a year, exceed $2,000. Assume a gross margin of 55%, so the product cost on a $4,000 piece is $1,800. Assume a chargeback fee of $25.
If a $4,000 order is lost to fraud, the shop loses the $1,800 product, roughly $60 in shipping and insurance, the $25 fee and the lost revenue. Call it $1,885 in hard loss. To replace that profit at a 55% margin on a $420 average order, you need to sell about eight more average orders at roughly $231 of margin each, just to get back to even on that one event. That is the real price of a single miss.
Now assume a review step takes 10 minutes per high-value order, and the shop reviews all 170 of them. That is about 28 hours a year. At a loaded staff cost of $35 an hour, it is around $1,000. Suppose it delays 15% of those orders by a day and costs you a few sales, say three orders at $2,500 and a 55% margin, or about $4,100 in margin lost. Even then, if the routine prevents two fraud losses a year, it saves about $3,770 in hard losses. The trade is close, which is the point: review only where the signals stack, not on every order.
These figures are illustrative. Your own order mix, margin and dispute history will change the answer, so run the same arithmetic with your numbers.
A routine a small team can run
Build a simple score you can apply in two minutes. Give one point for each signal: mismatch, first-time buyer, forwarder address, price-first browsing, urgency, odd email. Set a threshold, for example three points on any order over $2,000, that triggers manual review.
When an order trips the threshold, take these steps in order:
- Hold fulfillment, not the order. Tell the buyer you are completing a standard verification for high-value purchases.
- Contact the buyer on a channel you control. Call the phone number on the card issuer's record if you can obtain it, or ask the buyer to confirm by a call-back. A real buyer will usually tolerate this.
- Ask for a small, specific confirmation, such as the last four digits of the card and the billing address as the bank holds it.
- Ship with signature required and full insurance, to the verified billing address where possible.
- Document everything. Save the messages, tracking, delivery signature and photos of the packaged item. You need this evidence to contest a dispute.
For friendly fraud, the defense is evidence. Clear product photos, written descriptions that match what you ship, a plain return policy and signed delivery cut off most disputes before they begin.
Where this breaks down
Be honest about the limits. Gift buyers legitimately ship to other addresses. International customers legitimately differ from your usual profile. Some honest buyers will be annoyed by a verification call. If your review step rejects too many real orders, the cure is worse than the disease.
Track two numbers monthly: the share of reviewed orders that turn out to be fraud, and the share of reviewed orders that cancel for any reason. If the first stays near zero and the second climbs, loosen the threshold. If chargebacks keep landing on orders you passed, tighten it.
The practical takeaway is simple. Stop treating the checkout score as the verdict. Pick a threshold for high-value orders, define the five or six signals you will count, hold fulfillment when they stack, and keep the evidence for every shipment. That routine costs a few hours a month, and it turns an unpredictable loss into a managed one.