Screening checkout for fake orders — and the line where it stops

Most order-screening advice tells you what to check and stops before the part you actually need, which is where the checking runs out. For e-commerce that boundary arrives unusually early, so it is worth starting there.

What this cannot answer#

Whether the shipping address is real. Boundstone validates email addresses, phone numbers and IP addresses. It does not verify postal addresses, does not confirm a street exists, and does not check whether the delivery address matches the cardholder. If your fraud problem is parcels going to reshippers, that is an address-verification product and this is not one.

Whether the card is stolen. Not our layer, and never will be.

Whether the buyer is behind a VPN. The IP endpoint returns format, version and range classification. geolocation, asn, hosting_datacenter, proxy_vpn_tor and reputation all come back in checks.not_performed, because they need licensed data we do not ship.

Who the device belongs to. No fingerprinting, no behavioural scoring.

That is four of the six questions most order-fraud tooling is bought for. If those are your problem, buy a fraud platform and come back for the hygiene layer afterwards.

What it can answer#

Three signals, each real and each narrow.

The email is a throwaway. disposable: true means the domain is on a maintained list of throwaway providers. On a physical-goods order, a burner inbox is a genuinely odd choice — the buyer normally wants the tracking number. This is the strongest of the three for e-commerce, precisely because the innocent explanation is weaker here than it is at a SaaS signup.

The phone is a designated VoIP range. line_type: "voip" on an order where you will send delivery updates is worth noting. Metadata catches designated ranges, not a VoIP service that has ported onto a mobile range; resolving that needs a live dip with hlr:true at 5 credits on a paid plan.

The source IP is not a plausible public address. A private, loopback or reserved classification on a public checkout means something is misconfigured or forged. It is rare, and when it appears it is worth acting on.

Combine, then route#

None of these is a verdict on its own. A privacy-conscious buyer with a throwaway inbox and a VoIP number is a real customer, and blocking them costs you a sale you never learn about.

The useful shape is the same one that works at signup: count independent signals and let the count pick the response. One signal is worth logging. Two is worth a review before fulfilment — which for physical goods is a genuinely cheap intervention, because the parcel has not moved yet. Three, on a high-value order, is worth holding.

The advantage e-commerce has over SaaS here is time. You are not deciding in the 200 milliseconds before a page renders; you are deciding before a picker touches a shelf. Use it — a review queue that runs hourly is entirely sufficient, and it makes hard automatic blocks unnecessary.

Where the honest boundary helps you#

A screening rule you cannot explain is a rule you will eventually disable during a busy week. Because every Boundstone response names what it did and did not check, the rule you write is auditable: this order was held because the address was on a disposable list and the line type was VoIP — not because an opaque score said 0.82.

When a customer disputes a hold, that is a sentence you can say out loud.

The short version#

• Start with the boundary. No postal address verification, no card checks, no VPN or proxy detection, no device fingerprinting — four of the six questions order-fraud tooling is usually bought for.
• If parcels to reshippers are the problem, you need an address-verification product, not this one.
• Three real signals: disposable email, designated-VoIP line type, and a source IP classifying as private, loopback or reserved.
• Disposable email is the strongest for physical goods — a buyer expecting a parcel usually wants the tracking number.
• Count signals, then route. One is a log, two is a review before fulfilment, three on a high-value order is a hold.
• E-commerce has time that SaaS does not. Deciding before the picker moves means an hourly review queue beats an automatic block.
• An explainable hold survives a dispute. "Disposable domain plus VoIP line" is a sentence; "score 0.82" is not.