Card-not-present fraud cost United States merchants $10.16 billion in 2024, roughly double the 2019 total, and it now makes up 74% of all card fraud. CBD sellers absorb more of that pressure than most retailers. The category already has higher dispute rates, and every fraudulent order that slips through pushes a high-risk account closer to the chargeback limit that can end it. Fraud detection software is the layer that decides how many of those orders get stopped before they post.
The software works as a set of checks that run in the moment between a buyer clicking pay and the transaction settling, each one scoring the order against patterns that signal theft.
Inside a Fraud Engine
A fraud engine reads several signals at once. It scores the order with a machine-learning model trained on past transactions, both clean and fraudulent, and flags anything that deviates from the normal pattern. It runs velocity checks that count how many attempts come from one card, device, or address in a short window, which catches the rapid-fire testing fraudsters use on stolen numbers.
Other checks confirm the buyer holds the physical card. The card verification value asks for the three digits on the back. Address verification compares the billing address entered at checkout against the one on file with the issuing bank. A 3-D Secure prompt hands authentication back to that bank for a final identity check. Device fingerprinting records the hardware and browser behind the order, so a single machine cycling through cards is easy to flag. None of these is decisive alone. Together they produce a risk score the processor acts on.
The score does more than approve or block. A middle band routes an order to manual review, where a person checks the details before the sale completes, and the threshold for that band is something the merchant tunes as it learns which orders become disputes. A tighter threshold stops more fraud while sending more orders to review, and a looser one approves faster while letting more slip through. The setting is a business decision the merchant owns.
The Tight Margin on a High-Risk Account
For an ordinary retailer, a missed fraudulent order is a loss. For a CBD seller, it is a step toward losing the account. Visa and Mastercard treat a dispute rate above roughly 1% as excessive, and CBD already operates near that ceiling. Each fraudulent sale that posts becomes a likely chargeback weeks later, and a cluster of them can trip the network monitoring program that forces a processor to drop the merchant.
Each disputed order also costs the seller roughly $110 in handling, labor, and lost goods, so a single missed fraud runs to more than the price of the sale.
Fraud tools change that math by stopping the orders that would have become disputes. A high-risk processor that boards CBD usually requires the merchant to run them for that reason. The screen protects the processor’s standing with the networks as much as the seller’s revenue, which is why it is built into the account rather than offered as an extra.
The Processor’s Part in the Screen
The merchant does not assemble these tools alone. A payment processor for CBD brings the scoring engine, the verification checks, and the device data together inside the same system that moves the money, so the screen runs on every order without a separate integration. The merchant sets the thresholds, and the processor supplies the infrastructure behind them.
That setup matters most for a small CBD store without a fraud team. The provider’s model has seen far more transactions than any single merchant, and the shared view of patterns across many accounts catches schemes a lone store would miss.
Evolving Fraud Tactics
The threats keep changing. Generative tools now let fraudsters build convincing fake storefronts and run online shopping scams that are harder for a buyer or a basic filter to spot. The same technology writes cleaner phishing messages and generates synthetic identities that pass a shallow check.
Detection has moved in step. The fraud surge is now driven partly by AI fraud on the attacker’s side, which forces the scoring models to retrain on new patterns faster than before. For a CBD seller, the lesson is that a static rule set ages quickly, and a processor whose model updates is worth more than one whose checks were set years ago.
Stolen Numbers and Card Testing
Most fraudulent CBD orders start with data the buyer never owned. Large breaches put millions of card numbers into circulation, a pattern that has run from the Target cybercrime case to the steady stream of leaks since. Fraudsters buy those numbers cheaply and test them on small online purchases to see which still work.
CBD storefronts are a common testing ground because the products are inexpensive and the checkout is quick. Velocity checks and device fingerprinting are the tools that catch a testing run, flagging the burst of small attempts before they turn into approved charges and later disputes.
False Declines and the Cost of Caution
A screen set too tight creates its own loss. False declines, where a legitimate order is rejected as fraud, cost merchants far more in lost sales than fraud itself does. By some measures a business loses several times as much to wrongly blocked orders as to the fraud it stops, and the annual cost of false declines runs into the hundreds of billions. A CBD seller that blocks too aggressively trades a fraud problem for a revenue problem.
The balance is the point of good software. The aim is to protect a small business from theft without turning away the real customers who drive its growth. A model that scores accurately blocks the fraud and approves the genuine order, and that accuracy is what separates a useful fraud system from a blunt one.
The Stakes of an Unscreened Account
The stakes are not limited to the value of a few stolen orders. For a high-risk merchant, unchecked fraud feeds the dispute rate, and a dispute rate above the network ceiling ends the account the business depends on. A CBD store that skips fraud detection leaves open the one failure that can take its payment processing away entirely. Replacing a processor after a forced closure is slower and costlier than running the screen would have been, and the software keeps the account alive, which is the outcome that matters most to a business built on card sales.