Published on Oct 23, 2023Naqcho AliRead time: 13m1.5K viewer
What Ecommerce Fraud Really Costs Merchants

What Ecommerce Fraud Really Costs Merchants

Introduction

A chargeback report is the worst place to learn what fraud costs you, because it is the last page of a story that started weeks earlier. By the time a disputed order lands there, the payment has already been captured, the goods may have shipped, the processor has taken its fee, your support team has answered two tickets about an order they cannot find, and a genuine customer may have been turned away at checkout by a rule that did not trust them.

This is why the amount stolen and the cost of fraud are two different numbers. Merchants tend to track the first one because it is easy to find on a dashboard, then budget against it and wonder why the fraud line never seems to shrink. The gap between loss and cost is where the money actually goes: fees, refunds, labour, declined revenue, processor scrutiny and the slower damage of a checkout that feels risky to good customers.

That gap is also measurable, which is the useful part. Once you know the lines that make up the bill, you can attack them one at a time instead of buying a tool and hoping the graph moves.

A survey of 569 fraud and risk executives across US and Canadian retail and ecommerce companies found that US merchants absorb an average of $4.61 in total cost for every $1 of fraud they lose, and Canadian merchants $4.52 (LexisNexis Risk Solutions, True Cost of Fraud Study, 2025).

TL;DR

Fraud cost and fraud loss are not the same figure, and the difference is usually three to four times the loss itself. The money leaves through fees, refunds, reshipments, manual review hours, processor penalties, false declines and lost repeat business. Tracking only chargebacks hides most of it.

  • The visible loss (goods and cash) is the smallest part of the bill for most stores.
  • False declines and manual review often cost more than the fraud they prevent, and neither appears on a fraud report.
  • A short model with your own order volume, average order value and dispute rate turns the argument into a number your finance team accepts.
  • Most of these costs fall at the same few moments: account creation, checkout and fulfilment. Score the order there, before money moves, and every downstream line shrinks together.

Why the Loss Amount Is Only the Entry Ticket

When a fraudster uses a stolen card, the store loses the goods and eventually the money. That is the headline number. Everything else is administrative fallout: the acquirer charges for the dispute, the payment processor records your ratio, someone investigates and replies to the customer, the warehouse ships a replacement, and the inventory is gone either way.

Then there is the second half that almost nobody counts, which is the orders you refused. A fraud rule that blocks 100 orders to stop 3 bad ones has not saved 97 orders. It has lost 97 sales, some of them from customers who will not come back. That cost is invisible because the order never existed, and it is the single most common reason fraud programmes look successful while revenue quietly flattens.

The practical consequence is that fraud prevention is not a cost centre to minimise. It is an approval decision to optimise, and the target is the total of both errors, not just the one that gets reported.

The Direct Costs That Show Up on the Invoice

Start with the lines finance can see. They are real, and they usually add up to more than the stolen amount.

Chargeback fees and representment

Every dispute carries a fee from the acquirer, and the fee is charged whether you win or lose. Fighting a chargeback costs staff time and evidence gathering: order logs, delivery proof, device data, customer communication. Many merchants win the dispute and still lose money on the case, which is why so many small stores simply accept chargebacks below a threshold value.

Refunds, reshipments and lost inventory

Refund fraud and friendly fraud end with money returning to a customer while the product stays with them. Physical goods are worse: the item is gone, the shipping cost is gone, and a replacement shipment may follow. Digital goods are worse still, because a delivered licence or credit cannot be clawed back at all.

Card testing and gateway costs

Card testing attacks push hundreds of small authorisations through a checkout. Most fail, but each attempt is a request your gateway processes and a signal your processor records. The direct cost is small per attempt and meaningful per week, and the pattern damages your approval rate with issuers.

The Cost of Being Too Careful: False Declines

A declined good customer is a full price sale lost plus the lifetime value behind it, and unlike a chargeback, nothing records it. The customer sees a generic error, assumes the problem is your store, and leaves.

False declines concentrate in exactly the places where aggressive rules are easiest to write: mismatched billing and shipping countries, privacy VPN users, customers on mobile networks that geolocate to a data centre, prepaid cards, and new accounts buying high value items. Each of those produces a plausible reason to block a real buyer.

The better pattern is to separate the signal from the decision. A privacy VPN is a weak signal. A privacy VPN plus a brand new account plus a high value order plus a shipping address that does not match the card country is a different conversation. Proxy and VPN Detection gives you the first half of that reasoning, and step up authentication gives you a way to keep the sale instead of refusing it.

The Invisible Bill: Manual Review and Support Load

Every order queued for review costs a human ten to twenty minutes, and the queue grows whenever a new rule fires too broadly. Review teams then make inconsistent decisions, because two analysts looking at the same order will weigh the signals differently.

Support load is the same cost wearing a different hat. Fraud generates tickets about failed payments, blocked accounts and missing refunds, and every one of those arrives with a customer who is already annoyed. Adding review capacity feels like a fix until you notice the queue grows as fast as the team does.

The way out is to reserve human attention for the small share of orders where a person genuinely adds judgement, and let scored decisions handle the rest. That is what Payment Fraud Analysis is for: it returns a risk read on the transaction so the queue only sees the cases that deserve it.

What Fraud Does to Your Processing Terms

Processors price risk, and a merchant with a rising dispute ratio gets priced accordingly. Above the card networks' monitoring thresholds, a store can be placed in a monitoring programme, charged higher fees, asked to hold a rolling reserve, or in the worst case lose the ability to accept cards at all. That last outcome ends the business, which makes this the cheapest cost to avoid and the most expensive to ignore.

This is also why a fraud programme that only reacts to chargebacks is structurally too late. By the time disputes are filed, the ratio is already moving. Preventing the disputed transaction is the only thing that keeps the ratio flat.

Where Fraud Hides in a Normal Looking Order

Fraudulent orders are not obviously fraudulent. They look like a new customer buying something popular with a card that authorises correctly. What separates them is usually a small mismatch between data points that should agree with each other.

  • Card country against IP country. A card issued in one country, charged from another, shipped to a third is common in stolen card fraud and rare in honest shopping.
  • Card type and order value. Prepaid and gift cards on high value electronics or easily resold goods is a pattern worth scoring, not blocking.
  • Email age and quality. Disposable domains and freshly created addresses show up disproportionately in first order fraud. User Data Scoring scores the address and the phone number at the point they are collected.
  • BIN range behaviour. The first six or eight digits carry issuer, country, card type and a prepaid flag. BIN Lookup turns those digits into a cheap, fast check that the card is what the order claims it is.
  • Network character. Hosting providers, Tor exits, and known proxy ranges behave differently from residential connections, and that difference is stable enough to score on.
  • Velocity. Three orders in ten minutes from three cards to one address is not a customer, it is a script.

None of these is proof on its own. Together they form a profile, and profiles are what a rules engine should be reading instead of single flags.

A Simple Model You Can Run on Your Own Numbers

The numbers below are an illustration with stated assumptions, not a benchmark. Replace them with your own and the arithmetic still holds.

Assume 10,000 orders a month at a $90 average order value, and 0.6 percent of orders later disputed as fraud.

  • Sixty disputed orders means $5,400 of goods and cash lost.
  • Sixty chargeback fees at $15 each adds $900.
  • Review and support time for those cases, at ten minutes each, adds roughly 10 hours of work.
  • A two percent false decline rate on remaining traffic refuses 200 genuine orders worth $18,000 in revenue. At a 30 percent contribution margin that is $5,400 of lost profit.

Total measurable cost lands near $11,700 against a reported loss of $5,400, or about 2.2 times the loss. The LexisNexis survey figure of $4.61 per dollar sits above that, which is expected: their sample includes chargeback heavy categories, larger operations and the cost of fraud teams themselves.

The point of the exercise is not precision. It is to show that the false decline line is as large as the fraud line in a typical store, which changes what you ask a vendor to optimise.

Scoring an Order Before You Capture It

The cheapest place to influence every cost above is the moment before authorisation, because nothing has been lost yet. Greip's v2 API returns IP, card and identity signals for a single order in one round trip.

Check the card against the order first:

curl --request GET \
  --url 'https://v2.greipapi.com/lookup/bin/457173' \
  --header 'Authorization: Bearer $GREIP_API_KEY'

Then score the transaction itself, server side:

const res = await fetch("https://v2.greipapi.com/score/payments", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.GREIP_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    data: {
      action: "purchase",
      transaction_amount: 149.0,
      transaction_currency: "USD",
      customer_ip: order.ip,
      customer_email: order.email,
      customer_country: order.billingCountry,
      shipping_country: order.shippingCountry,
      is_digital_products: false,
    },
  }),
});
const score = await res.json();
if (!score.success) throw new Error(JSON.stringify(score));

Keep the key on the server. Browsers should never see it, and the verdict should reach your front end as your own decision, not as a raw Greip response. The request reference for both calls is in the docs: Payment Fraud Detection and BIN Lookup.

Send the customer IP you actually observed, not the one the customer typed. Client supplied values defeat the comparison that makes the signal useful.

Reading the Signals Without Guessing

A score is only useful if it maps to an action. The usual pattern is three outcomes, not two: allow, challenge, block. Allow the large middle, challenge the uncertain band with 3D Secure or a verification step, and block only the narrow top where several signals agree.

Two rules keep that structure honest. First, never let a single anonymisation flag block an order on its own, or you will refuse every privacy conscious customer you have. Second, review the challenge band for outcomes, not just for declines, because a challenge that fails for technical reasons is a false decline wearing a different label.

Thresholds also drift. Fraud patterns move between quarters, new BIN ranges appear, hosting providers get abused and then cleaned up. A review every month on your own dispute and challenge data keeps the bands calibrated; setting them once and leaving them for a year is how programmes quietly rot.

Building the Business Case Internally

Fraud work competes for budget against features that generate revenue, and it loses that argument when it is framed as protecting money already lost. It wins when it is framed as recovered revenue.

Three numbers do most of the work: your dispute rate and the fees attached to it, your estimated false decline rate and the margin behind it, and the hours your team spends on manual review. Put them on one page, show the total, then show what a scored checkout at the point of sale would change. A store that recovers half its false declines usually finds the budget conversation ends quickly.

Finance will trust the model more if you state the assumptions and mark the estimates. A range with visible inputs beats a single confident number every time.

Measuring Fraud Cost Without Fooling Yourself

Three measurement habits separate programmes that improve from programmes that only report.

Track declined good orders explicitly. Sample blocks and challenges, contact a share of those customers, and count the ones who were legitimate. Without that number, every tightening of the rules looks like a win.

Count cost per order, not cost per month. Monthly totals move with volume and hide whether the programme is working. A cost per thousand orders makes two quarters comparable.

Separate fraud types. Chargeback fraud, friendly fraud, card testing and account takeover have different economics and different fixes, and keeping them in one bucket means the biggest line never gets its own owner.

FAQs

How much does ecommerce fraud cost a merchant?

More than the amount stolen. LexisNexis puts the US average at $4.61 of total cost per $1 of fraud lost, with fees, refunds, review labour and lost sales making up the rest. Smaller stores usually sit lower, but the multiplier is still well above one.

Which fraud costs do merchants usually miss?

False declines and manual review time. Neither appears on a fraud or chargeback report, and together they often exceed the direct loss. Support hours spent on fraud related tickets are the third blind spot.

Are false declines really part of fraud cost?

Yes. They are the cost of preventing fraud badly. A refused genuine order loses the sale, the margin, and often the customer, and it is invisible because nothing was ever created to measure.

How do I calculate my own fraud cost per order?

Add chargeback fees, refunded and reshipped goods, review hours valued at your labour rate, and the margin on estimated false declines. Divide the total by order volume. Keep the assumptions written down so the number can be revisited as volume changes.

Does fraud affect payment processing rates?

It can. Rising dispute ratios can trigger monitoring programmes, higher fees, rolling reserves and in serious cases loss of card acceptance. That is the most expensive outcome and the reason prevention beats recovery.

What signals should I check before capturing an order?

Card country against IP and shipping country, BIN details including prepaid status, email and phone quality, the network behind the IP, and order velocity. No single signal should decide alone; the combination is the decision.

Conclusion

Fraud is expensive in ways that do not appear on the fraud report, and the merchants who get it under control are the ones who measure the whole bill instead of the visible slice of it. Start with your own numbers, put a cost on the orders you refuse, and move the decision to the moment before capture, where nothing has been lost yet.

Greip scores the transaction from IP, card, email and phone data in one call, so the same signal set that stops a bad order can approve a good one that a blunt rule would have refused. Details are in the Payment Fraud Analysis overview and the documentation.

Get started

Start protecting your business today

Our service is trusted by thousands of businesses worldwide.

  • 1,000 requests during trial
  • Cancel anytime

Related Articles