The Merchant Risk Council surveyed its members in December 2025 and published the results in its 2026 Global Ecommerce Payments & Fraud Report. There were 1,278 responses from 37 countries. Of those, 29% named minimising operational costs as their top fraud-management priority. Two years earlier, that figure was 10%. Practical Ecommerce covered the findings on 15 November 2026.

The article also says resolving a first-party dispute costs merchants more than $80 on average. And 62% of merchants said the time spent on fraud management is a moderate or major factor in future investment. More respondents also reported false positives than a year before.

The claim: cost now matters more than loss

The headline reading is that nearly one in three fraud professionals would rather cut cost than avoid loss. That sounds alarming, and it may be. I would first ask what the cost is. The article lists several. There are detection software fees, manual order reviews, investigations and staff time. There are chargebacks, which cost money even when the merchant wins. And there are false positives, where a good customer gets turned away.

The article gives a simple example. Say the average order is $100 and the margin is $30. If you wrongly reject 50 legitimate orders, you lose $1,500 in margin. A tool that stopped $1,000 of fraud has then cost you more than it saved.

So the 29% is not automatically a bad sign. Some of those people may have looked at their numbers and found that their screening costs more than the fraud it catches. What worries me is the opposite case, where cost-cutting is the default and nobody has done the sums.

Fraud is where a wrong call is expensive

My view on automation is that rules belong where being wrong is costly, and AI belongs where judgement is needed. Fraud fits both halves.

Some limits should never depend on a model’s mood that day. A refund ceiling, a cap on order value before manual review, a block after a set number of refund claims from one customer. These are money and promises. The system should enforce them every time, the same way. The rule lives in the platform, not in a prompt.

Other cases are not clear-cut. An order that looks slightly odd, a customer with an unusual pattern, a refund request that does not match the delivery record. A fixed rule either catches too much or too little here. This is where AI helps, by flagging the odd case. Then a person looks at it and decides. The AI narrows the pile. The person makes the call on the ones that matter.

This also speaks to the false positive problem. If every unusual order is blocked automatically, you pay in lost margin, as the article’s example shows. If unusual orders go to a person with the order details in front of them, you only spend review time on the cases worth it.

What this looks like for a small shop

Picture an online shop with a few hundred orders a week. The owner is tempted to turn on the strictest setting of a fraud tool, because a chargeback hurt last month. Or the owner is tempted to switch screening off, because the tool’s fees feel like a tax.

The article suggests a middle path, and I agree with it. Review only high-dollar or unusually risky orders instead of every order. Apply stricter refund checks to customers with repeated claims, not to everyone. Both are rules you can write down once. The system then applies them without anyone remembering to.

Then the owner needs to measure. The article recommends counting all of it: fraud losses, prevention spending, dispute costs, false positives and customer friction. Put them side by side for a month. If prevention costs more than the losses it stops, loosen it. If losses are climbing and the review queue is small, tighten it. Neither “spend more” nor “spend less” is a sensible default.

Where this connects to what I build

This is the same thinking behind AeroChat. The merchant sets the limits, such as how much it can refund and when a case goes to the team, and the platform enforces them. Anything beyond those limits goes to a person with the order details attached. I would apply the same split to fraud checks inside a customer conversation.

What a business should do

Write down your hard limits first. Decide the refund ceiling, the order value that triggers a manual look, and the number of repeat claims that triggers stricter checks. Put these in the system as rules.

Use AI or a scoring tool only to flag the odd cases, and keep a named person responsible for the final call on those.

Once a month, add up fraud losses, tool fees, dispute costs, review hours and the margin lost to wrongly rejected orders. Compare prevention cost with actual loss, and adjust from there.

Source: Practical Ecommerce: Is Fraud Prevention Worth the Cost?