Merchant underwriting used to be a one time human decision. Someone reviewed an application, checked a few boxes, made a call, and moved on to the next file. Machine learning is replacing that with something structurally different: a system that keeps reading a business’s actual transaction data for as long as the account stays open, not just once at signup.
That difference sounds small. It isn’t. A human reviewer physically cannot watch a merchant’s chargeback rate, refund pattern, and customer behavior every single day for years on end. A model can. It scores that data continuously, pulling in transaction velocity, refund concentration, device fingerprints, IP consistency, and dispute history, and flags changes the moment they appear instead of waiting for a monthly statement to reveal a problem after the exposure already happened. Industry research puts the share of payments risk teams already using AI somewhere in their underwriting process at close to two thirds, and that number reflects how far the old manual, one time review model has fallen behind what the job now requires.
Key Takeaways
- Merchant underwriting has evolved from a one-time decision to a continuous monitoring system using machine learning.
- This shift allows models to analyze real transaction data over time, providing a more accurate risk assessment.
- The old model failed many businesses due to reliance on category labels rather than actual performance metrics.
- Continuous underwriting enables fair evaluations of businesses in ‘blacklisted’ categories based on their current behaviors.
- High-risk underwriting is still not automatic; there’s a need for manual reviews in edge cases and transaction history remains essential.
Table of contents
Why the old model kept failing the same businesses
A one time review has one real weakness: it leans on shortcuts, because a single person doesn’t have the time or the tools to do anything else. The fastest shortcut is the category label. Subscription businesses, high ticket ecommerce sellers, nutraceutical brands, and direct sales companies have all been declined this way for years, not because of their actual numbers, but because their category carries a reputation, and a reviewer checking a box has no practical way to look past it before moving to the next application.
Continuous, data driven underwriting removes that shortcut entirely. Instead of one decision made off a label, the system is reading real behavior: chargeback rate, dispute pattern, transaction velocity, and customer complaint volume, all tracked over time rather than glanced at once. A business with a clean 1.5 percent dispute rate and one running at 5 percent stop getting the same answer, because the model is actually measuring the difference instead of grouping both under a single category stamp. That also changes what happens after approval. Instead of a merchant account that only gets revisited when something goes wrong, the account is monitored on an ongoing basis, so a shift in refund volume or a spike in complaints gets caught in days, not discovered six months later during a portfolio review.
What this looks like for network marketing specifically
Direct selling is one of the clearer examples of a category that got the blanket treatment. A network marketing company with a healthy distributor base and a normal dispute rate could still get declined by a standard acquirer running the old checklist model, simply because the word “MLM” ended the review before any transaction data got looked at. The company’s actual chargeback history, refund rate, and customer complaint volume never entered the conversation, because the category label made the decision first.
That is the exact problem a working MLM Merchant Account is built to solve: underwriting based on what a direct selling business actually does month over month, tracked continuously, rather than a single decision made once off the reputation of the business model. A distributor network with clean numbers gets treated like a business with clean numbers, not like the worst version of its industry.
Where the limits still are
None of this makes high risk underwriting automatic. A business with almost no transaction history still gives a model very little to work with, since there isn’t enough real behavior yet to score. Reserves and manual review for genuine edge cases aren’t going anywhere either, and they shouldn’t. What has changed is the default starting point. Underwriting used to start and end with the category label on the application. Now it can start with the numbers a business actually produces, which is the real mechanism letting well run companies in previously blacklisted categories get judged on their own performance instead of someone else’s history.
There’s also a documentation benefit that tends to get overlooked. A decision made off a category label is hard to defend later if a card network or regulator asks why a merchant was approved or declined. A decision built on tracked transaction data comes with its own paper trail: the actual chargeback rate, the actual dispute pattern, the actual behavior that justified the call. As card brand rules keep tightening, that record matters almost as much as the decision itself, because the businesses and the processors backing them both need to be able to show their work, not just state their conclusion.
