Almost every merchant can name a number. One and a half percent. Two percent. Whatever their processor told them on the phone once.
Almost none of them are actually being held to that number, and the reason is the part nobody explains.
The thresholds, as published
Visa consolidated its old dispute and fraud monitoring programs into a single program, the Acquirer Monitoring Program, generally abbreviated VAMP. Two things changed that matter.
First, the counting. Fraud reports and non-fraud disputes used to be tracked by two separate programs with two separate thresholds. Under VAMP they are added together into one ratio against settled transaction count. A business that was comfortably inside both old programs can be outside the new one without a single thing changing in how it operates.
Second, the number came down. As of 1 April 2026 the merchant level Excessive threshold is reported at 1.5 percent in most regions, down from 2.2 percent, with a monthly floor of counted events, reported at 1,500, before a merchant is assessed at all [1][2]. A small number of regions were still on the older figure when this was written, which is the first reason to confirm your own region rather than trusting a blog, including this one.
These figures come from the published analysis available at the time of writing, and the program has been revised more than once in two years. Thresholds, regional coverage and enforcement dates all move. Ask your acquirer for the current fact sheet in writing, and ask which region your merchant identification number sits in. Those are the only two numbers that apply to you.
The part that catches people out
The merchant line is not the only line. The acquirer is being measured too, and on a different, much tighter scale.
Acquirer level thresholds are reported at 0.5 percent for Above Standard and 0.7 percent for Excessive, measured across the acquirer's entire portfolio, not per merchant [2][3].
Run the arithmetic from the bank's side and the behaviour stops looking arbitrary. A portfolio sitting near its own line has to pull its worst files down, and the only lever it has is the merchants. So the processor sets an internal limit well below the published merchant figure, and acts on it quietly. That internal limit is the number you are actually being held to, and it is almost never the number on the sales call.
This is also why a merchant can be shut down at 0.9 percent while a competitor sits at 1.2 percent somewhere else and hears nothing. Same network rules, different portfolio load. The question that decides your outcome is not what your ratio is. It is what your bank's ratio is.
What you can actually ask for
Three questions get you further than any amount of arguing after the fact:
- What is my internal threshold, in writing? Not the network number. The one your acquirer will act on. Many will tell you if you ask directly.
- What is in the count? Fraud reports and disputes combined, against which denominator, over which period. You cannot manage a ratio you cannot reproduce.
- At what point do I get a warning rather than a notice? A bank willing to answer that is a bank that intends to work with you.
What actually moves the ratio
Dispute response wins money back. It does not generally remove the event from the count, which is the thing merchants discover too late. Representment and ratio management are two different jobs.
What moves a ratio is earlier than that:
- Refund before the dispute exists. A refund issued before a customer calls their bank is not a dispute and never enters the ratio. This is the single cheapest control available and most merchants under-use it because it feels like losing.
- The billing descriptor. A descriptor that does not look like the brand the customer bought from generates "I do not recognise this charge" disputes out of perfectly good sales. It is a one line fix that takes a day.
- A cancellation path as findable as the sign-up. For anything billing on repeat, this is both a dispute control and, in the United States, a consumer protection requirement [4].
- Prevention alerts and inquiry resolution, where the network programs and your processor support them, which stop a case before it is counted rather than arguing it after.
- Delivery and usage evidence captured at the time, not reconstructed later. Which is getting harder, for reasons we wrote about in who pays the chargeback when an AI agent buys the wrong thing.
Why this is an underwriting story, not a support story
A ratio problem does not stay a ratio problem. It becomes a reserve, then a monitoring period, then a closure, then a re-underwriting conversation at the next bank with a monitoring history attached to the file. That last step is the expensive one, because it changes what every future bank sees when it reads you.
Which is the whole argument for knowing your bank's position and not just your own. If you are in a vertical that is watched, the right time to ask where your acquirer's portfolio sits is while your numbers are still good.
FR