Black Friday is the week you most want to say yes to every order. This year, some of those orders will be placed by AI agents. Here’s how to keep the sales without having them disputed.

Adobe expects traffic from AI services to US retail sites to grow 130% this holiday season, with Black Friday alone expected to nearly double. Last year, during Cyber Week, AI and agents influenced 20% of all orders worldwide, according to Salesforce.

That’s good news for revenue. It also sharpens the oldest Black Friday dilemma: how do you approve as many orders as possible without those orders coming back as disputes?

Where Black Friday disputes come from

Most Black Friday disputes don’t involve a stolen card. They come from real customers: a charge they don’t recognise on their statement, a parcel stuck in peak-season delays, a gift sent to an address the cardholder didn’t expect, or plain buyer’s remorse.

In the Merchant Risk Council’s 2026 Global eCommerce Payments and Fraud Report, produced with Visa, 64% of merchants reported rising first-party misuse. A quarter of them saw it grow by 25% or more.

These disputes never show up on the Black Friday dashboard. Cardholders generally have up to 120 days to file, so by the time they arrive, the stock has shipped and the decision can’t be undone.

This year adds a new one to the list.

The new dispute: “my agent bought it”

Most AI shopping this season will still be assisted. A person asks a chatbot for options, clicks through and buys. That traffic is good news: in the same Adobe research, 69% of shoppers said they’re less likely to return something they bought with an AI assistant’s help.

Delegated purchasing is different. The person sets a goal and a budget; the agent finds the deal, picks the variant and checks out. Black Friday is exactly when that’s most tempting. Prices expire by the hour, the gift list is long, and nobody wants to refresh twelve tabs at midnight.

It also creates a claim that barely existed a year ago: my agent bought that, I didn’t. Sometimes it’s true. Agents pick the wrong size, stack a deal onto the wrong item, or buy from a store the person never named. Sometimes it isn’t. Either way it’s plausible, and the dispute system still asks only one question: did the cardholder authorise this purchase?

For a merchant, that’s the worst kind of claim. Legitimate often enough that you can’t dismiss it, unprovable often enough that you can’t win it.

Why a stricter checkout won’t save you

The instinctive response to a risky week is to tighten the rules. For Black Friday disputes, that backfires twice.

First, these disputes come from real customers using their own cards. Nothing looks wrong at checkout, because nothing is wrong yet.

Second, agents look exactly like what fraud filters were built to stop: fast sessions, no mouse movement, first-attempt success. Tighten the rules and you decline good agent orders in the one week you can least afford it. The disputes you were trying to avoid still come, from the orders you approved.

The protocols identify the agent, not the intent

The card networks have moved fast. Visa’s Trusted Agent Protocol lets merchants cryptographically recognise approved agents and tell them apart from malicious bots. Mastercard’s Verifiable Intent, built with Google, keeps a tamper-resistant record of what the cardholder authorised and what the agent was told. On October 9, Sierra published a draft of the Personal Agent Protocol, with Visa, Mastercard, PayPal and OpenAI among its design partners.

All of it is necessary, and all of it answers questions about the agent: who it is, who it represents, what instructions it carried.

None of it confirms that the person behind the card wants this order, as it’s about to ship. A trusted agent can still buy the wrong thing. A perfect record of the instructions shows the mismatch after the dispute. It doesn’t prevent it.

Approve the agent. Confirm the human.

The answer isn’t a stricter checkout. It’s a second decision, made later, with more information.

FUGU’s post-payment analysis approves the order at checkout and keeps evaluating it after the payment, across the order’s whole life. When real risk appears, FUGU verifies directly with the accountholder, before the order ships.

On Black Friday, that does three things.

You keep the sale. Orders are accepted at checkout, agent orders included. No filter reads machine speed as fraud, and verification only happens where the risk justifies it.

The dispute doesn’t happen. A cardholder who confirmed the order before it shipped has already answered the question a dispute would ask. “My agent did it” stops being an option once the person has said yes.

If one comes anyway, you can answer it. Your evidence is the cardholder’s own confirmation, not session data that describes an agent.

Before the traffic hits

1. Leave your fraud rules alone. Tightening them the week before Black Friday costs good orders, agent orders first.
2. Check your billing descriptor. A charge that doesn’t match your store name is one of the easiest disputes to prevent.
3. Decide which orders get confirmed before shipping. High-value carts, first-time buyers, agent-placed orders and late address changes are the usual candidates.
4. Keep delivery confirmation on every order. “It never arrived” is a peak-season classic.

Black Friday 2026 will be the first peak where a real share of orders is placed by software. The merchants who win it won’t block the agents or trust them blindly. They’ll approve the agent, confirm the human, and keep the sale.