AI in Payments: Where AI Helps Online Businesses Today
Who this is for: payments, finance and operations teams at online businesses who want to know where AI actually helps today, and where it doesn't yet.
"AI in payments" covers everything from real tools in production to slides. The useful question is narrower: which payment jobs are slow, rule-bound or done by hand, and can AI do them well with people still in charge? Here is where it works today, with examples from Soap.
Where does AI help in payments today?
| Where | What AI does | At Soap |
|---|---|---|
| Recovering blocked payments | Strict rules block real customers, for example when the name on the ID is a nickname of the name on the card. AI can judge these near-matches the way a person would. | Soap's AI name matching recovered over 4.8% of blocked payments in a soft launch with tens of thousands of checks. Read more |
| Answering chargebacks | Winning a dispute means pulling evidence from many systems and writing a clear case. AI can gather it and draft the response. | Soap's AI agent builds a representment evidence report in a few clicks, and the merchant submits it. Read more |
| Payments made by AI agents | AI agents are starting to buy things on a person's behalf. That needs limits on what the agent may buy and proof of what the person allowed. | Our guide explains how agentic payments work and the guardrails they need. Read more |
| Building payment software | AI coding agents can write, test and review software changes in parallel, with people deciding what ships. | Soap builds its own platform this way, with feature flags and fast rollbacks. Read more |
Can AI recover payments that rules block?
Payment checks are full of strict rules, and strict rules block real customers. A customer whose ID says Amanda and whose card says Mandi fails an exact name match, even though any person would see they are the same. This is a good job for AI: a narrow question, a clear yes or no, and guardrails around it. Soap's AI name matching was built that way, guardrails first, and recovered over 4.8% of blocked payments that would otherwise have been lost.
Can AI fight chargebacks?
Most of the work in a chargeback is gathering evidence: payment logs, device data, identity checks and messages with the customer, often from different tools. AI is good at collecting it and writing it up. Soap's AI chargeback representment agent builds the evidence report in a few clicks, and the merchant reviews and submits it. For what to look for in a tool, read chargeback management software: what to look for.
What changes when AI agents make the payment?
AI agents are starting to shop on people's behalf. The payment itself looks familiar, but the questions are new: what was the agent allowed to buy, who approved it, and what happens if the person disputes it? Our guide to agentic payments covers how they work and the guardrails they need.
Where should people still make the call?
- Decisions about money and risk, such as approving a new merchant.
- Anything the AI is unsure about: it should say so and hand it to a person.
- Submitting a dispute response or a compliance filing.
- Changes to the rules the AI works within.
The pattern that works is narrow questions for the AI, fixed rules around its answers, a record of every decision, and people deciding what matters. It is also how Soap builds its own software.
Frequently asked questions
How is AI used in payments?
AI is used to judge cases that strict rules get wrong, like near-matching names, to gather evidence and draft chargeback responses, to help detect fraud, and increasingly to let AI agents make purchases on a person's behalf.
Can AI reduce declined or blocked payments?
It can help with payments blocked by rigid rules. For example, a strict name check rejects a customer whose ID says Amanda and whose card says Mandi. AI can recognize that as the same person. Soap's AI name matching recovered over 4.8% of payments that would otherwise have been blocked.
Can AI fight chargebacks?
AI can do the slow part: gathering payment logs, device data, identity checks and customer messages, and turning them into an evidence report. The merchant still reviews and submits the response, and the card issuer decides the outcome.
Is it safe to let AI make payment decisions?
It depends on the guardrails. Good systems give AI narrow, well-defined questions, check its answers against fixed rules, keep a record of every decision, and leave the final call on money and risk to people.
What are agentic payments?
Agentic payments are purchases made by an AI agent on a person's behalf, within limits the person sets. They raise new questions about authorization and disputes, because the person who pays isn't the one who clicked buy.
Talk to the Soap team
Want to see where AI fits in your payments? Talk to the Soap team.