Where AI is used in the payments industry today
| Area | What AI does | Maturity |
|---|---|---|
| Fraud and risk | Scores transactions and accounts in real time | Mature, many vendors |
| Routing and acceptance | Picks the provider or route most likely to approve | Common at orchestrators |
| KYC and onboarding | Checks documents and identities | Mature, specialised vendors |
| Agentic commerce | AI agents that pay on a user's behalf | Early; a different topic from this guide |
| Support and operations | Resolves requests about stuck, disputed and missing payments | Least automated — mostly people |
The first three areas have products you can buy off the shelf. Support and operations usually still run on people reading tickets, checking several systems by hand and writing “we're checking” while they do. That is where an AI agent helps most.
The operations problem: requests about money that can't wait
A PSP serving merchants — casinos, brokers, marketplaces — gets the same few request types again and again. Each one needs a look into processing, the bank and sometimes the orchestrator before anyone can answer, and the merchant's own customer is waiting on the other end.
- a payout stuck at the partner bank;
- a deposit debited but not credited, with a receipt as proof;
- a method failing for many clients at once;
- a report that says “paid” when the money never arrived.
Four use cases for AI agents at a payment provider
1. Stuck payouts: status, reason and ETA in seconds
The agent finds the operation, works out whose side it is stuck on and why, and answers the merchant right away. If the bank needs a nudge, it prepares the request for a manager to confirm, then holds the deadline: reminds the bank, updates the ETA, escalates with the full thread if the deadline passes.
2. Disputed deposits and receipts
The agent reads the receipt and matches it against the request and the bank statement. Unreadable digits are never guessed — it asks for the original. A receipt that doesn't match is never credited; an edited or borrowed one stops the credit and goes to the anti-fraud team with a breakdown. If the money left but the provider shows the request as expired, the agent collects the evidence and prepares a claim.
3. Mass failures: one incident instead of twenty tickets
Similar requests arriving in a burst are grouped into one incident: which method, since when, who is affected. The tech team gets the picture, affected clients get a status and a workaround — before they hear about it from their own players.
4. Reconciliation: a transfer ID, not “we're checking”
When a client's report doesn't match, the agent reconciles the operations against the payout register, finds the missing transfer and prepares a retry for a manager to confirm.
The rule that makes it safe: the agent never moves money
Every action with money — a bank request, a payout retry, a status change — is prepared by the agent and confirmed by a manager with one button. The agent finds, checks, replies and drives the case; people approve anything that moves funds.
Data stays in your perimeter. The agent runs inside your infrastructure, our support works without access to your data, and neither volumes, merchants nor terms leave it.
What it connects to
Your processing and back office, payment orchestrators such as Praxis and PaymentIQ, KYC and anti-fraud such as Sumsub and Seon, ticketing such as Zendesk, Freshdesk and Intercom, and the channels clients use — email, Telegram, live chat. Your own system connects via API. The full product is described on the payment provider support page.
How to start
- Diagnostic. We look at your request flow and show what can be taken off the team and what it costs today.
- Test on past cases. The agent works through your closed requests first; it goes live once its decisions match your team's.
- Pilot on one request type. One queue, measured before and after.
For the wider picture beyond payments, see AI in iGaming: where AI agents actually work.
Example: a player payout stuck for three days
Stuck payout: finds where it is and follows through to an answer
Finds the operation, works out whose side it's stuck on and why, answers the client substantively and comes back with updates until the case is closed.
Results from our deployments
Live clients under NDA; numbers come from client reports.
A client or merchant writes — the AI agent answers at once: where the payment is, why it's stuck, when it will clear. At night, on weekends, at peaks. More clients and turnover with the same team.