Platform — AI services
Fraud stopsin flight
Fraud detection, intelligent automation and AI-assisted decisioning, running on the same platform that moves the payment — so a decision is made in the path of the transaction rather than after it.
01 — Fraud detection
Fraud caught before the money leaves.
A fraud report that arrives the next morning describes money that has already gone. On Pixely the scoring sits between the request and the rail, so the answer comes while there is still something to stop.
Transaction scoring
Every payment is judged against the payer's own history — the amount, the payee, the hour, the place — before it is routed.
Behaviour and device
How a session moves, which device it comes from and whether that device has been tampered with all count toward the same score.
Networks, not just payments
Money mules and rings show up in the shape of who pays whom. The platform reads that shape across accounts, not one payment at a time.
One view of the person
The signals from onboarding travel with the account, so the fraud model knows what identity verification already saw.
Reasons, not just scores
Every hold and every refusal carries the signals that drove it, in words a reviewer and a regulator can both read.
Your rules, in the loop
Scores feed your policy rather than replacing it: allow, hold for review or refuse, decided by thresholds you set and can change.
Decision streamSample
- payment.scoreusual payee, usual deviceALLOW
- velocitynew payees in quick successionHOLD
- device.intelemulator detectedREFUSE
- payment.scoresalary credit, expectedALLOW
- networkfunds fanned to new accountsHOLD
- sessionpasted credentials, new deviceREVIEW
- payment.scoremerchant, regular basketALLOW
- account.takeoverpassword reset, then payoutREFUSE
- payment.scorefirst payment to a supplierALLOW
- review.closedhold released by analystALLOW
02 — Intelligent automation
The work nobody should do by hand.
Much of what a finance operation does all day is reading, matching and sorting. The platform is built to do that part, so people spend their time on the cases that need judgement.
Reconciliation
Incoming money matched to the invoice it pays as it lands, so the books close while the month is still running.
Categorisation
Every movement labelled with what it was for, consistently, and exportable the moment it settles.
Documents read
Invoices, statements and registry extracts read into structured fields, with the source kept beside what was taken from it.
Cases routed
Alerts arrive at the right reviewer with the evidence already gathered — the account, the history, the signals and what they mean.
Questions answered
Ask about a customer, a payment or a balance in plain language and get the answer with the records it came from.
Reports drafted
The routine reports a finance or compliance team owes assembled from the ledger, ready for a person to check and send.
03 — How a decision is made
AI-assisted, never unaccountable.
Four movements, and the last one keeps the rest honest: every decision the platform helps make can be explained, reviewed and corrected.
- 01
Signal
The payment, the session, the device and the history around them gathered at the moment of the request.
- 02
Score
Models and rules judge the signals together, and each score keeps the reasons that produced it.
- 03
Decide
Your policy turns the score into an outcome — allow, hold or refuse — in the same call, before the rail.
- 04
Review
People take the cases that need judgement, and their outcomes are fed back into the models and the rules.
04 — Get started
Put it inthe path
Create an account and run payments through the sandbox's decisioning, or talk to an engineer about your fraud and your rules.