Rules anyone can read
Plain-English rules. Test them on last month, then switch them on.
Fraud prevention built for how Africa pays. Sieve checks every sign-up, login and payout, and stops fraud before the money leaves.
Your app asks before money moves. Sieve answers with a decision and the reasons. Tap a step to see how.
{ "type": "payout.requested", "customer_id": "cus_1001", "amount": 48500, "currency": "KES", "method": "mpesa" }
{ "decision": "block", "fraud_probability": 0.94, "reasons": [ { "reason": "Phone and payout account changed 16 minutes after signup" }, { "reason": "Payout account added 5 minutes ago" }, { "reason": "Withdrawing 99% of the balance" } ], "patterns": [{ "name": "New-account cash-out" }], "latency_ms": 16 }
Most fraud tools were made for card payments. Here, money moves through mobile wallets, agents and instant payouts, and fraud can happen in under an hour.
Works with how your customers pay

Also PalmPay, paybills, tills, USSD, POS, agents, bank transfers and cash.
SIM swapped, PIN reset, wallet emptied. All inside an hour.
Many small deposits from strangers, then one fast cash-out.
A quiet account wakes up on a new device and starts paying out.
Deposits kept just under the reporting limit.
Cash in, USDT out to a brand-new wallet, within minutes.
Festive sends, night shifts and travel stay approved.
Pick a country. Currency, phone format, payout rail and local time all follow it.
One device, basic ID checks. Nothing looks wrong yet. Most fraud starts like this.
From a new device. Sieve notes it: details changed right after signup.
It is added as the M-Pesa payout account. A second warning sign.
Sieve sees the whole story and holds the payout. The money never leaves.
Your app gets block and three plain reasons. Your team gets a written-up case.
Sieve decides before the money moves. Six tools, one view of each customer. See the platform →
Plain-English rules. Test them on last month, then switch them on.
Once fraud is confirmed, Sieve remembers its phones, devices and accounts.
Sanctions and PEP checks that re-run whenever lists change.
The AI gathers the evidence. Your analyst decides.
Scans your code for the payment bugs fraudsters look for.
const d = await sieve.events.create({ type: 'payout.requested', customer_id, amount, }) if (d.decision === 'block') hold()Read the setup guide
Every decision is logged in a chain that shows if anything was changed.
No surprises. See full pricing.
Try everything with test data.
Plus $0.10 per screened customer a month.
Your own cloud or servers, with guaranteed support.
100,000 decisions + 5,000 screened customers = $1,500 a month.
Something else? Talk to our team.
94.5% on our benchmark of 515 test cases, and above 90% for every business type. On your own traffic, Watch mode shows the results before anything is blocked.
Most teams send their first live event the same day. It is one API call before each sign-up, login and payout.
Not without you seeing it first. Test any rule on the last 30 days before you switch it on.
Yes. Amounts, phone formats and regulator reports follow your country.
Yes, fully inside your network, with a local AI model and no outside connections.
No. It writes up cases and recommends. A person makes every final call.
Lipana, a payments product from our founder, is our first customer.