The easiest way to stop an online fraudster is to stop everybody. No orders, no stolen cards, no chargebacks. A wonderfully clean dashboard. A rather empty business. Somewhere between that absurd extreme and accepting every purchase sits the problem Riskified sells to merchants: deciding which strangers deserve a yes.
- Riskified screens ecommerce orders and backs eligible fraud decisions with a chargeback guarantee.
- Merchants pay a negotiated percentage of approved sales for its core product.
- The useful scorecard counts good orders recovered, fraud losses and operating effort together.
A chargeback is a payment reversal following a cardholder dispute. For a merchant, fraud can mean losing the payment after the goods have gone. But a legitimate shopper rejected by a fraud filter leaves a quieter kind of loss. There is no angry bank notice. The sale simply disappears.
The customer your fraud filter never met
Eido Gal and Assaf Feldman founded Riskified in 2013. Entrée Capital, an investor, describes their starting conviction: merchants were turning away legitimate customers because they feared fraud. That is a useful place to begin a software company. The customer’s existing defense was also obstructing the customer’s business.
Riskified’s answer was to combine machine learning with an unusually tangible commitment. Its Chargeback Guarantee gives merchants approval or decline decisions. If a guaranteed, approved transaction produces an eligible fraud chargeback, Riskified reimburses the merchant under the contract. The decision comes with financial consequences for the company making it.
The machinery draws on device information, behavioral signals and transaction history across its merchant network. Riskified calls one part of that process Smart Linking: connecting clues across interactions. A merchant sees an unfamiliar order; a network can supply context that one storefront lacks. Fraud expertise, data science and software engineering meet in that small gap.
Eligible fraud loss → reimbursement
Simplified Chargeback Guarantee model. Coverage follows the merchant contract.
This creates a particular kind of sales conversation. A merchant is buying a decision, access to a larger pool of evidence and a promise about who absorbs an eligible loss. Those parts should be evaluated separately. A model can be accurate while a contract is a poor fit; a generous contract can still leave a team wrestling with awkward daily workflows.
The decline that could not be undone
Consider Meshki, the Australian fashion retailer. Its Riskified case study describes a customer service operation burdened by manual fraud and dispute work as the business expanded internationally. Before the change, the team could spend as many as 20 hours a week on tedious manual tasks. Growing sales also meant growing administrative drag.
“With our previous solution once an order was rejected it couldn’t be recovered.”Damien Park-Neilson, Head of Operations, Meshki
What Meshki wanted was automation with room for judgment. Riskified’s Control Center allowed another review of a declined order, giving the retailer a chance to recover a legitimate purchase. The detail matters. A fraud system can make its first decision quickly and still need a practical way to reconsider.
Meshki case study. Reported customer outcomes, not a forecast for every merchant.
A percentage, a promise, a catch
Riskified primarily charges for Chargeback Guarantee as a risk-adjusted percentage of approved merchandise value. Fees vary with the merchant, transaction population, geography, historical fraud and agreed approval rates. There is no single public rate to paste into a purchasing spreadsheet. Other products can use subscriptions or transaction fees.
The incentives are worth inspecting. More approved value earns Riskified more fees. More fraudulent approvals create reimbursement costs. Excessive caution can sacrifice revenue; excessive generosity can sacrifice margin. The commercial design asks the company to keep working on both mistakes.
This approach has company. Signifyd also offers financially guaranteed fraud decisions. Forter sells identity-based fraud management, payment optimization and abuse prevention. A buyer should compare coverage, approval performance, integration and total cost. The word “AI” does remarkably little of that work.
Riskified concentrates on enterprise merchants. Publicly featured customers include Booking.com, Prada and Gymshark. Its annual report defines enterprise merchants as businesses with more than $75 million in annual online sales. Those customers generated more than 90% of estimated 2025 billings. A high-volume checkout supplies plenty of opportunities for small improvements to matter.
The economics still need a pencil. A thin-margin seller with little fraud exposure may struggle to justify another fee. A merchant must supply data, integrate workflows and understand which losses qualify for reimbursement. Those are practical limits to the business case. A guarantee never makes its terms irrelevant.
A useful evaluation would start with a baseline: what did the existing system approve, what did those orders earn, and what did fraud handling consume? Then examine comparable transactions under the proposed arrangement. Compare similar markets and product mixes. Otherwise, a change in the shoppers arriving at the store can masquerade as an improvement in the software watching them.
The fraud follows the customer
The product range now follows problems beyond the original order. Adaptive Checkout adjusts the payment journey. Account Secure targets account takeovers. Policy Protect addresses abuse of returns, refunds and promotions. Dispute Resolve assembles and submits chargeback evidence, with editing controls for merchants who need to add their own material.
Partnerships put those decisions inside other systems. A commercetools connector brings fraud decisions into commerce workflows. A September 2026 Zendesk announcement brings identity risk signals into customer service, where a refund request can become the next consequential decision. ARIA, expanded that month inside Control Center, lets fraud teams question order data in natural language.
On October 8, Riskified announced Agent Identity Risk Intelligence to connect activity by AI personal assistants with the people behind them. Its limited enterprise beta is planned for December 2026, with general availability expected in 2027. The premise is straightforward: permission for an assistant to act does not establish whether its owner is trustworthy.
Measure the yes
Investors have funded the wager. A $165 million Series E in 2019 preceded a 2021 NYSE listing. Riskified reported $344.6 million in 2025 revenue. Second-quarter 2026 revenue reached $98.7 million, up 22% year over year; the quarter still recorded a GAAP net loss. Growth and profitability remain separate questions.
For merchants, the transferable lesson is to change the scorecard. Compare legitimate approvals, fraud expense and staff time over the same period. Calculate the contribution from recovered sales after fees and remaining costs. Keep a route for reviewing mistakes. A clean fraud dashboard feels reassuring. A customer who can actually buy something is worth counting, too.