Riskified

Service Overview

Riskified is an Israeli fraud-prevention provider founded in 2013 by Eido Gal and Assaf Feldman. Its stated positioning is making e-commerce safer through AI, but the business objective is stated more concretely: letting merchants approve more of the orders they would otherwise decline.

That sentence is the key to reading the company. Most risk providers sell better blocking, which carries an implicit trade-off - decline a bit more in order to hold the fraud rate down. Riskified sells the opposite: approve orders you would not dare approve, and let it carry part of the consequences. The arrangement behind that promise is chargeback liability shift: the merchant tunes strategy to its own risk tolerance, and fraud chargebacks inside the guaranteed scope land on Riskified.

So the right way to read it is not as fraud tooling but as insurance. Tooling value is decided by the model; insurance value is decided by the underwriting terms. Model quality is only the precondition for it being able to keep underwriting - a model that loses money tightens terms or raises price, and a model that holds up is what lets you get a wide guarantee.

Beyond payment fraud it covers account takeover protection, payment failure remediation at checkout, and abuse blocking that keeps consumer-friendly policies intact. All three connect to chargebacks: orders placed after account takeover become disputes later, failed payments lose orders immediately, and over-tightened policies push normal customers away. Putting them under one commitment is internally consistent.

Key Strengths

• Turns uncontrollable dispute risk into a budgetable line item: chargebacks cost more than money - they drag card-network monitoring metrics, fines, accounting adjustments and dispute handling labour. Under a guarantee that becomes a negotiable fee that finance can budget, and risk teams stop re-litigating each order.
• Approval strategy is a tunable dial: you are not handed a black-box threshold but a dial - tighten for lower exposure and narrower cover, loosen for higher approval rates and wider cover. Different margin profiles can each find their own economic sweet spot.
• Higher approval rates are direct revenue: false declines are invisible in reporting but very real under growth pressure. For long-repurchase, high-ticket categories, recovering declined good customers usually beats squeezing the fraud rate another notch.
• Network effects widen recognition: machine learning over large cross-merchant datasets means the behaviour of one identity elsewhere helps decide whether to approve it with you, which matters most where the new-customer share is high.
• Account takeover is covered as its own attack surface: ATO works through real accounts, real addresses and real credentials, so transaction features alone rarely expose it. Sharing one decision layer with payment fraud closes the gap where both sides disclaim responsibility.
• Abuse is blocked without tightening policy: many merchants clamp down on returns and refunds to prevent abuse and pay for it in customer experience. This approach identifies abusive behaviour while leaving the policy unchanged, which matters to brand-sensitive merchants.

Product Ecosystem

Payment Fraud Management

The core scenario: approve or decline at checkout, with fraud chargebacks inside the guaranteed scope absorbed by the provider. Decisions draw on cross-merchant data and model scores, and merchants generally stop layering manual review on top.

Account Takeover Protection

Covers login, credential and contact-detail changes plus the orders that follow. It shares the decision layer with payment fraud, which removes the seam between account security and transaction risk.

Checkout Payment Failure Remediation

Targets technical failures in the payment step - issuer declines, interrupted verification flows, missing retry logic - that silently lose orders. These never show in chargeback reporting but they do hit conversion.

Abuse Blocking

Identifies promo arbitrage and policy abuse while keeping consumer-friendly policies in place, instead of buying safety with blanket rule tightening.

Decision and Guarantee Interface

Integrated by API into checkout or order systems, returning a single decision and recording guarantee attribution so later disputes can be matched.

Limitations

• The guarantee is not comprehensive: contracts define covered categories, regions, ticket ceilings and exclusions (for example orders not executed per the decision, self-modified orders, or shipments outside agreed timing). Read the exclusion list line by line before signing, or you discover the gap only when a loss appears.
• You give up final authority: the guarantee presumes you follow the decision. Orders your risk team manually overrides outside an approval generally fall outside cover, which means paying a premium for protection you cannot use. This often collides with existing approval workflows and requires re-dividing authority.
• Pricing tracks dispute performance, weakening predictability: a deteriorating dispute mix pushes the rate up, and the mix is driven by category, promo cadence and delivery speed. Peak-season promotions can lift both chargeback rate and fee at once.
• Data sharing and cross-border compliance cost: order and consumer behaviour data must be transferred, and cross-border use requires GDPR and personal data export assessments, with separate export paths where applicable. Compliance review often outlasts integration.
• The model depends on volume and data quality: network effects cut both ways. Narrow verticals and low-volume merchants receive tighter guarantees and higher rates because the underwriter lacks samples.
• No coverage of regulated obligations: AML transaction monitoring, onboarding review and sanctions screening are outside the product, so regulated entities still need a separate system and end up running two risk stacks.

Use Cases

• Mid-to-large cross-border DTC stores (★★★★★): volume, mid-to-high tickets and meaningful chargeback cost make the guarantee model work - exposure becomes a budgetable expense while approval rates rise.
• Digital goods and virtual services (★★★★☆): no logistics and no delivery proof, so disputes are structurally hard to win and the guarantee is most valuable. Confirm the category is covered first.
• High-ticket and luxury e-commerce (★★★★☆): a single chargeback is large and manual review is expensive, so the guarantee materially reduces per-order handling pressure.
• Subscriptions and memberships (★★★☆☆): payment fraud and account takeover are covered, but recurring-billing disputes differ from one-off orders, so confirm whether that dispute type sits inside the guarantee.
• Mainland China operations (★★☆☆☆): usable when a mainland entity runs cross-border business, but transferring order and consumer data abroad must follow the Personal Information Protection Law and the applicable data export rules, so a compliance assessment comes first. Purely domestic business does not need overseas guaranteed services, has domestic equivalents, and avoids the export question entirely.
• Very low-volume new stores (★☆☆☆☆): least suitable. The guarantee premium plus cold-start network effects mean high unit cost and typically narrower cover, so returns do not cover the premium.
• Teams that must keep manual final approval (☆☆☆☆☆): least suitable. Manual overrides void the guarantee, which means buying insurance you cannot use.

Pricing

Cost drivers include reviewed order volume, GMV, guarantee scope (category, region, ticket), chargeback rate, whether manual investigation is included and the integration approach. Riskified does not publish standard rates; headings and any final rate follow the current official quote and contract terms.

FAQ

• What exactly does the guarantee cover, and what does it not?
  It covers fraud chargebacks inside the contractual scope, which typically carries boundaries on category, region, ticket size and shipping timing. It does not cover orders not executed per its decision, situations outside the agreed scope, or disputes it classifies as non-fraud. Ask for the full exclusion list and map it against your business before signing.

• Can we keep our own manual review?
  You can review, but the guarantee is affected. Cover presumes you follow the decision, so overridden approvals generally fall outside it. Either hand approval authority to the model and only work the referrals it sends you, or accept that those orders are uninsured - and state the choice in the contract.

• Is this the same as buying insurance?
  The mechanics are similar, with one difference: the underwriter is also the service provider. It must hold the fraud rate at a survivable level before it can afford to offer you wide cover. So negotiate two things together - how wide the terms are, and whether the model's historical performance on your traffic supports that width.

• Can a high-chargeback category use it?
  It can apply, but terms will be tighter. A high chargeback rate raises expected payouts, which usually means narrower cover or a higher rate. If the category already sits under card-network high-risk monitoring, fixing the operational causes - delivery proof, description accuracy, refund timelines - beats switching providers.

• What should a mainland China company watch out for?
  It can be evaluated, but data compliance comes first. Transferring order and consumer behaviour data abroad involves the Personal Information Protection Law and the applicable export rules, so confirm whether a security assessment, standard contract or certification path applies. Purely domestic business does not need this service and has no export question.

Source: original 16IDC content. Product capabilities, supported countries and fee definitions follow the official documentation and your contract.