E-commerce

Trust & Safety for Two-Sided Marketplaces

Fraud detection, content moderation, KYC/AML verification, and abuse prevention for platforms connecting buyers and sellers.

Platform engineering and trust & safety teams at two-sided marketplaces (goods, services, gig economy, rental) handling payments between strangers at scale. $2,000–$25,000/mo depending on GMV and verification volume. Sift charges per event (~$0.01–$0.03), Sumsub per verification ($1.50–$4), HUMAN by monthly MAU. Stripe Radar is included in Stripe pricing; Radar for Fraud Teams adds $0.02/screened charge. 📦 11 tools

Two-sided marketplaces face a unique trust & safety surface: fake listings, payment fraud, fake seller identities, stolen card usage, buyer-seller disputes, and illegal content. This stack covers the full lifecycle from seller onboarding (KYC/AML) through transaction fraud scoring (Sift, Signifyd), bot and fake account prevention (HUMAN Security, PerimeterX), and content moderation for user-generated text and images. The tools layer on top of each other — identity verification at signup prevents a class of fraud that no amount of transaction scoring can catch after the fact.

The Stack

Sift

— Account and transaction fraud scoring

Sift's machine-learning risk scores cover account takeover, fake account creation, payment fraud, and dispute abuse. The Score API integrates at account creation, login, and checkout with sub-50ms latency.

Alternatives: kount, accertify, feedzai, riskified, signifyd

Signifyd

— Chargeback guarantee and order decisioning optional

Signifyd provides financial liability shift: if an order they approve results in a chargeback, they pay. Eliminates manual review queues for low-to-medium risk orders.

HUMAN Security

— Bot detection and fake account prevention

HUMAN (formerly PerimeterX) protects account creation, login, and checkout flows from credential stuffing, carding attacks, and inventory hoarding bots.

Alternatives: datadog-asm, fastly-waf, cloudflare-waf

Hive Moderation

— AI content moderation for UGC

Hive's vision and text classifiers detect NSFW content, hate speech, spam listings, and counterfeit product signals in user-uploaded images and descriptions at scale without a large human moderation team.

Alternatives: activefence

Sumsub

— KYC and seller identity verification optional

Automated document verification, liveness check, and AML watchlist screening for seller onboarding. Covers 220+ countries and 6,000+ document types. Webhook-driven decision flow.

Alternatives: onfido, jumio, veriff, au10tix, idenfy, trulioo, smile-id

Onfido

— KYC identity verification (alternative) optional

Onfido's Atlas AI runs document and biometric verification with a strong US/EU presence. Good fit for regulated verticals (financial services marketplace, alcohol, cannabis) where audit trails matter.

Stripe

— Payment processing and Stripe Radar

Stripe Connect handles split payments and payouts to sellers. Stripe Radar provides ML-based card fraud scoring at checkout with customisable rules, reducing friendly-fraud chargebacks.

Accertify

— Enterprise fraud management optional

Accertify (an Amex company) provides device fingerprinting, link analysis, and a rules engine for high-volume marketplaces needing fine-grained manual review workflows.

Kount

— Digital identity trust network optional

Kount's global identity network links device, email, and payment signals across 60B+ interactions to score both buyers and sellers before the first transaction.

Twilio

— Phone verification and 2FA

Twilio Verify provides phone number verification and TOTP/SMS 2FA for seller and buyer account hardening. Phone verification at signup significantly reduces fake account rates.

Datadog

— Fraud signal monitoring and alerting optional

Custom dashboards tracking chargeback rate, dispute rate by seller cohort, and moderation queue backlogs. Anomaly detection alerts on sudden spikes in fraud signals.

Gotchas

⚠️ False-positive friction kills conversion: aggressive fraud blocks frustrate legitimate users. Use risk tiering — auto-approve low-risk, step-up verify medium-risk, manual review high-risk — rather than binary block/allow.
⚠️ Seller KYC vs. buyer KYC: full KYC (document scan + liveness) is appropriate for sellers receiving payouts but adds too much friction for buyers. Implement proportional verification based on transaction value and risk signals.
⚠️ AML thresholds: in the US, payments platforms must file CTRs for transactions over $10,000 and SARs for suspicious activity. Ensure your KYC provider covers AML screening, not just identity verification.
⚠️ Content moderation latency: synchronous ML moderation on listing creation can slow down seller UX. Use async moderation with a hold-for-review state for flagged content rather than blocking the listing creation API call.
⚠️ Dispute rate benchmarks: card networks flag merchants at >1% dispute rate (Visa) or >0.9% (Mastercard). Track dispute rate by seller and proactively offboard high-dispute sellers before the marketplace account is flagged.
⚠️ Network effects for fraud: fraud rings test stolen cards on low-friction platforms first. Implement velocity checks (same card across multiple accounts, multiple failed attempts) in addition to per-transaction scoring.