Coris

Proprietary risk models score merchants, agents decide alerts and AI assistants research merchants and summarize investigations so analysts decide in minutes. A merchant database with manual review queues, the operation being replaced, is what would be left.

Agents decide alerts, pause payouts and resolve routine cases with a person in the loop and escalation for edge cases. What separates routine from edge, the confidence that triggers escalation and how quickly a paused merchant reaches a person are not published.

Review time down 75 to 80 percent measures work removed, not how many removed reviews would have changed an outcome. Precision, missed fraud and a false positive rate are all unpublished, though the platform both approves merchants and stops their money.

Weave cut daily reviews by 89 percent, Tekmetric reduced manual risk reviews by 70 percent and Zift cut review effort by 30 percent, each named with executives quoted. The figures measure review effort rather than fraud caught or merchants wrongly declined, with no baseline period.

Many merchants hold accounts with several platforms, so a risk signal raised at one payment facilitator could follow a merchant to the next. Insights from deidentified usage data may go to third parties under the terms, a permission rather than a boundary, and training is not addressed.

Merchant and business information processed through the platform is Customer Data under a privacy policy updated August 20, 2026, with Coris as processor and individual requests routed to the customer. Retention runs as long as needed without periods, and subprocessors are not named.

SOC 2 Type II compliance is stated, and a trust center on Vanta is linked. Its audit period is not given.

Compliance violations are said to be detected without naming which, and no supervisor, statute or instrument appears. Sponsor banks, a named buyer type, carry supervisory duties for overseeing the platforms they sponsor, the very risk this product manages.

Paused or declined merchants are often small businesses whose cash flow stops that day, and thin digital footprints produce weaker profiles for new, offline heavy or foreign firms. Common name false positives affect impersonation detection, and error analysis by merchant type or geography is unpublished.

Full audit trails on agent actions let a platform reconstruct and defend why a payout was paused. Terms updated April 11, 2024 provide the service as is with no accuracy warranty and cap liability at twelve months of fees, and the merchant whose money stops has no described route to contest it.

Mastercard's terminated merchant database, Emailage and ScamAdviser are named outside sources alongside proprietary models. Language models behind the agents, and a subprocessor list, are not identified.

Stripe Connect, Adyen for Platforms and TSYS are named on the processing side, with Salesforce and Zendesk for cases and Snowflake as a warehouse, each with a stated function. Developer documentation sits behind a login at docs.coris.ai.

Processing happens in the United States and other locations around the world, per the privacy policy, with no hosting provider, region or residency commitment named. Banks overseeing third party processing will want that location documented.

A demonstration is the route to terms, and no price is published. Setup is handled by the company and rules are configured by its own specialists under a managed risk offering, while the charging basis goes unstated.

Payment facilitators, sponsor banks and marketplaces are among eight named buyer types that own merchant risk directly, with coverage from onboarding through monitoring to disputes. Everything serves the one function of merchant risk, and material behind each buyer type is uneven.