The Center for Data Innovation recently spoke with Arnaud Schwartz, CEO of CheckMarble, a Paris-based company developing an AI-powered platform and tools to help financial institutions detect and investigate financial crime. Schwartz explained how Marble’s platform helps banks and financial technology companies (fintechs) adapt their compliance systems, identify suspicious activity, and streamline investigations.
David Kertai: What problem is CheckMarble solving?
Arnaud Schwartz: Financial institutions often struggle to keep pace with financial crime due to their compliance systems evolving more slowly than the risks they are designed to detect. Compliance teams should be able to respond quickly to a new payment product, regulatory requirement, or money-laundering technique. However, updating their systems often requires lengthy information-technology projects.
Investigators also spend much of their time gathering information from disconnected systems, spreadsheets, notes, and past cases before they can assess potential risks. As a result, compliance teams struggle to keep up with increasingly sophisticated criminal activity, while criminals use AI-enabled tactics to expand their operations.
We built Marble to help close this gap. The platform allows banks and fintechs to update detection rules more quickly, identify suspicious patterns across customer and transaction networks, and streamline investigations. It also includes AI tools such as the Virtual Analyst, which helps investigators prepare and review cases. By adapting to each institution’s existing data and processes, Marble reduces the time investigators spend assembling information and allows them to focus on assessing risks and making informed decisions.
Kertai: How does Marble integrate with financial institutions’ existing systems?
Schwartz: Many compliance-technology projects become difficult because financial institutions organize and label their data differently. Traditional compliance vendors often require customers to translate their data into a predefined structure before integrating it with a new platform. This process can take months and introduce unnecessary complexities.
Marble takes a different approach. Institutions define their own data model, the way they organize and describe information, using the language and structure they already use internally. Marble then connects customers, accounts, transactions, and other records into a unified view. Integration typically involves defining the institution’s data model, establishing data ingestion, and connecting through our application programming interface. Marble also works alongside existing tools rather than replacing them. Data from Know Your Customer systems, which verify customer identities and assess potential risks, risk-scoring tools, and third-party providers can feed directly into the platform.
Kertai: How does your AI-powered Virtual Analyst guide investigators from an initial alert through the investigation process?
Schwartz: The Virtual Analyst is one of several AI-powered capabilities within Marble. It focuses on the investigative stages of fraud and anti-money laundering, efforts to detect and prevent the use of illegally obtained money. In a traditional investigation, analysts spend much of their time collecting information before they can assess risk. They review transaction histories, examine the individuals or organizations involved in transactions, search previous cases, and gather evidence from multiple systems.
The Virtual Analyst automates much of this preparation. Before an investigator opens a case, the system gathers relevant context, highlights important risk signals, and proposes a risk level with supporting evidence. Every conclusion links to the underlying records through a feature we call the proof view, allowing investigators to verify the evidence directly. This lets analysts spend more time evaluating risk rather than collecting information.
Our platform also follows each institution’s existing procedures by routing cases to the appropriate teams, escalating investigations when necessary, and preparing supporting documentation for regulatory reports that financial institutions submit to authorities when they identify certain suspicious activities.
Kertai: How does your company connect unrelated transactions and customer activities to uncover broader financial crime networks?
Schwartz: Financial crime networks are often designed so that individual activities appear harmless in isolation. Two companies may never transact with one another directly yet still share a device, payment processor, or another hidden connection. Looking only at individual transactions can cause investigators to miss these relationships.
Marble identifies these connections in two ways. First, our AI-powered detection rules tool flags activities involving shared identifiers. For example, if several customers with no apparent connection use the same device to access their accounts or initiate transactions, Marble can flag that connection for investigators to examine. A transaction involving a flagged International Bank Account Number, a standardized number used to identify a bank account across countries, or a shared device fingerprint can also trigger an alert.
Second, our graph visualization tools map relationships between customers, accounts, and organizations, helping investigators understand the networks surrounding a particular customer. The platform also explains why each connection exists, further helping investigators identify relationships between shared devices, IP addresses, or platforms.
Kertai: How does Marble ensure its AI systems remain accurate, trustworthy, and effective as financial crime tactics evolve?
Schwartz: Financial crime is an adversarial problem. Once criminals learn how a system works, they adjust their behavior to avoid detection. Compliance tools therefore need to evolve as quickly as the threats they face.
Marble gives compliance teams the ability to update detection rules quickly. For example, if investigators notice that several customers are using the same device to make suspicious transactions, they can describe that pattern in plain language and use our AI Rule Builder to generate a rule based on their institution’s own data. Teams can test rules against transaction data before activating them and maintain a record of who made each change and why. The resulting rules are deterministic, meaning they produce consistent results when given the same inputs, and explainable.
Trust also depends on how Marble handles investigations. The platform combines deterministic detection with evidence grounding and human oversight. While the rules engine identifies activity that warrants investigation, AI tools help analyze and summarize alerts. Every AI-generated conclusion links to supporting data so investigators can verify the evidence themselves.
Looking ahead, we are developing AI agents to analyze confirmed fraud cases and false positives, or alerts that incorrectly flag legitimate activity, and recommend improvements for compliance teams to review before deployment. Our goal is to help teams adapt as quickly as the threats they face without replacing human expertise.
