Revolutionizing KYC:
A 5-Step Guide to Streamlining Perpetual Customer Behavior Monitoring
whitepaper
Navigating the constantly shifting landscape of regulatory requirements presents a daunting task for banks. At the core of this challenge is the Know Your Customer (KYC) process, which has grown increasingly complex and demanding in recent years. The necessity for significant resources and expertise only heightens the difficulty. Financial institutions, including banks, must adopt these measures to mitigate the risks tied to financial crime.
To effectively tackle these concerns, banks need to gain a deep understanding of the nature and purpose of customer relationships, develop customer risk profiles, and conduct ongoing monitoring to detect enduring behavioral changes, as well as identify and report suspicious activities. However, the intricacy and variety of customer data sources often make this process challenging and time-consuming, particularly for compliance teams.
In this whitepaper, we delve into the complexities of the KYC process and offer practical guidance for streamlining this crucial aspect of establishing and understanding customer profiles, as well as pinpointing when a customer has undergone a permanent behavioral shift that warrants renewed Customer Due Diligence (CDD).
The Need to Reinvent the Approach to KYC
Banks must develop a deeper understanding of customer behavior to accurately assess customer risk profiles. Traditionally, banks have relied on customer declarations; basic information such as client type, industry, client geography; and products/services approved to use. This information is normally supplemented with manual reports of customers’ expected behavior, such as the number of transactions they plan to conduct, or the volume, direction, and value of those transactions, while neglecting their actual behaviors and relationships with other counterparties over time. This approach is considered inefficient, since customers often behave differently than they claim to or than is evidenced from historical data.
The challenge for banks lies not only in accurately determining a customer's authentic profile and identifying evolving behavior, but also in defining a permanent change and integrating the nature and purpose of their relationships. This necessitates analyzing trends over time, rather than focusing on individual, isolated changes in behavior, and viewing the customer as part of an entire network of interconnected relationships.
While banks increasingly focus on cost-effectiveness and implement automated processes to enhance efficiency, they continue to adopt machine learning. However, a significant obstacle remains in fostering collaboration between data scientists and business users.
These two groups have unique perspectives, which makes it challenging to achieve a shared understanding and effectively apply machine learning that produces comprehensible output for compliance staff, ultimately hindering the process.
The sheer complexity of the customer risk profile compels banks to leverage modern technologies such as graph analysis and machine learning. Modern graph analysis captures the nature of relationships between billions of entities and data points, while AI (machine learning) fed by the output from graph analyses effectively categorizes customers based on their behaviors, statistically and contextually.
The Knowledge Graph – A Single View of Bank Data
A critical step in the KYC process involves constructing a comprehensive data universe, which integrates various sources such as transactions, KYC data, customer surveys, products, segments, and any additional information. Yet data and analytics leaders frequently face challenges in developing a cohesive view of the business, since much of the essential business domain knowledge is entrenched in traditional database schemas that utilize obscure language and exhibit ambiguous connections to business reality. This is the primary reason why modern knowledge graphs, as graph-structured data models that integrate data, have emerged as a game-changing solution.
The entire data universe is streamlined, organizing all data around business-related sets, including entities, bank accounts, transactions, products, etc. Furthermore, it establishes relationships between these sets, allowing both technical and non-technical users to communicate using the same language. Knowledge graphs serve as the foundation for designing, testing, and launching the complete profile monitoring process. (See Figure 1.)
Figure 1. An example of a knowledge graph where all bank data is connected
5 Steps for Streamlining Perpetual Customer Behavior Monitoring
Figure 2. Entity resolution keys generated by various techniques
Step 1 – Select segments
After consolidating and integrating data through the knowledge graph, it's essential to categorize customers into groups such as small businesses, payment service providers, non-profits, enterprises, mid-sized corporations, wholesalers, and financial institutions. (See Figure 2.) This segmentation is crucial for establishing relevant benchmarks, as different customer groups display distinct behaviors. It's important that this process does not entail months of coding; rather, to ensure accuracy, the segmentation should be fine-tuned in a sandbox-like environment.
Figure 2. Entity resolution keys generated by various techniques
Step 2 – Create characteristics
After segmentation, the next step entails generating a collection of variables or features that, when combined, can define the reference profile for each segment separately, both for individual and business clients. Typically, the data scope for such analysis and feature creation spans a period of 9-12 months, and information on behavior trends is captured over periods of three, six, nine, or twelve months.
These features can include:
Figure 2. Entity resolution keys generated by various techniques
The basic features have to be supplemented with the features reflecting the nature of relationships between entities such as:
Figure 2. Entity resolution keys generated by various techniques
Having a high number of features can be overwhelming and difficult to interpret for compliance personnel. To streamline profile interpretation and create a common language for data scientists and non-technical users (including compliance professionals and KYC analysts), these features should be consolidated into business characteristics that align with compliance hotspots. (See Figure 3.) These aggregated variables can be organized around the following activities:
- General Transactions — those that reflect the customer's overall activities
- Foreign Transactions — those that reflect the customer's transactional activities with foreign accounts
- Cash Transactions — those that reflect the customer's cash purchase activities
- Credit Card Transactions — those that reflect the customer's credit purchase activities
- High-Risk Countries Transactions — those sent to or received from high-risk countries
- Very High-Risk Countries Transactions — those sent to or received from ultimate high-risk countries
- Entity Relationships — values and types of relationships between parties
- Others
Figure 2. Entity resolution keys generated by various techniques
Step 3 - Generate the reference profile
Once the characteristics have been created, the next step is to generate the reference profile, using a series of techniques. A machine learning model is supplied with all relevant characteristics, features, and their values. First, the KMeans algorithm is employed to cluster the values of each characteristic separately for each segment. Then, the clustering output is interpreted using a decision tree, assigning business ontology labels to the characteristic's values, such as "low" or "high" activity, primarily sent or received transactions, etc. (See Figure 4.) This method allows for the generation of a self-explanatory customer profile determined by the AI/ML model, which fosters communication among KYC analysts, data scientists, compliance experts, and FIU (Authority) through a unified language.
Figure 2. Entity resolution keys generated by various techniques
Step 4 - Re-calculate the profile with new data
After generating the reference profile, the next step is to automate its monitoring. Instead of requiring machine learning experts to recalculate reference profiles for customers every month, the process should be automated, continuously comparing reference profiles to new data to determine if the customer's reference characteristic values have changed over time. Thereafter, it would be essential to ascertain whether customer behaviors have undergone permanent changes, which would necessitate replacing the existing reference profile with a new one. This can be achieved by adopting a broader perspective, gathering characteristic values from multiple observation periods and evaluating the number and significance of changes observed (e.g., a shift from low to medium or from very low to very high).
Parameters that determine an automatic reference profile update would be:
- Number of characteristics values changes over 4 periods
- Number of periods during which the characteristics values have changed
- Weighted characteristics values change over 4 periods
Figure 2. Entity resolution keys generated by various techniques
Example:
If a customer's cash transaction pattern is initially classified as low and, in the subsequent four monitoring periods, they begin conducting frequent high-value cash transactions that significantly impact the trend, the system will automatically update their cash transaction characteristic from low to high. This adjustment will revise their entire reference profile and generate an alert.
Step 5 - Run the alert to renew the CDD
The final step is to design triggers for notifications or alerts when the CDD for a customer should be renewed. The alert definition may include specific rules based on monitored behavior changes including several parameters such as whether the:
- Customer reference profile has changed permanently
- Customer CDD review is not planned within the next 3 months
- Customer CDD renewal has been accomplished more than 3 months ago
- Customer has adverse media
- Customer risk has changed
When an alert is triggered, a concise summary will help your analysts in decision-making when considering further escalation, showcasing both the customer reference profile and a 4-month observation period.
Figure 2. Entity resolution keys generated by various techniques
Implementing a streamlined approach to KYC, as described in the five steps above, automates customer profile monitoring and dynamically evaluates customer behaviors. By utilizing graph and machine learning techniques, and encouraging collaboration among stakeholders, the process ensures that customer profiles are always current, which helps to maximize regulatory compliance while managing costs effectively. This innovative solution simplifies decision-making for analysts and improves communication among KYC analysts, data scientists, compliance experts, and FIU authorities, resulting in more efficient and effective risk management.
Learn More About DataWalk pKYC Software >
Register to Get PDF
Solutions
Product
Partners
Company
Resources
Quick Links