Generative AI or GenAI is a seriously trending global topic. The reason for its success is the ease of use in a multitude of cases. Four of the most common are: content creation; translation; writing code and debugging code; and good old-fashioned learning.
Although GenAI has taken the world by storm, it’s actually not a new concept. There have been models developed by OpenAI and others able to perform similar tasks for some time, so why does it have so much traction now?
The main reason is that the latest GenAI, like ChatGPT, has the ability to readily understand natural language. Understanding and evolving responses based on user directions is something to behold. GenAI can determine the meaning as well as the intent of its users and that’s a game-changer.
ChatGPT, for instance, can follow an entire conversation and maintain context, so a user could use pronouns, refer to various segments of the conversation, ask it to summarize the conversation or update the output to sound funny or poetic, and it will be able to follow and respond accordingly.
GenAI has demonstrated the effectiveness of a conversation-based paradigm for human learning and understanding in assisted writing and coding, creating diet plans, booking travel itineraries, and so on. And KYC processes can similarly benefit from GenAI chat-based workflows - providing a natural, human-friendly user experience.
With this kind of chat-based AI in a KYC workflow, interactive investigations and intelligent screening of entities becomes possible. It focuses on the strengths of human cognition - asking the right questions and forming rational answers (as opposed to memorizing facts and code syntax).
Although GenAI such as ChatGPT are exciting technologies that understand natural language, there are limitations:
Moody’s believes harnessing GenAI technology and integrating it with existing proprietary datasets and workflow technology are the best way to support and develop smarter KYC workflows with natural-language-based solutions embedded.
Public GenAI tools, like ChatGPT, only respond with publicly available information – for example training data for ChatGPT that is on the web. These instances of GenAI don’t know and can’t therefore include information from KYC databases such as Moody’s Grid and Orbis, and our official registries datasets from Kompany, which we own. These datasets are Moody’s proprietary information. Any KYC workflow for a person or entity is incomplete without access to trusted data like this.
Ideal responses to KYC queries need to include trusted, verified, accurate and up-to-the-minute data, presented to risk and compliance professionals to ensure a well-judged and fair outcome. To achieve this, an AI chat interface within a KYC platform would need to be established to support and inform decision making. Like a co-pilot operating as a virtual colleague who is trained on all proprietary data, and can be trusted to use, process, and share that data appropriately - and instantly.
Currently, the suite of Moody’s KYC products is working to provide a structured input and output, with Large Language Models (LLMs) acting behind the scenes in various pipelines that ingest media articles to support intelligent screening.
In the future, there will be many more customer experience improvements from natural-language-based querying and generative responses powered by Moody’s industry-leading data and KYC workflow automation platform.
Using AI technology to enhance KYC processes includes Gen AI, of course, but it encompasses a lot more besides.
Effective know your customer (KYC) processes are essential for compliance and third-party risk management in a range of sectors, from financial services to corporates to fintechs and beyond.
The manual KYC methods, which could be error-prone and were often time-consuming or labor-intensive, have been phased out in many industries. Through digital transformation and the introduction of RegTech, organizations can carry out automated risk assessments at onboarding and throughout a customer lifecycle.
KYC platforms, like Moody's, have revolutionised areas of anti-money laundering compliance, rules based risk management, and transaction monitoring - able to offer workflow automation, integration with global data, and full case management for human oversight.
Advancements in artificial intelligence (AI) and machine learning (ML) are continuing this expansion of digital KYC – offering more solutions for automated identity verification, security enhancements such as liveness tests, and intelligent screening that leads to more efficient name matching and entity verification.
AI and ML technologies can help instantly authenticate identities; screen government IDs and bank statements; and help detect patterns of fraud. These technologies also enable continuous monitoring for suspicious activities or changes in risk profiles, flagging high-risk entities within a counterparty network who can then become subject to enhanced due diligence (EDD).
KYC processes are designed to help organizations understand who they are doing business with. They help prevent financial crime, like money laundering and fraud. And there are typically 3 steps in a KYC process:
Each of these three steps could be explored using a GenAI chat-based KYC workflow. Moody’s, for example, could create a KYC workflow containing GenAI chat as the user endpoint. Users could then get screening-related information via a free-flowing conversation, although it would currently be challenging to perform bulk screening this way. Nevertheless, it provides a good KYC investigation tool to serve as a second line of defense.
In the first line of defense, KYC professionals need automation to process data differently, carrying out intelligent screening to detect patterns and analyze behaviors that indicate risk, rather than asking questions. Applications of AI and ML in the first line might include:
The question-and-answer format provided by GenAI chat could then help compliance teams deal with KYC alerts that require escalation to the second line of defense, which is where investigations happen, and practitioners have to interrogate data and risk profiles at a more in-depth level.
AI and ML in KYC workflows could support with processing the vast amounts of data need to understand a world of risk and generate profiles that enable better decision making. While GenAI’s chat-based model could be used as an interactive and transparent investigation and research tool surrounding an entity being assessed, for example in an enhanced due diligence (EDD) process.
AI's role in KYC is set to expand, offering organizations significant benefits such as reduced costs, heightened accuracy, and an optimized customer experience. By leveraging AI for identity verification and risk assessment, compliance teams can allocate resources more effectively, focusing on areas of highest risk and importance.
1. What is the difference between AI and ML?
Artificial intelligence (AI) and machine learning (ML) are closely related but distinct concepts within the field of computer science. AI refers to the broader discipline of developing intelligent machines capable of simulating human intelligence. AI aims to mimic or perform a task that would normally require human engagement to make decisions or take actions. Machine learning, on the other hand, is a subset of AI that focuses on using large sets of data and the nuanced patterns within the data to generate software models used to solve problems and derive insights.
ML algorithms can be updated to improve their performance over time through the addition of new data, allowing systems to recognize patterns, make predictions, or perform tasks without explicit instructions. While ML is a powerful tool in AI, it is just one component of the broader field, which also includes areas like natural language processing (NLP), computer vision, and robotics.
At Moody’s, we deploy a supervised learning model to maximize screening efficiency. In this type of model, the algorithm is trained using labeled data, where the input data is paired with corresponding desired outputs. The model learns to map inputs to outputs by generalizing patterns and relationships in the data.
2. How can this technology be adopted into a know your customer (KYC) process?
Moody's AI Review technology creates an AI-powered Alert Score for each entity an organization, such as a bank or corporation, wants to screen. This 0-1 score represents the confidence level of the screened name with 0.00 as a no-match and 1.00 as a match.
Moody’s Grid is the dataset that feeds AI Review. Over 12.1 million rows of data were used to train our AI Review global model. Additionally, firm-specific data can be applied for a local model to be deployed within a company to better reflect decisions based on its company policies and analysts’ behavior.
Setting an alert threshold to adhere to a firm’s unique needs
The Alert Score can also be used to filter results. For example, firms can choose to further analyze only results that have a score of greater than 0.25 to help reduce false positives. This tunable screening, defined specifically by a company, can be configured to drive efficiency, while maintaining control over the screening process.
Every alert is scored by the model on a scale of 0–1. An alert with a score closer to 1 signals that the screened name is higher risk and can be sent to level-2 analysts for additional investigation. By setting a threshold, scores on the lower end signify less likelihood of an alert and can be automatically filtered. This means the ML technology can act as a level-1 analyst providing repeatable decisioning in sorting out false positives, so compliance teams can focus on more important work of investigating higher risk alerts.
Monitoring the ML model to implement proper governance
It is important that any company deploying AI/ML technology understands the inner workings of how it is being used within the organization’s processes and procedures, and to have governance around its design and implementation.
Our approach is to deploy models that are static, meaning the model does not learn from customer decisions in real-time without human intervention. This is purposeful and by design. Otherwise, the continuous, automated training that the machine learning model does will present a significant oversight burden for regulated financial institutions and multi-national companies conducting KYC/AML processes. Instead, we monitor our models for any drift that may indicate the model is no longer fit for purpose.
3. What are the common challenges for adoption of AI in KYC workflows
The adoption of AI into KYC processes is influenced by several barriers. Here are some common challenges:
Using innovative technology while keep humans in the loop is a great way to help you answer three key questions:
Moody's is creating solutions so you can understand risks and make decisions with confidence. Get in touch any time to talk to us about how we can help with your KYC workflows - we would love to hear from you.