Enhancing Compliance and Data Management in Financial Marketing for Visa with an LLM Constructor

Visa, as a global leader in payment processing within the financial services industry, sought to enhance its data management and compliance capabilities in multi-channel marketing operations. Neurons Lab partnered with Visa to develop an LLM Constructor that streamlines localized marketing while ensuring adherence to financial regulations and data security standards.

Partner Overview

If you have a debit or credit card, there’s a good chance it’s one of the 4.3 billion cards issued by Visa. As the world’s leading digital payments network, Visa connects businesses, financial institutions, governments, and individuals across 200 countries, processing 283 billion payment transactions globally in 2023 alone.

Visa facilitates a wide variety of payment types – from P2P and B2C to B2B and G2C. Visa’s global network includes banks, governments, businesses of all sizes, and individual consumers worldwide. Therefore, its messaging must always be tailored to the target audience’s context and location to deliver maximum impact.

Given the sensitive nature of financial data and the strict regulations governing financial communications, Visa requires marketing messages that not only resonate with diverse audiences but also comply with financial industry regulations and data privacy laws.

Project Overview

To help marketing teams overcome these challenges, Visa enlisted our help in creating a simplified constructor powered by Large Language Model (LLM) technology. The constructor would be available to marketing managers to streamline message creation, translation, and optimization.

To maximize marketers’ productivity, the generative AI module needed to be seamlessly integrated with its Campaign Atlas portal and allow for customizable parameter input (e.g., target language, channel, text length, and tone of voice selection).

The new tool had to support 20 different languages. This included 9 must-have languages and 11 nice-to-have languages, plus copy optimization for social media platforms (Facebook, Instagram, LinkedIn, Twitter), messengers (Viber, Telegram), email, and sales scripts.

Additionally, the solution needed to ensure that all marketing communications adhered to financial regulations and data privacy standards, such as GDPR and PCI DSS, to maintain compliance across different jurisdictions.

Business Challenges

Before working with Neurons Lab, Visa had to tailor its messaging to each location and leverage various outreach channels. It is a very time-consuming endeavor as marketing specialists have to grapple with the following challenges:

• Translating marketing copies accurately
• Adapting them to specific communication channels
• Ensuring they remain consistent in tone and messaging with the global corporate guidelines

This efficiency gap in marketing was the key challenge Visa turned to us with. The company wanted to leverage GenAI to increase the marketing teams’ productivity by tackling:

• Translation localization across 20 different languages
• Quality assurance to prevent spelling and grammar mistakes
• Copy optimization for each marketing channel
• Tone and messaging consistency, especially for smaller bank marketing specialists

Furthermore, Visa needed to ensure that all marketing communications complied with regional financial regulations and data privacy laws. 

The complexity of managing sensitive financial data across multiple regions posed significant challenges in maintaining compliance and data security. Marketing messages had to be carefully crafted to avoid any misrepresentation or violation of financial industry regulations.

Solution

The solution to these challenges was a simplified constructor that empowers marketing managers to streamline and optimize marketing messages using LLM technology.

The constructor includes pre-built templates and customizable parameters that users can leverage to generate effective messaging for various communication channels efficiently.

The LLM Constructor was designed not only to enhance productivity but also to embed compliance checks within the marketing message generation process. By incorporating regulatory guidelines into the AI model, the solution ensured that all generated content adhered to financial industry regulations and data privacy standards. This was crucial for maintaining compliance across Visa’s global operations.

To achieve this, we leveraged AWS services such as Amazon SageMaker for developing and deploying the machine learning models, AWS Secrets Manager for secure management of sensitive data, and AWS Artifact for accessing AWS compliance reports and agreements.

LLM architecture

The user inputs a query using the web app dashboard in the Campaign Atlas portal, including:

• Query
• Language selection
• Tone selection
• Channel selection

Next, the natural language processing (NLP) module built on LangChain receives the input and activates components for tone and channel augmentation, as well as translation.

Prompting options for tone include informative, authoritative, enthusiastic, inspirational, and so on. Channel choice includes social media, email, sales rep script, digital ad, and more.

Later in the chain, the NLP module handles translations. Initially, it leverages Anthropic Claude, followed by custom models such as MarianMT and MBart from Hugging Face to improve accuracy and quality.

• Near the end of the chain, the user receives several different intermediate output responses on the Campaign Atlas portal before choosing a text operation.
• These post-processing text operations include requests to make the text shorter or longer, add emojis, simplify the language, improve the writing, etc.
• This response enters the NLP module to perform the post-processing request, and the user receives the final output.

The platform stores a history of all intermediate and final outputs in an extensive database, allowing users to fetch past responses again in the future.

To ensure scalability, security, and compliance, we built the LLM model leveraging AWS services. 

AWS SageMaker was used to develop and train the machine learning models, allowing for efficient handling of the diverse language requirements. AWS Secrets Manager was employed to securely manage sensitive data, including API keys and access tokens, ensuring that all data handling complied with Visa’s stringent security policies and financial industry regulations.

API and Cloud architecture

To address Visa’s challenges in data management, compliance, and security within the financial services industry, Neurons Lab designed a robust and scalable cloud architecture leveraging AWS services tailored for the financial sector.

1. Implementation of the LLM Model:

• Neurons Lab implemented the LLM using AWS SageMaker, which enables building, training, and deploying machine learning models at scale while adhering to strict financial industry regulations.
• By utilizing AWS SageMaker, we ensured that the model operates within a secure and compliant AWS environment, meeting Visa’s requirements for data protection and regulatory compliance such as PCI DSS and GDPR.

2. Integration with LangChain and LangServe:

• The LLM model utilizes LangChain to manage natural language processing tasks between user inputs and outputs. We wrapped the model using LangServe, a Python framework well-suited for creating APIs, allowing seamless integration with Visa’s Campaign Atlas portal.
• The main chain checks the input configuration, ensuring that all user inputs meet the necessary security and compliance requirements before processing.
• The sub-chains handle tone augmentation, channel adaptation, and text customization while maintaining adherence to financial regulations.

3. Deployment Using AWS Services:

To deploy the API solution securely and efficiently, Neurons Lab leveraged several AWS services:

• AWS Elastic Container Service (ECS): Used for container orchestration, enabling scalable and reliable deployment of the application in a manner compliant with Visa’s operational requirements.
• AWS Identity and Access Management (IAM): Implemented to manage user access and permissions securely, ensuring that only authorized personnel can access sensitive components of the system.
• AWS Key Management Service (KMS): Utilized for encryption and secure key management, protecting sensitive financial data both in transit and at rest.
• AWS CloudTrail and AWS CloudWatch: Employed for monitoring, logging, and auditing system activity, which is crucial for compliance and security in the financial services industry.
• AWS Artifact: Accessed to obtain AWS’s compliance reports and agreements, providing evidence of AWS’s security controls and certifications necessary for regulatory compliance.

By leveraging these AWS services, we ensured that the solution not only met Visa’s operational needs but also complied with stringent financial industry regulations.

4. Integration with LangSmith:

• We integrated LangSmith for prompt storage logs and evaluation. This dashboard supports prompt management and provides an interactive playground while offering exceptional log and chain tracking capabilities.
• This integration allows Visa to maintain comprehensive records of all interactions, supporting regulatory requirements for data retention and auditability.
• It ensures transparency and traceability of all AI-generated content, which is essential for compliance with financial industry standards.

5. Security and Compliance Measures:

Throughout the architecture, Neurons Lab ensured that all components adhere to Visa’s stringent security requirements and industry regulations:

• Data Encryption: All data is encrypted using AWS KMS, both at rest and in transit, ensuring compliance with PCI DSS and protecting sensitive financial information.
• Secure Landing Zone: We established a secure AWS Landing Zone tailored for financial services, providing a multi-account AWS environment that is scalable, secure, and compliant.
• Compliance Alignment: The solution aligns with PCI DSS and GDPR requirements, ensuring that Visa’s operations remain compliant with international financial regulations.
• Monitoring and Logging: Implemented comprehensive monitoring using AWS CloudWatch, enabling real-time insights and alerts for any security incidents or compliance breaches.
• Audit Trails: With AWS CloudTrail, we provided detailed logs of all API calls and user activities, facilitating auditing and forensic analysis as required in the financial sector.

Client Feedback

“The Neurons Lab team demonstrated deep expertise in GenAI and worked diligently to build a tailored solution that has significantly improved our marketing personalization.”

Anna Melnychuk, Senior Marketing Manager VISA

About us: Neurons Lab

Neurons Lab is an AI consultancy that provides end-to-end services – from identifying high-impact AI applications to integrating and scaling the technology. We empower companies to capitalize on AI’s capabilities.

As an AWS Advanced Partner, our global team comprises data scientists, subject matter experts, and cloud specialists supported by an extensive talent pool of 500 experts. We solve the most complex AI challenges, mobilizing and delivering with outstanding speed to support urgent priorities and strategic long-term needs.

Ready to leverage AI for your business? Get in touch with the team here.