Mastercard, Visa, And AI Supremacy
AI is already changing the fintech industry, but is this relationship really working?
It seems like every day, we're hearing about another major player in the industry jumping on the AI bandwagon. And who can blame them? The way AI is transforming every sector is simply too great to ignore.
But as someone closely following the AI space, I've noticed a trend lately: the increased adoption of generative AI in fintech. 57% of banking and financial markets CEOs believe that gaining a competitive edge will depend on who has the most advanced generative AI.
From chatbots and fraud detection to risk assessment and beyond, AI is changing how financial institutions operate and interact with their customers.
However, AI implementation is not without challenges. Lack of quality data, ever-changing regulations, and bias in the training data are some of the major hurdles financial institutions face.
In today's newsletter, we'll explore:
Recent developments on how AI is used in fintech
Common challenges of implementing AI in fintech
How Indecomm automates mortgage document processing with Amazon Textract
Let’s get started!
AI in fintech: Recent developments and their impact
AI is automating everything from document verification to client acquisition. To better understand the real-world implications let's see some big developments.
1. Visa invests $100 Million in generative AI for commerce and payments
Visa has launched a $100 million generative AI venture initiative. This initiative aims to invest in companies developing generative AI technologies and applications specifically for commerce and payments.
Led by Visa Ventures, the company's global corporate investment arm, this move signifies Visa's commitment to driving innovation in the industry.
2. Mastercard uses generative AI for faster fraud detection
Mastercard is deploying AI predictive technology to scan transaction data across billions of cards and millions of merchants at exceptional speeds.
This advancement has enabled the company to double the detection rate of compromised cards, reduce false positives by up to 200%, and increase the speed of identifying at-risk merchants by 300%.
3. Visa launches gen AI-powered fraud solution
Visa has introduced a new generative AI-powered fraud solution called Visa Account Attack Intelligence (VAAI) Score to combat account and enumeration attacks in card-not-present transactions.
The VAAI Score assigns a real-time risk score to each transaction, using Gen AI to learn normal and abnormal transaction patterns. Compared to other risk models, it has helped reduce the false positive rate by 85% and has been trained on over 15 billion VisaNet transactions.
4. BBVA's strategic partnership with OpenAI
BBVA, a leading European bank, has taken a major leap in AI by partnering with OpenAI. It aims to deploy generative AI across its operations and empower its employees with advanced tools.
The bank has prioritized around 100 projects to be developed using OpenAI's technology, focusing on expediting processes, improving productivity, and enabling innovation. It has also distributed 3,000 ChatGPT Enterprise licenses to its employees.
As more companies embrace AI solutions, we can expect improved security, customer experiences, and operational efficiency.
However, the successful implementation of AI in fintech also requires addressing its challenges.
Common challenges of implementing AI in fintech
Implementing AI in solutions and workflows is not always easy. Organizations often face several hurdles to scaling AI solutions effectively.
But with the right strategies and solutions, you can overcome these challenges. Let's explore some of these challenges and potential solutions.
1. Data dilemmas: quality, access, and drift
AI models are only as good as the data they're trained on. Data is often scattered across siloed databases in unstructured formats, making it a nightmare to access, integrate, and prepare for AI use cases.
Plus, many AI models in fintech are initially trained on historical data, which can lead to performance degradation if the statistical characteristics of the data change over time (a.k.a. data drift).
Potential solutions:
Establish robust data governance, metadata management, and data integration pipelines.
Use APIs to automatically fetch fresh data from different sources and build a custom storage solution to quickly and efficiently retrieve historical and new data in the required format.
Employ synthetic data generation techniques to imitate real-world data characteristics and patterns, augmenting the available dataset and improving model performance.
2. Navigating the regulatory maze
The financial sector is heavily regulated, with strict rules around data privacy, security, model transparency, ethical practices, and audit trails.
Non-compliance can result in massive fines, and ensuring that AI solutions adhere to these rules is a significant challenge.
Moreover, AI's "black box" nature makes it difficult to explain results and instill confidence, especially for high-stakes decisions like lending approvals or insurance underwriting.
Potential solutions:
Follow model governance frameworks like OMRM by embedding compliance checks into the AI lifecycle, starting from the design phase.
Collaborate closely with legal and compliance teams to validate AI models' compliance with relevant guidelines.
Adopt explainable AI techniques that enable traceability into model decision-making logic.
Ensure human oversight for AI systems handling critical processes and use simplified machine learning techniques like decision trees that are more interpretable.
3. Bias from training data
In 2019, Apple Card was investigated for claims that its credit assessment algorithm was gender-biased against women. Inherent or not, it was a violation.
AI models can inadvertently perpetuate and amplify historical biases in training data related to gender, race, income levels, etc. If the training data reflects discriminatory patterns from the past, it can lead to unfair outcomes, such as lending.
Potential solutions:
Implement bias testing as part of model validation.
Use debiasing techniques like adversarial debiasing, counterfactual evaluation, reweighing training data, and discrimination-aware data mining.
Learn more about how AI is used in the fintech industry here.
How Indecomm automates mortgage document processing with Amazon Textract
Indecomm, a leading SaaS provider in the mortgage industry, has developed an innovative solution called Intelligent Document Extraction (IDX) to streamline the time-consuming and critical tasks associated with mortgage loan origination.
Indecomm has significantly reduced the cost and time spent reviewing mortgage origination documents using Amazon Textract, a machine learning service from Amazon Web Services (AWS).
IDX uses Amazon Textract to automate document review and extract data from images and text for analysis, making the process more efficient and accurate.
Indecomm chose Amazon Textract for its scalability, integration with serverless tools like AWS Lambda, and cost-effectiveness.
Indecomm's approach to automating complex mortgage document processing is threefold:
Intelligent document classification: IDX goes beyond simple optical character recognition (OCR) by accurately identifying and classifying various types of mortgage documents.
This enables the system to understand the context and purpose of each document, ensuring proper data extraction and analysis.
Data extraction and validation: Once documents are classified, IDX extracts relevant data fields and validates the information against predefined rules and criteria.
This process ensures that the extracted data is accurate, consistent, and ready for further analysis.
Data enrichment and integration: IDX enriches the extracted data by combining it with information from other sources, such as borrower databases or external APIs.
This enriched data is then seamlessly integrated into Indecomm's Genius product suite, which includes IncomeGenius, DecisionGenius, and AuditGenius, to provide a comprehensive view of the mortgage loan application.
By implementing IDX with Amazon Textract and AWS Lambda, Indecomm has achieved impressive results:
Data classification and extraction time reduced from 30 minutes to 5–7 minutes for a 100-page document
50–60% reduction in manual document intervention
The average total cost per page processed lowered to just 2 cents
Automated scaling with parallel processing and serverless architecture
100% data classification accuracy and 97% data extraction accuracy
Moreover, Indecomm's clients have experienced significant benefits from using IDX, including a 50% reduction in underwriting and mortgage origination time, improved data accuracy, and a predictable, affordable costing model.
The integration of IDX with Indecomm's Genius product suite has enabled lenders to optimize their business processes, boost efficiency, and focus on providing a better customer experience.
As the mortgage industry demand for faster, more accurate loan processing grows, solutions like Indecomm's IDX will play an important role in the future of mortgage lending.
Find out more about Indecomm's ML solution here.