Global financial giant American Express has inaugurated its largest global office in Gurugram, India, covering nearly one million square feet. This new headquarters is located in Sector 74A and will see employees moving in stages, starting by the end of this month. 

The Gurugram campus is part of a strategic initiative to boost the company’s presence in India by combining global expertise with local talent. This investment aims to construct a modern, environmentally friendly office space. The facility has earned LEED Gold certification because of its sustainable architecture and construction. Environmental considerations are integral for the company, featuring energy-efficient LED lighting, smart building solutions, and electric vehicle charging stations. It also involves the use of renewable energy and effective waste management practices.

The design of the campus prioritises the health and well-being of its staff, incorporating green spaces, ergonomic workstations, and areas for relaxation such as quiet rooms and lounges. The intent is to create an environment that enhances productivity and employee satisfaction.

“The facility is a fitting reflection of our brand and the kind of workplace where our colleagues can thrive,” said Gagandeep Singh, senior vice president of global real estate and workplace experience at American Express. Sanjay Khanna, CEO and country manager for American Express in India, highlighted the campus’s role in fostering innovation and adding global value.

Additional features of the campus include a dynamic cafeteria with a live kitchen, a fitness centre, outdoor sports facilities, and terraces, all designed to promote community and collaborative engagement among employees. Access to campus amenities is streamlined through user-friendly apps, facilitating easy integration and flexibility for the workforce.

Apart from Gurgaon, American Express operates from multiple offices in India including Bengaluru, Mumbai, Chennai, Pune and New Delhi.

The company strategically leverages generative AI to improve its customer service and operational efficiency, opting for partnerships with established AI models rather than developing its own LLMs. This approach allows them to integrate such analytics and customer sentiment analysis into their services, improving predictive capabilities and personalising the customer experience. 

On the first day of the Association of Data Scientist’s (ADaSci) Deep Learning DevCon 2021 (DLDC), Radhakrishnan G, Head- Global Commercial and Merchant Risk Decision Science at American Express (Amex), spoke about how his company helps small businesses with real-time credit decisioning using machine learning and artificial intelligence.

Radhakrishnan is an alumnus of Management Development Institute, Gurugram. He kick-started his career as an Assistant Manager at Reliance Industries Limited before joining Amex in 2002. Throughout his almost two-decade-long ongoing stint at American Express, Radhakrishnan has been associated with risk management. His current role as the Head of Global Commercial and Merchant Risk Data Science and Risk Models across customer life cycle for card and non-card portfolios involves leading a team of more than 80 data and decision scientists across the globe. 

Leveraging ML at Amex

Radhakrishnan began his talk by introducing the audience by providing insights into the financial services company American Express. He revealed that as of 2020, Amex has 112 million cards in force, with 63,700 employees across the globe managing a worldwide billed business of $1.01 trillion and generating an annual revenue of $36.1 billion. 

• Customer service
• Customer management
• Responsible lending actions and risk decisions 
• Information management 
• Commercial underwriting 
• Loyalty marketing 

It does so with a privacy framework. 

Assessing Commercial Credit Risk

Further, Radhakrishnan talked about the dimensions for assessing commercial credit risk. He laid down the three facets for assessing risk: 

American Express leverages machine learning techniques and big technology for:

• Enhancing and managing new customer marketing
• Company profile: This includes information relating to the industry of the company, business tenure, the company’s management experience, its online presence, and public records. 
• Capacity: Under this, the company’s financial ratios (leverage and cash flows) are considered, its business revenue and limit on external trades. 
• Creditworthiness: Business credit scores, financial ratios (liquidity and profitability), historical performance on credit products and owner’s FICO. 

Radhakrishnan gave the example of the company ABC General Trading LLC seeking credit from American Express to substantiate this. Suppose the company functions in the e-commerce industry and has an income of $120,000 with an external limit of $30,000 and a debt of $100,000, with no past derogs and public records; and unavailability of revenue data. While the absence of past derogs and income of $120,000 works in favour of the company, low business tenure, involvement of large revolve behaviour on external trades, and the absence of enough data to predict its capacity to pay are con. Thus, the next ideal step would be to ask for more numbers and figure out the revenue and industry capacity. Once the bank statements are collected, it will provide American Express to get a better view of ABC General Trading LLC capacity to pay, leading to the approval for credit. However, the bank statement collection and reviewing was traditionally done manually and are a time-consuming process. This is where automating underwriting using ML and AI comes in. 

Credit Risk Models

American Express, or any other financial institution, needs AI automation to extract information from documents and ML technologies to leverage information from multiple sources to enable real-time decision making. 

• For assessing the company profile, Amex uses ML to arbitrate the best industry match. It uses name-matching algorithms to uncover related entities. 
• It uses ML for revenue estimation and optimal limit estimation. 
• Finally, it extracts information from documents using AI. For this, it uses time series intelligence using RNN (recurrent neural network). 

Filling information gaps using ML:

Industry estimation: 

Industry intelligence is an essential part of risk management, and ML techniques can help identify this with greater accuracy and speed. 

• Applicant’s company industry can be identified using industry estimation algorithms such as Smart SIC. Asking the customer or relying on the bureau is slow and tedious. Using ML, on the other hand, arbitrates between different sources based on the recency and accuracy of the source. On this, financial institutes can incorporate real-time feedback and enable business activities. This results in faster decision making. 

Revenue estimation: 

• Customer’s revenue can be estimated using a revenue estimation algorithm or Smart Revenue. This model uses data from multiple sources and makes a decision based on the recency and accuracy of the sources. 

Radhakrishnan further talks about the intelligence extraction pipeline. He says that non-standard text is cleaned up using custom text processing pipelines and Word2Vec. Furthermore, an LSTM model is used to predict transaction categories. Real-time features, along with other ML-powered features, are fed into ML models, resulting in superior decision making, more credit and faster time to market. 

A large volume of information travels through papers in an organisation, and knowing the structure of documents enables the extraction of relevant and useful data. Documents can be in a variety of forms and types, including native PDFs, web pages and scanned images. They also have a variety of templates, making document processing and interpretation a chore. Financial papers, in particular, such as audited reports, bank statements, financial reports, exchange filings, and so on, are critical documents that must be assessed for a variety of compliance and risk-related applications, including underwriting, risk rating, and more.

At the DLDC 2021 organised by the Association of Data Scientists (ADaSci) — Rahul Ghosh, VP of AI Research and Services at American Express AI Labs, spoke on AI-Powered Document Intelligence for Enterprises. Additionally, Rahul also gave a sneak peek into the R&D efforts at American Express and demonstrated how Document AI-enabled products could drive innovation and efficiency at scale.

To start with, document intelligence is nothing but the way that allows us to tap into the opportunities offered by unstructured document data and unlock the potential for faster and more informed decisions, increase operational efficiency, data governance, integrity and compliance, and enhance customer experience. Next, Rahul presented a general document AI stack for an enterprise. The stack is shown below.

This was followed by a simple explanation of the different types of documents, which includes:

• Form type documents are short documents. Typically, they are like one to two, or less than five pages, and have a very well-defined structure and a layout present —for example, invoices or bank statements. Use cases can be understanding invoice spend patterns, getting cash flow insights from bank statements, etc.
• Verbose documents are longer and have a lot more information put together in a single place. For example, a construction contract agreement – where you have data in the form of text, images, tables and more. Marketing creatives can review use cases before the campaign launch, highlight key clauses from documents, etc.

Talking about the challenges in information extraction from verbose documents, Rahul discussed a research paper for the case. A research paper consists of certain text, tables, titles, etc. Moreover, tables come with different types of cell formats, so, as the paper changes, the content will change completely. Hence, a simple rule-based or template-based approach will fail to extract table types that vary across different documents. There are few extraction challenges attached in form type documents, which includes:

• Form type documents also have diverse templates.
• Rule-based approaches can’t handle unseen templates and are difficult to manage.
• NLP-based approaches assign tags to each portion of the text, while CNN-based approaches can capture irrespective of variations in templates.
• Both NLP/CNN approaches have limitations for the cases where information is embedded in the spatial arrangement of the layout, not the text itself.

A typical extraction pipeline for extracting information out of documents is shown below.

Rahul said, “RCNN is an approach for extracting information from images and other things that are there. Although the approach was good, it was computationally expensive and had a longer training time. To remove this bottleneck, Microsoft research came up with a fast RCNN. With the fast RCNN, the problem was that at the testing time, you needed to explicitly tell where the objects were located. Then, somebody had to manually feed that information. That itself is a bottleneck.”

On top of faster RCNN, there are some improvement areas that can be taken into account. The addition of iterative refinement helps in identifying the location of objects. Even after this, there are more complicated document use cases where even with iterative refinement, one may not be able to capture all the details accurately. The results obtained by the American Express team are shown below.

Documents are the key source of unstructured data for an enterprise. AI-powered document intelligence can understand the structure of a document and extract contents, leading to significant process efficiency. In addition, deep learning approaches borrowed from image processing, computer vision and other related disciplines can significantly outperform naive rule-based approaches.

The premier global professional body of data science and machine learning professionals — the Association of Data Scientists (ADaSci), has come out with its much-awaited Deep Learning DevCon 2021 (DLDC). The two-day virtual conference on deep learning will be held on 23rd and 24th September, bringing influential professionals and researchers in the deep learning domain on a single platform.

There will be seminars, paper presentations, exhibitions, and hackathons at the summit. A full-day training on deep learning will also be offered, with attendees receiving a certificate of attendance. Moreover, it provides you with the unique opportunity to network with fellow attendees, talk to them, and meet companies virtually. DLDC 2021 has a strong lineup of speakers. Below are a few sessions that one must not miss:

1| State of AI and Deep Learning

When: 23 September, 09:45 – 10:30

By: Mohan Silaparasetty, Head – Technology Programs, Times Professional Learning

Artificial Intelligence and Deep Learning are evolving rapidly with some new and exciting advances every day. There is also a race among the countries to establish a dominant position in AI. This keynote is about the latest advances in Deep Learning and the latest applications of AI worldwide.

2| Understanding and Leveraging Differential Privacy

When: 23 September, 10:35 – 11:15

By: Manoj Kumar Rajendran, Principal Data Scientist, MiQ Digital India

With privacy being the buzzword in data collection and analysis, how should the tech world be prepared for a differentially private world? In this session, Manoj will present how Differential privacy allows digital companies to acquire and share aggregate information about user habits while protecting individual users’ privacy.

3| Lap Estimate Optimizer: Transforming race-day strategy with AI

When: 23 September, 11:20 – 12:00

By: Vikas Behrani, Vice President – Data Science, Genpact

Formula E has gained popularity as a sustainability-conscious sport that originates innovations to improve electric vehicles. The premise behind Formula E is not only that the cars are fully electric, but that the 11 teams, each with two drivers, compete in identically set-up, electric battery-powered race cars. The purpose of this exercise is to define the approach to use historical data to predict the number of laps a car would finish in 45min for an upcoming race. The team at Genpact built an ensemble model with a combination of an intuitive mathematical model and an instinctive deep learning model to predict the number of laps at the end of every race.

4| Dealing with Data imbalance in classification problems

When: 23 September, 14:00 – 14:40

By: Raghavendra Nagaraja Rao Data Science Academic Lead at Times Professional Learning

Most of the real-world data around classification problems are cursed with the imbalance of the target column. ML models are biased towards the majority class and result in incorrect predictions. Different techniques like up-sampling, down-sampling, SMOTE etc. are used to deal with such imbalance data which in turn enhances the performance of the classification model

5| To data prep or to data science. That’s the question

When: 23 September, 14:45 – 15:25

By: Swagata Maiti, Technology Architect, IP & Data Products & Shaji Thomas Vice President, Cloud & Data Engineering, both at Ugam, A Merkle Company

Both the experts, Shaji Thomas and Swagata Maiti from Ugam, a Merkle company, will deep-dive into seven techniques that will help data scientists build a scalable data platform. These techniques include automated data validation, reusable feature stores, streaming ingestion, the transformation of IoT sensor data, and more. Join the session to get an understanding of challenges faced by data scientists, how to address these challenges, and Snowflake capabilities that simplify building a scalable cloud data platform.

6| AI-Powered Document Intelligence for Enterprises

When: 23 September, 16:15 – 16:55

By: Rahul Ghosh, VP of AI Research and Services, American Express AI Labs

A huge volume of the information in an enterprise flows through documents and understanding the structure of documents allows extracting relevant and meaningful information. The focus of the talk is on Document AI, i.e., AI-powered automated analysis of documents. He will share the R&D efforts at American Express and demonstrate how Document AI-enabled products can drive innovation and efficiency at scale.

7| [Paper Presentation] Time Expression Extraction and Normalization in Industrial Setting

When: 24 September, 12:05 – 12:25

By: Piyush Arora, Senior AI researcher, American Express AI Labs

We present TEEN, an industry-grade solution to the problem of time expression extraction and normalization (Timex). Extraction and normalization of temporal units is a challenging problem due to several factors, e.g.,

• same time units may be expressed in different ways
• inherent ambiguity in natural languages leading to multiple interpretations
• context-sensitive nature of natural languages

8| [Workshop] Industrializing AI/ML: Hands-on Model Deployment

When: 24 September, 14:10 – 16:10

By: Jatindra Singh Deo, Senior Technical Architect & Abhilash NVS, Data Scientist, Genpact

This working session will look at a hands-on approach to pipelines and their orchestration using TFX/Airflow. API/SDK approach to model deployment as a web service with Flask Pre-requisite: Laptop with a minimum of 8 GB ram with Windows/Linux/macOS Anaconda (individual edition) installed good internet connection to download coding stubs and pretrained model Docker desktop installed Basic familiarity with google collab with a google account.

9| [Paper Presentation] Global-Local Scalable Explanations Using Linear Model Tree

When: 24 September, 16:40 – 17:00

By: Narayanan Unny E., Head of Machine Learning Research, American Express AI Lab

A Generative Adversarial Network is employed for generating synthetic data, while a piecewise linear model in the form of Linear Model Trees is used as the surrogate model. The combination of these two techniques provides a powerful yet intuitive data structure to explain complex machine learning models. The novelty of this data structure is that it provides an explanation in the form of both decision rules and feature attributions.

10| [Paper Presentation] Predicting Custom Ad Performance Metric using Contextual Features

When: 24 September, 17:05 – 17:25

By: Divyaprabha M, Data Scientist, MiQ Digital

Digital advertising enables advertisers to promote their products on various online and digital channels. Real-Time Bidding is an advanced advertising method that allows advertisers to target potential buyers and acquire ad space on websites in the form of programmatic auctions. The paper proposes a machine learning-based approach to predicting future ad-campaign performance by focusing on contextual features such as browser, operating system, device type, and so on.

Grab the chance to interact and learn from the experienced bunch of data scientists going to present their sessions in the coming days. For more details and schedules, one can visit here.

The American Express Company, also known as Amex, is an multinational banking and financial services corporation headquartered in New York City. The company is a financial giant with 114 million cards in force, 64,000 employees worldwide and $1.24 trillion worldwide billed business. 

For Amex, there are billions of transactions going through its system every month. With such a volume of card transactions, it is not just the dollar amount which is high, the network also generates massive amounts of data, including trillions of transactional data combinations which need to be analysed in almost real time

In such a situation, advanced techniques in machine learning and deep learning are essential, and the company uses them extensively in detecting and preventing frauds. But how does Amex do that on such a large scale?

To understand better, Analytics India Magazine reached out to Dr Manish Gupta, Vice President, Machine Learning & Data Science Research, and Head of Risk COE Bangalore at American Express.

According to Dr Gupta, Amex has built and deployed best in class ML models, leveraging state-of-the-art technologies like deep learning, machine learning for various business decision-making processes. 

How would you define your leadership role at American Express in terms of the initiatives you are driving?

Dr Gupta: I lead the machine learning and data science team that builds state-of-the-art machine learning solution frameworks and leverages them in risk and analytics decisions on behalf of our customers across the globe. 

How did the journey of Amex in machine learning begin?

Dr Gupta: When it comes to American Express, early on, our leadership recognised the value of big data analytics and data to drive better decision-making and ultimately support risk management. As a 170-year old company, it has instilled a culture that embraces innovation, combining its technology and infrastructure with new computing techniques to make better and faster decisions. 

This mindset is a critical advantage to American Express since it began its work in machine learning nearly 10 years ago. We have recently started to explore deep learning techniques to generate the next set of data innovations by deriving intelligence from the global, integrated network of American Express. We are challenging ourselves to leverage deep learning capabilities to bring human-level intuition from this structured data. Deep learning has helped us to improve credit and fraud decisions and elevate the payment experience for millions of Card Members across the globe.

How does American Express use machine learning to maintain a competitive edge? Can you elaborate with a few examples?

Dr Gupta: American Express has three unique advantages: its data, advanced machine learning and deep learning techniques, and decision science talent. American Express has deployed what we believe to be one of the most advanced machine learning systems in the financial services industry. Our machine learning algorithms across credit and fraud risk are used to monitor in real-time more than $1.2 trillion worth of transactions annually around the world. We use sophisticated tools and methods to evaluate data available only to American Express since we operate as a card issuer, merchant acquirer and a network.

As a result of deploying machine learning within our fraud models back in 2014, we have continued to maintain the lowest fraud rates in the credit card industry (half that of our competitors), according to the February 2020 Nilson Report. Having our card members backs is our top priority and keeping our fraud rates low is key to achieving this goal.

Please tell us about the deep learning approach at Amex to counter fraudulent transactions. Can you discuss some brief details about data science techniques you use to create fraud analytics solutions? 

Dr Gupta: We have been leveraging advanced embedding techniques and deep learning generative and sequential models to better train our models to identify genuine transaction patterns from fraudulent ones, which has led to a significant drop in fraud losses and improvements in customer experience. We have recently invested in high-performance deep learning infrastructure based on the GPUs to implement real-time inferencing frameworks.

How do you test your models to simulate real-time frauds?

Dr Gupta: Our models are real-time, allowing us to be proactive about fraud prevention. Each incoming transaction is evaluated for its propensity of being fraudulent so that we can stop it before any losses are incurred. Our state-of-the-art decision engines help in evaluating models using complex feature calculations in a few milliseconds.

What according to you is the most significant breakthrough in AI/ML in the last five years, and why?

Dr Gupta: I’ll give you my top three. First, we have the Deep Learning Frameworks such as Tensorflow, Pytorch etc., which helped democratise AI rapidly. Now, people can build AI models very easily. Second, there are GPUs that make AI economical: One of the big reasons AI is now such a big deal is because of the cost of crunching, so much data has become affordable. Finally, there are Neural Architectures such as BERT, ResNet etc. has accelerated the innovation in NLP and Computer Vision Domains.