StartUP Webinar Series Recap: how Fintech can harness the power of AI/ML for document processing
Document processing remains a huge hurdle for businesses today. Manual document processing is a time and cost consuming process. Companies in the Banking, Financial Services and Insurance industries have to handle reams of invoices, loan applications, contracts and KYC processes.
However, the solution to these problems often involves manual labour, third party solutions, or building document processing services themselves – all of which have significant drawbacks in terms of accuracy, security, cost and compliance.
This is where AWS’ AI/ML services come into play. There are 3 main reasons why customers choose AI/ML solutions from AWS – choice, flexibility and innovation. Moreover, over 100,000 customers are using ML services, AWS has built 250+ new capabilities for AI/ML in the last year and 92 percent of deep learning in the cloud is on AWS.
To illustrate how companies can leverage AWS’ AI/ML services to streamline document extraction processes, YourStory, in association with AWS, hosted the StartUP webinar session on How Fintech can harness the power of AWS AI/ML.
The session covered automated document processing, identity verification, customer success stories, the three-level AWS ML learning stack, AWS’ AI/ML services and live demos.
Here’s what you can expect in the session:
Application of Automated Document Processing
Arun started the session by explaining what automated document processing is – extracting useful information from documents. He shared some the use cases where document processing was helpful, such as loan documents, claims processing, regulatory filings and more. For each of these use cases, he shared that there are 2 actions they do – Optical Character Recognition (OCR) and Natural Language Processing (NLP) to mine and extract useful information from the documents. He also covered the kinds of challenges organisations faced when implementing a document understanding solution, like the need for manpower, restriction in scalability etc.
Applications of Identity Verification
Identity verification has many applications across the fintech industry – including new account opening (KYC), ATM, payments, loan processing and more. This section looked at the different ways facial recognition could be used for identity. Arun cited the recent RBI concession of allowing people to do video KYC, allowing for comparisons between an individual’s photograph and their video. He also shared how banks are considering facial recognition as an additional layer of security for verification.
Arun also highlighted the key advantages of facial recognition, such as security (recognition based on the customer, instead of PINs and passwords) and a simplified customer experience. He then took attendees through implementing facial recognition through the lens of the KYC process. Additionally, he covered key considerations like security and model governance when building fintech applications.
This part of the session covered different customer examples, breaking them down into the situation the customer faced, and how AWS helped. For instance, how Biz2x – a fully-fledged digital lending platform used Amazon Textract and Amazon Rekognition to automate and optimise their KYC and document extraction processes. The session also covered use cases from Perfios (who used Amazon SageMaker to build a product for Debit Card Scores). Finally, the session looked at intuit – the global financial platform, who improved the accuracy of their data extraction from complex financial forms.
AWS AI/ML services
After covering the importance of document processing, identity verification and the uses cases, Arun spoke about the different AI/ML services offered by AWS. He displayed the three-level AWS Machine Learning stack, explaining that it was designed so that everyone – from data scientists to developers – could leverage the platform and build applications with ease. The bottom layer consisted of framework support and infrastructure GPUs and CPUs to choose from, as well as low-cost alternatives that offer similar performance. The middle layer consisting of Sagemaker Studio – 20+ services across different sections of ML workflow. The top layer – AI services – which are API driven.
Arun also specified the four services that will solve the challenges brought about by document processing. He discussed the functions, features, capabilities, followed by quick demos of each service.
Amazon Textract: an OCR service, it allows customers to extract data from virtually any document, without depending on document structure or orientation.
Amazon Rekognition: an AI service, which allows you to mine insights from images as well as videos. Users can identify labels and text in the images and videos. They can also identify and compare people for facial similarities. When it comes to identity verification, the service uses a few features as parameters, such as demographic data, facial landmarks, image quality, general attributes and more.
Amazon Comprehend: A Natural Language Processing service, Comprehend allows users to extract information from text documents.
Amazon Augmented AI: A service, from Amazon SageMaker, which allows users to bring in a human to review and validate your predictions or extractions from documents.
The session also featured 2 demos of use cases and reference architecture, namely:
Video KYC Reference Architecture: The demo covered the following features:
- Liveliness Detection
- Upload Documents
- Validation and Summary
Automated Loan Processing: The demo covered 2 components:
- Filling out the loan application
- Extracting data from documents and presenting it
How AWS can help
The session wrapped with a section on the support programmes and campaigns that AWS provides, in terms of workshops, immersion days, training, certification, AWS credits and more.