Software runs every part of an enterprise, and the leaders are the ones who leverage cloud, data, and AI incorporate into everyday operations. Suprakash Dutta is a Senior Solutions Architect at Amazon Web Services who often serves as a CTO-style advisor to major enterprise accounts. He helps Fortune 500 companies modernize and scale, aligning architecture with business goals so systems run faster, more reliably, and at lower cost. He designs roadmaps, leads migrations, and applies AI and machine learning to automate work and improve decisions. He also translates complex technology into clear choices for executives and guides teams from idea to production. In this conversation, we dig into Intelligent Document Processing and the results it delivers. We talk through the key challenges companies often face when transitioning to automated document processing, and we look ahead to where generative AI, multimodal AI, and composable architecture are taking digital transformation next.
Suprakash, you’ve developed an unique solution for automating document processing — Intelligent Document Processing (IDP). Can you tell us how this technology helps businesses eliminate paperwork and accelerate processes?
For starters, I want to note that everything we are discussing represents my personal views and does not reflect the views of my employer — this is an important point. Intelligent Document Processing (IDP) goes beyond traditional Optical Character Recognition (OCR), which basically just converts text from images or scanned documents into editable, searchable text. By contrast, IDP combines managed services into an end-to-end workflow, covering every stage of document processing. This comprehensive pipeline is what makes our approach unique, and it effectively improves how organizations handle documents.
The workflow starts with the secure cloud ingestion of PDFs, images, and scans. Then we classify documents with Amazon Comprehend, extract tables, forms, and text with Amazon Textract, and extract custom entities that basic OCR would miss. We enable natural language queries such as “What is the invoice total?” to help users get accurate answers across formats.
Then we enrich and govern documents by automatically redacting sensitive data using position data, classifying by content, and pulling insights from unstructured text. When model confidence is low, Amazon A2I routes items for quick human review.
As a result, organizations can process documents at scale with minimal human input. In practice, the timeframe drops from days to minutes, accuracy and compliance improve, and teams can instead focus on higher value work.
What enabled you to achieve a noticeable reduction in document processing time?
The reduction comes from four main design choices. First, we automate manual steps. For example, classification now runs with Amazon Comprehend. We train a model on your document types, and it then classifies your organization’s specific sets. Second, we process documents in parallel on serverless services that scale to thousands at once. By contrast, traditional processes often involve multiple handoffs between departments that take days. Third, we use purpose built AI in Amazon Textract (AnalyzeExpense/AnalyzeLending) for higher accuracy and fewer manual corrections. Finally, we flag only low confidence content for human review. Together, this does more than speed things up; it transforms the workflow.
Can you share examples of how implementing IDP has helped your clients reduce time and costs — you can omit specific names.
Absolutely. There are several compelling public examples across different industries that demonstrate the real impact of intelligent document processing.
A major mortgage lender publicly shared impressive results from their AI-powered document processing platform. They automatically identify nearly 70% of the more than 1.5 million documents they receive monthly, which saved over 5,000 hours of manual work for underwriters in just one month. Of the 4.3 million data points they extract from documents like W-2s and bank statements, nearly 90% are automatically processed, saving an additional 4,000 hours of manual work. This automation helped reduce loan interactions by nearly 25% year-over-year and decreased closing times by 25%, enabling them to close loans nearly 2.5 times faster than the industry average.
A leading UK financial services firm transformed their document processing for over 5 million pension customers who submit more than 1 million paper documents annually. Using Amazon Textract and Amazon SageMaker, they can now process 1,000 documents concurrently in just 30 minutes with their serverless architecture. This improvement in speed and accuracy helped them meet regulatory SLA requirements more consistently and allowed them to redeploy five employees to higher-value work, significantly improving business productivity.
A global fintech company automated their proxy voting data processing tool that handles SEC filings and extracts 130 proxy data points from 300,000 meetings. They achieved 70% automation accuracy using Amazon Textract and Amazon Comprehend, expecting to save thousands of hours annually. Their automated process reduced manual delivery time from 1-7 days to near real-time processing, enabling them to deliver faster insights to customers while maintaining their high accuracy standards.
A major health insurance provider automated 80% of their claims processing workflow using Amazon Textract, transforming a manual process that took an average of 20 minutes per claim. They now process thousands of claims daily with automated extraction and classification, expecting to reach 90% automation or higher.
These public case studies demonstrate consistent patterns across industries: dramatic time reductions, improved accuracy, cost savings, and the ability to redeploy staff to more strategic work. The key is that these aren’t just efficiency gains—they represent fundamental workflow transformations that create competitive advantages.
Tell us more about the workshop you’re organizing with your team. What is the goal of this project, and who will benefit from it?
Our Intelligent Document Processing (IDP) workshop is a practical, hands-on program that explains document automation and helps teams build their own IDP solutions with AWS AI services. Basically, it closes the gap between AI concepts and real implementation.
First up, Level 1 covers the fundamentals: classification with Amazon Comprehend, extraction with Amazon Textract, enrichment with custom entities, and human review with Amazon A2I. Participants build these step by step with real world documents like invoices and medical forms.
Next, Level 2 focuses on industry use cases—insurance claims, mortgages, and healthcare records—adapting the core components to sector requirements and compliance needs.
Then, Level 3 is about scale and operations: use AWS CDK and IDP constructs to build production ready, end to end workflows that handle enterprise scale reliably.
Finally, Level 4 adds generative AI with Amazon Bedrock, including zero shot classification, data normalization with large language models, and enrichment like summarization and translation.
What makes this workshop valuable is its open source model. We’ve made all materials publicly available on GitHub, allowing organizations to adapt and extend the solutions to fit their needs. The hands-on approach ensures participants both understand the concepts and gain practical experience they can apply right away in their organizations.
What are the key challenges companies typically face when transitioning to automated document processing, and how can they be overcome?
Companies run into a few common challenges; each of them needs a clear plan to address it.
First, document variability. Files come in all formats and quality levels. The solution is to use a flexible classifier and start with your most common types; Amazon Comprehend learns from examples and gets better over time, even with imperfect documents.
Second, legacy systems. Many existing tools weren’t built for AI. We address this with an architecture that uses well-defined APIs and standardized outputs to connect to almost any system. The key is to treat IDP as a modular component that enhances existing workflows and to expand it step by step as confidence grows.
Third, accuracy and trust. People worry about machine-extracted data. In this case, we use a hybrid approach: AI plus humans in the loop. With Amazon A2I thresholds, only low-confidence items go to review, and that share drops as models improve.
Fourth, sensitive data and compliance. Regulated organizations worry about documents that contain sensitive information. We address this with security controls, including automatic redaction, fine-grained access controls, and audit trails. Our architecture keeps data sovereignty by processing documents in the customer’s own AWS account, not in third-party services.
Last but not least, change management. Bring stakeholders in early, show how automation removes tedious work, and set up a center of excellence across business and IT.
By addressing these challenges systematically, organizations can move to automated document processing and realize the benefits of speed, accuracy, and scale.
By the way, what brought you into the field of digital transformation? Has this always been your career path?
My career path has not been a straight line. I started as a software developer at DXC Technology (formerly CSC), building cloud orchestration and management platforms. That taught me code, and how tech drives real business efficiency.
Work on CSC’s Agility Cloud Platform sparked the interest. Then at Fujitsu America, I architected AWS and Azure landing zones for a legacy bank and migrated more than 150 production applications. Seeing cloud modernize a bank—and advising executives—showed me transformation is both technology and organizational alignment.
At AWS, I leaned into machine learning and AI for intelligent document processing. On a telecom project, we achieved over 98% classification accuracy and cut onboarding time by more than 90%, which proved AI can remove manual work while improving accuracy and compliance.
Across roles—from developer to project lead to solutions architect—I’ve focused on both technology and business outcomes. Work with retail enterprises reinforced that, especially leading a cloud native point of sale platform for a large specialty retailer across 2,400+ locations.
Effective digital transformation needs this blend: understanding what is technically possible and what drives business impact. That is why I focus on intelligent document processing and generative AI. They sit at this intersection and use advanced capabilities to solve real business challenges across many industries.
Speaking more broadly about digital transformation, what technologies and approaches do you believe will be defining in the coming years?
The next wave of digital transformation will be driven by several converging technologies and approaches that change how organizations operate. Among them, generative AI stands out. It’s already changing how we handle document processing and knowledge work. With Amazon Bedrock foundation models, we can do things that were hard with traditional ML, like zero shot classification, automatic data normalization, and natural language reasoning. Bedrock Data Automation also streamlines building applications and automates workflows for documents, images, audio, and video, so teams ship multimodal, data centric automation faster. In short, a few trends are converging, and generative AI leads the pack.
We’ll also see AI copilots and agents across the business. Our virtual IT troubleshooting assistant with Amazon Q Business is a good example: it doesn’t replace IT staff; it gives them fast access to scattered knowledge and speeds up resolution.
Multimodal AI matters too. Processing text, images, audio, and video together changes how we handle complex content, from property assessments to medical records.
Beyond AI, ambient computing and immersive tech like spatial computing, AR, and VR will reshape how people work together and automate physical tasks.
And architecture-wise, composable beats monolithic. Organizations that thrive will treat these technologies as one integrated capability fabric, not separate initiatives. Winners will build digital platforms that combine them to deliver new customer experiences and operating models. This requires a shift in architecture: from building single applications to cultivating digital ecosystems that evolve continuously. That’s the foundation for whatever comes next.
