Rushabh Sheth, CEO of Docsumo, shares how Document AI is revolutionizing manual workflows and enabling 90%+ automation for high-volume, complex documents.
Rushabh, we’re delighted to have you at AITechPark. Could you please share a bit about your professional journey and what inspired you to co-found Docsumo?
Rushabh Sheth: I started my career working on consulting projects, and in 2019 my co-founder Bikram and I were trying to solve a document data capture problem for a Canadian bank. In the process, we realized this was a much bigger, widespread pain point – virtually every large enterprise was struggling with manual data entry and unstructured documents. Bikram had firsthand experience with this pain; he actually worked as a data entry clerk early on, so he knew how tedious and error-prone those manual processes can be. That really inspired us to build Docsumo as a solution to free businesses from mundane paperwork and help them automate document workflows using AI.
It wasn’t a straight path by any means – we faced our share of early challenges. In fact, about nine months in, we were nearly out of cash and had even told our two employees to start looking for other jobs. Then we caught a lucky break when we landed our first customer, PayU, which gave us the lifeline (and confidence) to keep going. From that point, our journey gained momentum: we got into the Techstars accelerator in 2020, which really broadened our ambitions to build a global SaaS product. So the inspiration behind Docsumo comes from those experiences – seeing a critical problem up close, almost falling down but then validating our vision with that first customer. It’s been a journey fueled by the belief that we can help organizations by automating the tedious parts of work and letting people focus on more meaningful tasks.
How did the collaboration with National Debt Relief come to life, and what do you think makes this partnership so impactful?
Rushabh Sheth: National Debt Relief (NDR) had a massive challenge that made our collaboration a natural fit. They were dealing with over 350,000 debt settlement letters a year, handled by a team of more than 50 people. When we learned about this volume of paperwork, we immediately saw how our Document AI technology could make a huge difference. We reached out and demonstrated that we could automate the extraction and processing of those settlement letters. Essentially, the partnership came to life from a shared recognition – NDR needed to streamline a labor-intensive process, and we had the exact solution to do that.
What makes this partnership so impactful is the scale of improvement we’ve achieved together. By deploying Docsumo, NDR is now automating over 90% of the processing of these complex letters. To put it in perspective, what used to take a staff member more than 20 minutes to review and input now takes under 20 seconds with AI. That speed-up is incredible. It means NDR can help a lot more people get out of debt, much faster than before. In an industry where timely help can change lives, cutting the response time down so dramatically is game-changing. This efficiency not only reduces operational strain on NDR’s team, but it directly benefits their customers – people seeking financial freedom get resolutions quicker. For us at Docsumo, it’s immensely rewarding to see our technology enabling NDR to impact so many individuals. We’re basically helping “fast-track” financial freedom for thousands of people, and that shared mission is what makes the partnership special.
Docsumo’s Document AI technology has brought remarkable efficiency to processing debt settlement letters. Could you kindly walk us through the standout features of the technology and how it achieves such impressive results?
Rushabh Sheth: Absolutely. At its core, Docsumo’s Document AI is an end-to-end platform that takes an unstructured document and processes it with minimal human intervention. One standout feature is how it automatically extracts all the crucial information from a debt settlement letter. For NDR’s use case, for example, our system pulls out the customer’s name, account details, creditor information, settlement amount, and even the payment installment plans from each letter. And it doesn’t stop at simple extraction – it understands the context. The platform performs smart validations on the data, like checking that phone numbers are in the right format, verifying that the totals of payments add up correctly, and even doing straight-through processing (STP) checks on the data consistency. These built-in checks are a big reason we can guarantee both speed and accuracy.
To achieve such results, we use a combination of advanced OCR and AI techniques. First, when a document comes in, we pre-process it – this means cleaning up the image, splitting or rotating pages if needed, reducing noise – so that the text is as clear as possible for extraction. Then our OCR module, which is optimized for a variety of fonts and scans, kicks in to extract the text with a very high degree of accuracy. On top of that, we have a proprietary NLP-based component that understands the structure and context of the document, so it can locate key-value pairs and line items even if every creditor’s letter looks a bit different. After extracting the data, we run it through a rules-based validation engine that applies contextual checks – for example, it will automatically flag if a sum of payments doesn’t match the total settlement amount, or if a date is out of expected range. This ensures the output data is reliable. Finally, integration is seamless: Docsumo can feed the results directly into NDR’s systems (like their CRM or Salesforce) via API, so the whole process from receiving a letter to having structured data in the workflow is automated end-to-end.
All these features working together are how we achieved the drastic improvement in efficiency. By the time we’re done, each letter is processed in seconds with exceptional accuracy. In the case of NDR, we went from a 20-minute manual task to a 20-second automated task without sacrificing quality. It’s a combination of clever engineering and machine learning – ingesting documents directly, extracting and interpreting their contents, validating the information, and handing it off to downstream systems. This holistic approach is what makes Docsumo’s Document AI so powerful in streamlining document-heavy workflows.
Achieving over 90% automation in processing complex documents is a significant accomplishment. What were some of the challenges Docsumo faced, and how did your team overcome them?
Rushabh Sheth: One of our biggest challenges was dealing with the sheer variety of unstructured and semi-structured documents across different industries. Debt settlement letters were one example, but we’ve also worked on automating lease abstracts, rent rolls, and medical records, each with their own unique complexities.
For instance, in commercial real estate, lease abstracts summarize critical information from long lease agreements. The challenge here is that no two leases are the same—terms can be hidden in legal jargon, spread across multiple clauses, and formatted differently across landlords and property managers. A traditional rule-based approach simply couldn’t handle this variability. So, we built context-aware AI models powered by Large Language Models (LLMs) to read and extract key details like rent escalation clauses, lease terms, and renewal options—even if they were worded differently in each document.
Another industry where we saw similar challenges was healthcare, specifically in processing medical records and insurance claims. Medical documents are often dense, filled with abbreviations, and written in free-form text by different physicians. Our AI had to not only extract structured information but also understand the medical context—for example, differentiating between “history of diabetes” (a past condition) and “at risk for diabetes” (a precautionary mention). Here, LLMs played a crucial role by bringing contextual understanding, allowing our AI to make sense of such nuances.
A key breakthrough for us was combining traditional NLP models with LLMs. Our legacy OCR and NLP pipelines were great at extracting data, but they lacked deeper comprehension. With LLMs, we were able to contextualize information—so instead of just pulling numbers from a rent roll, the AI could now explain what they meant in relation to lease terms, or instead of just extracting medication names from a prescription, it could determine the prescribed dosage and frequency.
Another major hurdle was ensuring high accuracy while scaling. In financial and healthcare documents, a small error can have serious consequences. For that reason, we built a multi-layered validation system, where AI performs automated checks on extracted data (e.g., checking if lease terms match property management systems or verifying patient insurance details against policy records). When in doubt, the system flags issues for human-in-the-loop verification, ensuring accuracy without slowing down automation.
Ultimately, our ability to blend LLMs with domain-specific AI models allowed us to move beyond basic data extraction and into a deeper level of document comprehension. This shift has been key to achieving high automation rates while maintaining accuracy across a variety of document types.
National Debt Relief’s Whole Human Finance™ approach is truly inspiring. How does Docsumo’s technology complement and enhance their mission?
Rushabh Sheth: National Debt Relief’s Whole Human Finance™ philosophy is about giving people real financial freedom, not just settling debt. Our role in this mission is simple: remove inefficiencies so that NDR’s team can focus more on helping clients, not processing paperwork.
Before Docsumo, handling debt settlement letters was time-consuming, often taking 20+ minutes per letter. Now, with 90%+ automation, NDR’s staff can focus on providing financial guidance instead of entering data manually. This means faster resolutions for clients, more personalized support, and scalability for NDR—allowing them to help more people in less time.
Beyond just data extraction, our AI provides validation and compliance checks, reducing errors and ensuring clients get the right financial solutions. By removing the bottleneck of document processing, we enable NDR to stay focused on what truly matters—empowering individuals toward a debt-free future.
In your view, how is Document AI reshaping the future of workflow automation, particularly in industries like financial services?
Rushabh Sheth: Document AI is fundamentally transforming financial services by eliminating manual data entry and enabling real-time decision-making. Financial institutions deal with massive volumes of unstructured data—bank statements, invoices, debt settlement letters—and AI is making these workflows faster, smarter, and more cost-efficient.
Take Hitachi Payment Services, which used to manually process 3,000+ bank statements monthly across 50+ formats. With Docsumo, they cut processing time from 2 hours to 2 minutes per statement, achieving 99% accuracy and saving 6,000+ hours per month. In lending, Grid Finance reduced manual effort by 90%, accelerating loan approvals while ensuring 93%+ accuracy in financial document verification. National Debt Relief and ClearOne Advantage automated 350,000+ debt settlement letters annually, cutting per-letter processing time from 20 minutes to under a minute, improving response times for their clients.
Beyond just extracting text, AI is now “understanding” documents, thanks to Large Language Models (LLMs). This allows financial firms to detect anomalies, validate transactions, and automate complex decisions in real-time—whether it’s flagging risks in a loan application or reconciling bank records instantly. Expense management firms like Huddle have also leveraged AI to automate bill processing, achieving 95.5% accuracy and reducing per-bill processing time from 15 minutes to 1 minute.
With AI, financial workflows are shifting from manual to real-time, from reactive to predictive. Companies that embrace Document AI are not just streamlining operations but unlocking new levels of efficiency, accuracy, and customer satisfaction.
What strategies have been essential for you in driving Docsumo’s growth and maintaining its innovative edge?
Rushabh Sheth: One of our biggest strategic bets has been on Large Language Models (LLMs). We realized early that LLMs would redefine document processing by enabling AI to truly understand context, not just extract text. That’s why we’ve integrated LLMs into our AI stack—so our models don’t just extract numbers from financial statements but also interpret patterns, anomalies, and insights. This shift from “reading” to “reasoning” has been a game-changer in how we approach automation.
Beyond LLMs, our growth has been driven by three key strategies:
- Solving a real pain point, not just building cool AI.
We started by focusing on financial services and automating high-friction workflows like bank statement analysis, rent roll processing, and insurance claims validation. This focus helped us land our first major customers, and from there, we expanded into adjacent industries. - Building a highly flexible and scalable AI platform.
One of the things that set Docsumo apart is that our AI is not template-dependent. Many legacy solutions require customers to define rigid templates, but we built pre-trained models that learn from as little as 10-15 documents. This adaptability has been crucial in winning enterprise clients who need scalable solutions across multiple document types. - A culture of continuous learning and iteration.
We constantly iterate on our models, incorporating customer feedback and real-world edge cases into our training data. Our R&D team stays ahead by experimenting with new AI techniques, including multimodal AI that combines vision and language models.
Finally, having the right team and mentors has been instrumental. Getting into Techstars London in 2020 helped us expand globally, and surrounding ourselves with strong advisors has kept us focused. Growth isn’t just about scaling revenue—it’s about staying innovative, solving real problems, and continuously pushing the boundaries of what AI can do.
For organizations looking to adopt AI-driven solutions, what advice would you offer to help them ensure smooth integration and tangible benefits?
Rushabh Sheth: My biggest piece of advice is to start with a focused use case. Don’t try to boil the ocean on day one. Identify one workflow or process that is particularly painful or time-consuming, and begin by automating that. Starting small not only makes the project more manageable, but it also helps you win early support internally by showing quick results. For instance, instead of attempting an enterprise-wide AI transformation, you might start with automating invoice data entry or, as in NDR’s case, settlement letter processing. Once you see success there, you can scale up to other processes.
Another important strategy is to make friends with your IT team and choose the right tools. Look for AI solutions that play nicely with your existing systems – many modern AI platforms offer APIs or even low-code integrations that make deployment easier. If you pick technology that fits into your current workflow (rather than completely overhauling everything), the integration will be much smoother. Also, consider the usability of the solution: does it require extensive training to use, or is it intuitive? A user-friendly AI tool will get adopted faster by your team.
Don’t underestimate the human aspect of AI adoption. In my experience, the organizations that see the best results are the ones that bring their people along on the journey. That means investing in training and change management. Make sure the team understands why the new AI solution is being introduced and how it will make their lives easier. Sometimes there’s natural fear about AI or resistance to new tech, so communication is key. Show your team the benefits and perhaps even involve some of them in the implementation process so they feel ownership. Maybe have a small group pilot the tool, gather feedback, and then champion its rollout to others.
Finally, align the AI initiative with your business goals and do your homework before you start. Ask: what metric do we want to improve? Is it processing time, error rate, customer satisfaction, cost reduction? Having clear goals will help you measure tangible benefits. And do a needs assessment upfront – ensure you have the necessary data quality, and that the AI solution you’ve chosen actually fits the problem you’re trying to solve. Sometimes organizations get excited about AI as a buzzword but picking the right use case and tool is crucial to see real ROI. In short, start small and smart, integrate seamlessly, take your people with you, and stay focused on the business outcome. If you do that, AI can become a smooth addition to your operations and you’ll quickly notice the gains in efficiency and value.
As a pioneer in Document AI, where do you see the technology heading in the near future, and how is Docsumo preparing to lead in that direction?
Rushabh Sheth: I see Document AI heading toward a future of invisible, ubiquitous integration. What I mean is that Document AI will become so ingrained in business processes that people might not even realize when they’re using it – it will just quietly do its job in the background. We’re moving towards an era of almost “touchless” processing, where documents can be handled end-to-end by AI with minimal human intervention, except for oversight. In the near future, I think we’ll see AI not only extracting data from documents, but truly understanding them. For example, AI will grasp context and intent: reading a contract and knowing the key obligations, or scanning a financial report and providing an analysis, not just raw numbers. Large language models (the kind of AI behind GPT-style technologies) are already showing how machines can interpret nuance and generate summaries. I expect Document AI will leverage these advances to handle even more complex tasks like summarization, cross-document correlation, and anomaly detection.
At Docsumo, we’re actively preparing to lead this future. Our mission from day one has been to let organizations read and analyze documents without manual work, and we’re continuously pushing toward that mission. Concretely, we are investing in R&D around the latest AI techniques. We’re exploring how generative AI can be combined with our platform – imagine an AI that not only extracts data but also answers questions about a batch of documents or drafts a response letter based on the content it processed. We’re also focused on expanding the types of documents and languages our system can handle, because the future of Document AI will demand versatility. Another aspect of leading the way is reliability and accuracy; we’re working on ways to make our models even more resilient to poor-quality scans or unseen document layouts, so that the AI remains accurate without extensive re-training.
Additionally, I believe the future of Document AI involves more real-time processing and decision-making. So we’re enhancing our platform to provide insights instantly as documents come in. For example, not only extracting data from, say, a loan application document, but immediately telling a lender whether certain criteria are met or if there are red flags, all in one go. Essentially, merging document processing with analytics and decision workflows. Docsumo is preparing for that by integrating more deeply into our clients’ decision engines and offering analytics on top of the extracted data. We also anticipate stricter compliance and data privacy norms around AI, so we’re building with those in mind – making sure our solutions are secure, auditable, and trustworthy. As pioneers, our job is to stay a few steps ahead: we’re constantly talking to our customers about what documents or processes they wish AI could do next, and that often guides us on where to innovate. In the near future, I see Document AI moving from a niche tool to a default feature in all content-centric processes, and we intend for Docsumo to remain at the forefront of that evolution, setting the benchmark for what’s possible.
Finally, what message would you like to share with our readers about embracing AI as a transformative tool for growth and empowerment?
Rushabh Sheth: My message would be simple: don’t fear AI, embrace it – but do so thoughtfully. AI truly is a transformative tool, and when used right, it can amplify what humans are capable of. I’ve seen companies where AI takes over the drudgery (like data entry, reconciling documents, etc.), and it frees up the team to focus on creative, strategic, and high-value work. Instead of replacing people, in the best cases it elevates them. Employees move from being data gatherers to decision-makers. Organizations that have embraced AI in this way often find that their teams are actually more empowered and happier, because people get to work on interesting problems rather than repetitive tasks. In terms of growth, AI can be a huge accelerator – it can help you scale operations, serve more customers, and enter new markets without linear increases in cost.
Of course, adopting AI comes with a learning curve and change, and it’s natural to have concerns – whether it’s about job roles or trust in the technology. To that I’d say: educate yourself and your team about what AI can and cannot do. Start small, measure the impact, and you’ll likely see the benefits first-hand. In my experience, once the initial hesitations are overcome, the benefits of AI – the efficiency gains, the innovation it sparks – far outweigh the challenges. We can and should approach AI with optimism and responsibility. As a society and in business, we have the chance to shape this technology into a force for good.AI is here to stay, much like computers or the internet – it’s the next layer of how we interact with information. So my advice is to approach it proactively. Be curious: experiment with AI tools, encourage your teams to learn new skills alongside AI, and create a vision for how AI can advance your mission. Those who embrace AI as a partner will find themselves at an advantage. It’s like having a superpower at your disposal – one that can handle the heavy lifting of data and computation at lightning speed. That leaves us humans with more time and insight to do what we excel at: creativity, empathy, and complex decision-making. So, I encourage readers to see AI not as a threat, but as a transformative ally. If you lean into it, you’ll likely unlock new growth opportunities and empower both your business and your people in ways you might not have thought possible. Embracing AI with the right mindset can truly be transformative.
Rushabh Sheth
Co-founder & CEO at Docsumo
Rushabh Sheth: Co-founder & CEO of Docsumo, Rushabh is passionate about improving people’s lives through AI & automation. Over the last 10 years, he has worked around the globe in data science consulting, e-commerce, classifieds and document analytics.
Explore AITechPark for the latest advancements in AI, IOT, Cybersecurity, AITech News, and insightful updates from industry experts!