February 2022 Summaries

Building on its simplicity, Excel table has been the most predominant way of storing any structured data digitally. The seemingly simple spreadsheets are actually tightly linked to the daily data processing by large corporations and organizations. In a few clicks, companies can now distribute tasks to different workers, keep track of budgets from multiple cash flows, and even make accurate predictions from past data. However, extracting data from pre-existing tables, scans, or images in the first place isn't easy. Delivering error-free tabular data extraction seems to be something so close yet so difficult to achieve. The task of extracting tabular data can be divided into two sub-problems: 1) extracting tables from scans/images/PDF documents where the format is not recognizable by machines and 2) understanding/interpreting the words inside table cells so that it could be properly imported into CSV files for spreadsheets. This process involves various use cases such as business cash flow tracking, cross-business record transfer, and accounting firms. To tackle this task, high-level steps involve using deep learning concepts like CNNs and RNNs to classify documents, detect tables, and perform optical character recognition. The process also requires converting PDF files into image formats, finding tables within images, and extracting content via Google Vision API or other OCR services. For non-technical users, Nanonets offers a user-friendly interface for extracting tabular data from invoices, receipts, and other documents without requiring coding knowledge.

Nanonets has secured $10 million in Series A funding, led by Elevation Capital and notable tech leaders, to further its mission of creating a frictionless document communication platform using Deep Learning. Over recent years, Nanonets has enabled hundreds of companies to automate document management and communication processes, enhancing productivity by transforming repetitive tasks into strategic decision-making roles. The company envisions a future where AI will automate complex business processes, reducing human labor in data entry and verification, and has introduced an academy to train workers as AI Process Managers. Nanonets is positioning itself as a pioneer in AI Business Process Management Platforms, aiming to simplify AI deployment for business automation and making it accessible without the need for extensive in-house expertise. The company is optimistic about its growth and is committed to supporting its clients with cutting-edge AI solutions.

The invoice-to-pay process is a formal request for payment related to goods or services, typically starting with the receipt of an invoice from a vendor/seller, followed by checking its validity and approval processes. Automation of this process has become increasingly important due to the volume and complexity of transactions, offering benefits such as rapid processing, better vendor-client relationships, prevention of errors, and improved transparency. The automation involves data capture, coding invoices with correct account and cost center information, approving invoices, matching them to purchase orders, and posting for payments. Software solutions like Nanonets use AI-enabled invoice readers to intelligently capture relevant data with minimal errors, streamlining the process, reducing paper clutter, and enhancing productivity. The choice of an automation tool depends on factors such as budget allocation, scale of business, bottom line, investment potential, ease of use, tech support, and integration with existing digital tools.

The article provides a comprehensive guide on converting information from scanned PDFs to Excel through various techniques, highlighting the challenges and solutions in this process. With the exponential growth of data, PDFs have become a prevalent format for storing text-related data, yet extracting information from them into Excel remains a complex task due to the lack of inherent table structures in PDFs. The guide explores methods such as Optical Character Recognition (OCR) and Deep Learning for automating the extraction process, emphasizing the importance of identifying electronically generated versus scanned PDFs. It reviews tools like Nanonets, EasePDF, and Adobe Acrobat, discussing their advantages and limitations in automating PDF to Excel conversion, and outlines business benefits such as improved efficiency and data integration. The article also addresses common issues like algorithm selection and post-processing challenges while offering insights into building robust deep learning pipelines for this conversion task.