The Technology Landscape of Intelligent Automation

There are many challenges for people to solve during their automation journey. Over the years, methods to tackle these challenges have evolved. Repetitive manual processes initially were automated using Macros, a pre-recorded set of rules, that execute tedious repetitive manual jobs. Then, RDA and RPA technologies took automation to the next level, and each of them was more successful in minimizing manual work performed by a human.

The latest frontier of automation, Intelligent Automation (IA), is promising to eliminate the remaining limitations in automation by solving unsettled issues with AI and ML Technologies that aim to automate any cognitive tasks - such as reading and understanding documents, speech and text commands, and generating voice and text context in response. The technology landscape of Intelligent Automation is composed of tools and platforms such as Business Process Mining, Business Process Orchestration, ML and AI, Natural Language Processing (NLP), and Intelligent Document Processing (IDP).

While new technologies evolve, automation of data extraction from documents remains a significant barrier, preventing many companies from the successful implementation of IA solutions. Data extraction from relatively simple documents (such as simple tables or forms) has been implemented successfully industry-wide, with low rates of unrecognized documents. However, more complex documents, such as invoices and contracts, hold all kinds of important data that cannot be extracted fully automatically. Some examples of such cases in different industries include but are not limited to:

Banking, Investment, Asset Management and Insurance:

• Loan Contract Processing
• Financial Statement Processing
• Invoice Data Processing
• Mortgage Document Data Extraction
• Document Content Search
• Confirmations and Pre/Post Matching
• Customer Onboarding, Account Opening
• Loan Applications
• Compliance-Related Processes
• Receipt Processing
• Vendor Onboarding
• Claims Handling
• Mortgage Processing

Manufacturing, Supply Chain

• Sales Order Processing
• Accounts Payable/Receivable
• Parts Requests from Customers
• Remittance Processing
• Order Scheduling and Tracking of Shipments
• Bill of Landing
• Transport Notes

Media

• Contract Management

Healthcare

• Billings and Claims Management
• Insurance Processing

IDP is a solution that uses ML, Computer Vision and NLP technologies to capture, categorize, and extract data from PDFs, images, and office documents, and to map data to structured data sources. Due to the use of “smart” technologies, the quality of data extraction from complex documents could be significantly improved.

IDP is usually integrated with RPA, BPM and internal systems and applications, or provided as a Software as a Service platform. There is a variety of technologies that could be used to implement IDP solutions:

Open Source Solutions

• spaCy
• AllenNLP
• Stanford University NLP
• Tabula
• Excalibur

Pros

• Mostly free of charge
• Could be fully integrated into existing infrastructure without external endpoints

Cons

• Lower recognition quality out of the box, as these models and libraries are not trained on exabytes of data
• Requires qualified specialists to deploy in a local infrastructure, to create API, to tune and to train

Machine Learning as a Service (MLaaS) Solutions:

• Amazon Comprehend
• Google Cloud Platform Natural Language
• IBM Watson Discovery
• Microsoft Azure Cognitive Natural Language Processing

Pros

• Fast and easy jump-start, no need to deploy infrastructure first, API available
• Pre-trained on exabytes of data, which provides much better results out of the box
• Transparent pricing that depends on the volumes of data, gradually decreasing
• Convenient integrated annotation service

Cons

• Constant expenditure compared to opensource solution deployed on-premises
• Working with tables is a challenge

Software as a Service (SaaS) Solutions:

• IBM Watson Compare & Comply
• AWS Textract

Pros

• Works out of the box. Decent quality of tables recognition and some key-value pairs
• Hassle-free integration via API

Cons

• Impossible or hard to tune
• If the key-value pair is not recognized, requires additional efforts to extract it
Download a Detailed RPA Case Study


DataArt helps companies implement IDP solutions by combining open-source, MLaaS and SaaS solutions, as well as our own IP and solution accelerators. We offer our customers IDP consulting services, pilot implementation, integration with RPA, BPM and internal systems.