Automating automation: Machine learning behind the curtain

Vice President, Marketing and Product Management, Parascript

Robotic process automation (RPA) can be the true antidote to manual, rote work, or it can be our worst nightmare if you listen to all the drama or the hype. RPA centers on the use of artificial intelligence (AI) to apply human-like thinking to streamline a typically manually intensive process or activity; and whether we like it or not, it’s here to stay.

Take, for instance, the process of data extraction from documents such as invoices. Application of advanced optical character recognition (OCR) and intelligent document recognition can automate a significant amount of the job of data entry typically performed by clerks or specialized data entry staff.

Machine learning for automating configuration

Interestingly, human effort is still involved with attaining the ability to hand off a process or task to a machine. Whether the process is discovery of configuration rules for the email server, training a speech recognition engine to understand a request or configuring data extraction rules for invoices, human effort is required. The work is complex because the individual doing the work needs to understand the range of input required and how to measure results. This work is not a one-time affair. An investment in time needs to be made each time a new automation task is necessary.

What if the initial configuration could also be automated using machine learning, which is a part of AI? Such a task would almost certainly benefit anyone wanting to implement AI-based automation of any kind. In the age of big data, the input required would be readily available. Consider this opportunity when using invoice capture. The current approach for invoice data extraction typically uses one of two models: 

  • Rigid rules based on specific vendors using zones
  • Flexible rules based on keywords and field patterns—for example, amounts that always have dollars and decimals 

These models are implemented in one of two ways: either tuning beforehand using samples of documents or tuning after the fact as they are encountered in production. In practice, both are typically done. One trick some vendors employ is to use established rules for identifying those invoices that underperform and introduce those examples into a workflow that has a staff member identify the location of fields. This data either goes into the rules-based approach or creates a new rigid template. With enough time or effort, the library of invoice rules will have sufficient coverage. a vendor wanted to offer something out of the box, they take this challenge on themselves by gathering samples, analyzing each, developing rules and so on to minimize the effort for businesses. This process is what many vendors do. One approach requires a lot of staff time to produce good results, and the other requires a lot of developer time. Both techniques could technically be called AI, but are miles behind the current capabilities of machine learning.

Automation for big data and beyond

If these vendors were to focus on automating processes using big data and machine learning to automate the development itself, what would that effort look like? The answer is, it would look a lot like the IBM Watson web application programming interfaces (APIs). It starts with obtaining representative samples and associated truth data and inputting them into a machine learning system. The developer or data scientist observes results and—depending on performance requirements—makes adjustments to the system parameters. The result is a system that approaches the level of tuning a mature production capture system delivered out of the box.

This same organization could even make this algorithm public and build a community that further enhances the performance by submitting its own samples and truth data to continue the machine learning cycle on a more diverse and voluminous data set. Taking this route, a developer-centric effort transitions to one that relies on machine learning, big data and a community. Even though the process may appear more complex, the overall effect is greater effectiveness and efficiency by automating what was once a time-intensive and complex task: development of tuned data extraction algorithms.

Automating automation may seem far-fetched, but this approach is happening today through unprecedented access to machine learning platforms and community-driven data. The result is more expanded access to higher quality data.

Fast track your data and secure your competitive advantage with machine learning. Register for the June 22nd livestream.