Artificial Intelligence and Invoice Processing: What Do They Have in Common?, Part 1
Artificial intelligence – without a doubt one of today’s hottest buzzwords. This little keyword sometimes carries with it massive expectations. Some market forecasters are already predicting a revolution or sketching out overwhelming threats a la Skynet. If you’re thinking of the 1984 hit “Terminator,” then you’re right on target.
But what exactly does the phrase mean? “The term artificial intelligence, or AI for short, stands for computer systems that imitate human intelligence. […]” This means it’s an umbrella term for a group of technologies that expand the classic boundaries of IT systems. That, in any case, is how the “Spiegel” defined the term in its 4.1.2017 edition.
Details in Focus?
In my experience, and considering developments in digital invoice processing over the past few years, there seems to be a high level of maturity among the solutions available on the market today. Therefore, the development of products for invoice processing seems to be driven more today by detail optimization than by sensational leaps in development.
Today, users consider web-based access and specialized apps that integrate mobile devices seamlessly into the review and approval process standard features. These kinds of services are broadly available today. No group company nowadays is without its own solution. Even among mid-sized companies, the majority of companies have already or are currently in the process of introducing such invoicing solutions. Ideally, these solutions would be dynamically linked to the company’s IT landscape and work seamlessly with back end systems.
However, AI-based methods have already snuck into our everyday lives, often without notice, becoming a natural part of our environment in both our private and professional lives. This also applies, but is certainly not limited to, solutions for electronic invoice processing. The trend might become clearest if we look at a few examples for how these methods have already become established.
Let’s start with the electronic capturing of paper documents and provision of information for further processing. Even today’s widely available solution components offer amazingly good recognition rates and simple operation. These have continuously pushed the boundaries of technology over time. Just a few years ago, poor templates were a major problem, for example because they used thin and transparent paper or because folds, signatures, or stamps made automated reading of such documents more difficult.
Modern systems today capture a large percentage of these documents accurately. They do so by combining specialized algorithms: picture recognition algorithms use AI mechanisms to eliminate errors and optimize image quality. Error-tolerant systems (fuzzy logic) identify information even with incomplete input values. And multi-dimensional vector systems learn through training with example data, for example with unknown address formats.
Learning from People Means … Optimizing Algorithms
More and more, self-learning systems will increase their performance capabilities and recognition rates solely by observing human users while they work. This means the algorithms will optimize themselves through continuously comparing input information with the information recorded by human specialists for further processing.
In the second part of my post, I’ll take a closer look at the issue of intelligence, for instance in the context of incoming mail solutions or assistance systems.