Collaborations abound. Don’t want to listen to Miley Cyrus while driving in a cab? Then simply order an Uber and enjoy your own Spotify playlist in the back. Or guaranteed to stand out on your next stroll through town? A piece of cake with the wacky Yeezys from Kanye West and Nike – just two of the numerous examples of collaborations. It’s fair to ask whether there’s a need for more. The answer is: maybe one last more, especially since this one is also a bit different. Namely, the collaboration between man and machine, the so-called Human in the Loop (HITL).

Purely manual work makes no one happy

Automation is intended to serve people in their everyday work and relieve them of manual tasks. These manual tasks are very often – rightly – criticized for being time-consuming, cost-intensive and error-prone. This can be seen very specifically in the case of document processing, for example: if carried out manually, these are enormously time-consuming and therefore cost-intensive, which at the same time will make the executive suite nervous and give employees little added value due to these tedious and lengthy tasks. In addition, if the work is boring, repetitive, and thus not very demanding, then there is often a higher susceptibility to errors, which then also makes no one happy. But what and where exactly to automate?

Parashift: Autonomous, accurate, painless and fully validated document extraction

Automation in invoice processing

As with so much, context is of course important. The entire workflow in a company must be considered, in this example document processing and specifically invoice processing with the usual data such as:

  • Invoice number
  • Vendor
  • Payment date
  • Invoice amount
  • and so on

Accordingly, automation is needed when it comes to the document inflow so that invoices do not come into contact with employees in the first place.

Human in the Loop – machine processing and human post-processing control

Nouveau call-to-action

The machine a.k.a. intelligent, AI-based OCR

This is where intelligent, AI-based OCR (Optical Character Recognition) comes in. The intelligent, AI-based OCR automatically captures the incoming invoices, improves the quality (including page separation if required), classifies the invoices and finally extracts all relevant data automatically. What remains is structured data that can be further processed automatically.

Data for the machine-human in the loop for follow-up control

Every AI system needs humans in the loop. At least to some degree. Of course, intelligent OCR is only as intelligent as the fed data allows, from which the different models learn and based on which they create their classifications. In order to validate the performance of the classifications at all, the AI needs a dataset already collected in the cloud, based on which it

a) has a ground truth available to know which values are correct or are actually on a document

b) is able to successively select parameters in such a way that errors can be minimized

c) continuously has the ability to recreate parameters to be optimized for a specific metric.

If a classification cannot be made or improved, for example due to missing data, the AI needs additional context. It gets that from the human in the loop. Incorrect data or errors that lead to label changes can be quickly detected and corrected thanks to the human in the loop. With the human in the loop, the classification can be validated and, if necessary, labeled differently, so that in the future these classifications can be improved and, in turn, made automatically and correctly.

The Human in the Loop confirms, denies or flags the output of the machine if required

Zero post-processing required on the part of the customer

What the intelligent, AI-based OCR does not do automatically is therefore taken care of by the human in the loop, so that the customer can sit back and enjoy the coffee. Or can take care of much more important things like customers and trading partners.

Massive advantages through Human in the Loop

So the collaboration of machine and human in the loop for follow-up control offers numerous advantages:

  • Fast processing (when interactions are necessary, the human in the loop is there to help).
  • High cost reduction (previously manual work on the part of employees is eliminated)
  • No more document interaction necessary
  • Increase in straight-through processing due to correct data (reconciliations, postings, orders, etc.)

Even though ‘human in the loop’ sounds a lot like the machine subservient at first glance, this is not (yet) the case. The successful process requires the cooperation of both.