Many organisations trekking through their Robotics journey find that the road to achieving full automation is blocked due to the inability to digitally read hand written, cursive text located in manually filled in forms. With Intelligent Character Recognition (ICR) technologies now being injected into the market, this is now possible. What is ICR technology? Intelligent Character Recognition: An advancement of the more commonly known technology Optical Character Recognition (OCR), where we are no longer limited to ‘computer-reading’ printed text found in images, but are now able to also read handwritten, cursive text that on occasion even people cannot decipher. How can a technology read hand written text that even humans can’t read? ICR is able to do so by two key mechanisms. Training Sets: When handwritten text is being processed, ICR draws on data, or in machine learning speak, training sets, where it compares each character to a variety of different handwriting styles and formats, matches it and determines a transcription it is most confident in Learning Algorithms: Machine learning algorithms further enhances its ability to accurately read human handwriting as it is used and further taught correct transcriptions. An example of what ICR technologies can do: Can you read the horribly written sentences – ‘horrible sentence” 1, 2 and 3? ICR can! See below for the output of its transcription: ICR in the Office Place It is not new news to hear that many organisations are eager to kick start their robotics journey. Though, at times many find their ability to automate processes heavily dependent on having required input data digitised and in a computer readable format. This often applies to processes that revolve around paper/scanned forms as inputs (e.g. home loan applications, registration forms, cheques etc.) For example, a common type of form that could be received are those that have been physically printed, hand populated, and then posted or scanned and e-mailed. In a lot of cases, companies dedicate whole teams to entering data from these documents into systems. With the emergence of smarter and more accurate character recognition capability, many organisations now have the means to automatically transcribe data into a structured, computer readable format. By coupling ICR with other automation technologies, such as robotic process automation that can mimic human interactions with desktop interfaces, allows the ability to automate processes end to end. Organisations can remove, or at least reduce, the effort spent on mandrolic data entry activities. ICR in the industry Companies have been facing the perils of having dependencies on hand written forms to execute their processes and together embarked on a journey to explore ICR capabilities and its application in all its realism. At times, some handwriting styles are even illegible to people. Top 5 Tips for Leveraging ICR: 1. Understand variability in the process In a perfect world, all processes and inputs received from our customers would be streamlined and adhere to standards. Though, more often, this isn’t the case, and we can only get as close as we can to having processes with little variance. ICR technology can handle some degree of variance with its ability to identify the type of form it needs to process (E.g. is it a home loan or personal loan application?) and handle structured (e.g. forms with fields to populate) or unstructured forms (e.g. free text written documents). However, ICR can only read what it has been taught to read, and any other variant forms that exist would need to be manually processed. Ideally, processes with a finite number of form types would be selected first when automating using ICR technology. 2. Clarify the boundaries of ICR capability Imagine you were filling out a form. Do you always write within the bounding box of a field? Would you write a few side notes if you’ve run out of room in the field area you were given to write your answer in? If your answer is no, oh boy, would ICR configurators like you! If your answer is yes, like what is imagined to be most people, you have introduced field level variance in the process you are filling in the form for. There are so many ways a form can be filled in. ICR vendors have identified that this level of variance is inevitable and that there has to be means to identify such variances and prevent incorrect transcriptions. Some ICR products are more mature in this identification than others. Therefore, it is imperative to fully understand what the selected ICR tool can and can’t transcribe and for those field variances that it can’t transcribe, implement necessary measures to ensure mis-transcribed information does not impact operations. 3. Determine the right confidence threshold Many ICR tools assign confidence percentages to each transcription it has made from a form. This comes useful as it is essentially the computer telling you how confident it is in the accuracy of its transcription. With this confidence percentage, you can assign a certain acceptance threshold, in which any fields that have been transcribed with a confidence level below this threshold, would fall into a manual inspection bucket. This threshold could be at 95% or even 65%. With any data that falls below this confidence threshold, thinking would need to be done around where these scenarios would need to be handled to and whether manual intervention is required. 4. Enhance with enriched training sets As mentioned before, ICR technologies achieve higher accuracy rates by drawing on data sets that it has been trained on. The best kind of data sets to use when training the ICR technology are: Real forms that have been received from customers Geography specific data sets Specific use case data sets (e.g. medication names that aren’t often found in a standard dictionary) Also, the more forms you train the tool on that showcases the varying ways forms can be filled in, the better it will know how to handle different scenarios. In an environment where data is sensitive, it is imperative to begin working through security requirements as early as possible so ICR developers would have access to data. 5. Remember ICR technologies are evolving fast Just within the few weeks of exploring ICR technologies, improvements in the way certain text was transcribed was achieved. On day one of processing forms for transcription, accuracy rates may not be 100%. Though, as the tool gets further used, and when machine learning kicks in, accuracy rates will increase. All-in-all, expect a period when ICR is first established and heavy manual inspection is required. As the technology becomes more accurate and confident, the need for manual inspection will reduce. In addition to this, ICR vendors are evolving their products fast, further adding to text recognition capability and expanding use cases it can handle. Conclusion Digitisation technologies, such as smart forms can eradicate the existence of handwritten forms which in this day and age, is the ideal data submission process. Though, as long as ink pens exist, the familiar ineligible hand written form would still make an appearance. Overall, an integrated automation where multiple robotics and cognitive technologies are used across various points along a value chain brings us closer to really disrupting the way we work and fueling the transformation of our workforce. It becomes important to keep in mind how this will change your organisation and the implications on ‘Work, Worker and Workplace’. Automation is happening, and as capabilities like ICR advances, there are really not too many limits left on automation opportunity.