Optical character recognition (OCR) and robotic process automation (RPA) are commonly used terms in today’s business world, however, linking these words to Artificial intelligence (AI) suddenly poses the question of what to do once a potential use case has been identified?
In our last Blog article, we tackled the task of identifying use cases within an organization. In this article we are going to take one step further and look at the specific use case of Intelligent character recognition in the Human Resources (HR) department, addressing the following questions:
- What is ICR as an AI solution and how is it different from using a robot that uses OCR?
- How is ICR applied in practice?
- How can an ICR solution be implemented and rolled out?
How can we use AI in everday operations? – Evolving from OCR to ICR
A specific use case for AI is Intelligent Character Recognition. At this point making a distinction between Optical Character Recognition and ICR is key. Whereas OCR refers to recognizing certain elements from structured documents that have minimum variation in terms of layout and language, ICR adds an intelligent dimension and learns how to extract document elements, despite variations in layout, language and type. An ICR robot can hence classify document types and process the relevant data as trained. This enables respective employees to focus on tasks that are more value adding.
Thereby, the application areas are widespread, and some include:
- Receipt recognition for travel expenses
- Invoice processing within the procurement department
- Application handling within the HR department
- And many more…
In this example, our focus is on the process of handling applications within the HR department using an ICR robot. Therefore, the question arises, how do organizations process applications? For most companies, the answer will be the same; manually! This draws employees’ time away for activities with little value. Many of our clients face the same issue, hence making it important to add intelligence into this process.
To grasp the full advantage of AI in this context it is very useful to make a comparison between two digital workers in the form of an OCR and an ICR robot. On one hand, we have the RPA robot that can download applications from a certain E-Mail account for example, store it in a defined location, forward it or even extract elements such as the applicant’s name by using OCR. However, the pre-condition is that there are no variations between CVs, which is not realistic in an organizational setting. This is where an ICR robot makes the difference.
A learning ICR robot can assess applications of any layout and language, extract relevant data from these and upload this information across applications (for example into an ERP system). With an accurate extraction of data, decision and administrative tasks can be shared between a human and a robot. The human can focus on the truly value adding components, namely contacting the candidates in person. The robot on the other hand can assist with the pre-selection process of candidates, but also take over the administrative tasks completely, by downloading documents, scanning these for relevant information and uploading the data across applications. To achieve this, the machine is trained to learn variations between documents, hence eliminating the need for human intervention over time.
What do you need to get ICR started? – From minimum viable product (MVP) to roll-out
Depending on which software your company decides to use, the training process varies, however, the implementation approach for an ICR prototype can be divided into three stages. Bear in mind that training a prototype requires data to assure accurate results. Specifically, these three stages are MVP classification and field training, prototype creation through validation and scalability of prototypes.
Stage 1: The prototype is trained to classify documents based on certain page elements so that it recognizes similar layouts, whereas field training teaches the machine which data to extract for a certain classifier. The result is an MVP.
Stage 2: The MVP is continuously trained with new data to extract the required fields, improving its accuracy. With a lack of data more interaction is necessary to validate the extracted fields.
Stage 3: Starting with an MVP and validating the prototype allows to scale it for various layouts. After the prototype has been validated sufficiently it can be utilized in every day operations.
Comparing digital employees – OCR and ICR
To see ICR in action, see our prototype:
How to procceed?
Using ICR as an AI solution has significant benefits for organizations. Not only does it reduce processing times but is also scalable to other business processes. Thereby, implementing AI is best undertaken through selecting a use case, locating a focus topic such as ICR and initiating its implementation with an MVP. The ultimate target should always be to scale it across functions and see AI integration as an incremental process.