How to speed up work in the operational departments
Data input and processing activities in an organization can be extremely time-consuming, tedious, monotonous, and costly, mainly if they occur in a manual manner that carries a high risk of errors. With the right OCR systems, not only can such processes be accelerated, but also automated, reducing workforce usage by up to 30%! In the following article, we talk about robotized data transformation and machine learning, as well as the benefits of both solutions.
What are OCR systems, or optical character recognition systems
The main task of OCR (Optical Character Recognition) systems is to convert text that we are not able to read in digital form (e.g., on photos or graphics) into an entirely digitized structure. This is achieved through a set of techniques or specific software, and such ‘processing’ is usually carried out with the aid of scanned documents (forms, invoices, policies, orders, etc.).
The comparison between the classic and intelligent OCR
Classic OCR systems are designed to recognize printed content and save it in digital form. They support structured documents, i.e., in a uniform format. Have you ever wondered why there are “windows” in official letters and why specific fields (e.g., with first and last name) are always in the same place in a given pattern? This layout dramatically facilitates the digitization process, which then does not require advanced solutions. The result here will be, e.g., a word file, which we can edit freely.
What can intelligent OCR do?
Intelligent OCR systems, on the other hand, have much more extensive capabilities. They allow, among other things:
- to identify document types based on elements characteristic for them (for an invoice it will be, e.g., a number, as each has one),
- to recognize written content (especially in capital letters),
- to create a workflow of documents,
- to handle semi-structured documents (i.e., having the same set of data, but located in different places, as is the case with invoices or receipts) – in their case, the OCR system’s implementation will bring the most significant benefits,
- to handle unstructured documents, i.e., consolidated texts, such as, e.g., commission contracts (to find information in them you need words or key characters, as well as creating algorithms and appropriate logic in the OCR system),
- to handle sets of documents (e.g., applications with a set of attachments),
- self-learning, i.e., continuous improvement and assimilation of new elements (e.g., invoices from a new supplier) by machines,
- to integrate with different data sources,
- to read complex parts (e.g., tables with multiple columns and rows).
iOCR – and much ado about nothing? A few words about the benefits
iOCR is one system to handle many types of documents. Its primary purpose is to turn them into structured data and minimize the time the operator has to spend on the file. It is unlikely to happen that human involvement will not be needed at all. Always at the stage of iOCR implementation, a person will be required to be responsible for the quality of the data (if the scans have poor resolution, the system may not recognize certain elements). However, Mindbox’s implementation teams always strive to create an almost fully automated process in which human input is negligible. It makes it possible to increase its quality, save time and resources, and scalability and acceleration. Not to mention the simultaneous release of energy of people who can engage in more exciting and creative tasks.
We use software (the so-called robot), which contains an algorithm of conduct, i.e., the same steps that people would do on the applications they work on (more about RPA, i.e., robotization, and its benefits you can read here). This is possible without profound interference with systems that are already running in the company (the “robot” performs its tasks in the background, but you can see them on the computer that is assigned to it).
What else can we get?
Moreover, it is possible to integrate the data “produced” by the OCR system with the IT systems we use (e.g., CRM or ERP) as well as external systems. We can do this using API, electronic data exchange, or databases (products, etc.). If advanced processes are involved, which use many systems in the company, and we use a robot with implemented algorithms, then after receiving data from the OCR system, it will know exactly what to do with this data. For example, the robot can download data from an XML file and enter into CRM or web applications.
An even more sophisticated solution is to use machine learning models in this whole jigsaw. They allow us to understand, contextualize, and interpret the data provided by the OCR system. Then this data can be transferred to a robot that will process it properly.
In the processes that we automate in Mindbox, we try not to focus on just one of the available solutions, but rather to create a combination of OCR systems and artificial intelligence elements using the appropriate software. The connection possibilities here are endless, with the implementation process developing with the organization and its needs. Individual algorithms and logic can be expanded with new elements, while operators who work with a given application daily can add new rules or providers themselves. We carry out iOCR projects from start to finish and accompany our customers through all the steps, with over 60 automated processes in more than 30 countries. More information you will find in the catalog of business automation projects.
How does it work in practice? A case study
To better illustrate the scheme, let’s look at the automated recording of purchase invoices. It seems more or less like this:
- Input data passes through the OCR system, which then processes, structures, and verifies it (e.g., using ABBYY® FlexiCapture® software, which supports all major languages). It extracts the elements we need (e.g., product names on the invoice).
- The next step is to interpret the data through machine learning and an appropriate program (e.g., APPLICA.AI). It allows the system to know how the data should be entered where to go and what decrees to use. There is no such thing as the unification of names on documents – each supplier can use its naming. However, we managed to create a solution which, based on an extensive database, passed through the machine learning model, made it possible to create a kind of classifier. Thanks to it, we can create an assignment to a given “class” based on different contents. For instance, if the name “Channel lift” appears in the invoice line, our machine learning model will know that it should be recorded, for example, on the account 1281, to the branch 901, the department 850 and the project no. 2. If the name “Channel lift service” appears on the invoice from another provider, which sounds a little different, our algorithm will know from the available database that it should be recorded in the same way. Accommodation service, accommodation, and hotel are the other examples. If these words are found in the learners’ data, in the database that we first fed our machine learning model, the algorithm will be able to interpret these data from the invoice line correctly.
- The last element of the process is the robot, which integrates everything. Apart from the machine learning algorithm, we can apply any number of business rules, which will help determine how to record the documents (the final step). At this stage, the robot processes one document after another completes the necessary data and selects a person to whom it should send the final file for acceptance or possible corrections. So the user logs in to the application, where his task is only to add a few values and click on a few windows. The full process – from the receipt of the invoice to its final acceptance can only take 2-5 hours, while in the absence of such a system, it takes an average of 5 days!
For those who want to know more
Mindbox has been on the market for 13 years, employing 10 certified RPA developers and successfully implementing intelligent OCR technology, conducting projects using Artificial Intelligence models, and providing support in the managed services model. Learn from their experience in the Intelligent Process Automation Academy training. You will learn i.a. how to plan and implement the process automation into your organization.
If you have any questions about automation or optical character recognition systems or are thinking about introducing such solutions in your company, please contact Magdalena Kwaśny firstname.lastname@example.org. We also recommend to contact us via phone on +48 572773537. We will be happy to talk about your needs and provide you with professional advice.