How Verigram came to text recognition and OCR technology
The COVID-19 pandemic has forced us to adapt to the new realities of life and reduce direct contact to a minimum. To keep customers safe and provide remote services, more companies are looking to move their business online as soon as possible. Now more than ever, it's important that customers receive services remotely, from the comfort of their homes. At the same time, in order to place an order and receive services online, the customer must register online, which primarily requires new methods of remote identity verification. Using new ways of online identity verification requires a minimum of effort from customers - it is enough to have a smartphone for online transactions.
Biometric verification became possible with the development of artificial intelligence and machine learning (AI). Thanks to new technologies, machines have learned to see, understand and classify images and videos just like humans. Face recognition, liveness detection, and text recognition (OCR) are some of the areas of AI. The mass use of these services has become possible due to the development of computing power. For the development of verification services, a large amount of data is collected, which is used to train artificial neural networks.
An example of one of our clients change their business with the help of digitalization
Initially the company served its customers offline, operating only in large communities during fixed hours. To receive services, the company's customers had to spend time: attend the branch and wait in line. Situations where a client forgot the necessary documents and failed to return on time before the end of the working day happened. Because of customer claims and competition, the company was forced to increase the staff, rent more offices to open additional branches and extend the working hours of employees. The company spent a lot of resources to improve customer service.
Several years ago, the company's management made a strategic decision to invest in the digitalization of processes. Gradually all offline processes were moved to the Internet (mobile application and website). The main and challenging task was the need to move the customer registration process online while complying with the legally required identity verification steps.
To ensure a smooth transition, Verigram and the client decided to digitize the enterprise step by step, starting with improving the efficiency of the offline branches. A document recognition service (automatic scanning of text data from an ID document) was created and used to increase managers' productivity.
If previously manager spent 5-7 minutes for manual entry of personal data into the system (transferring full name and other important information from an ID, a copy of both sides of the document, checking the correctness of the entered data), with the help of the document recognition service the process was reduced to 5 seconds. The slight reduction in time had an effect on the scale: reducing the process by 3 minutes allowed managers to increase throughput by 15-20%. As a result, customers spend less time in line, so the company needs fewer employees. All this makes it possible to operate with fewer branches.
The second stage of digitalization was full automation of the customer registration process without a manager. To do this, Verigram implemented a mobile version of the document recognition service. So the client didn't have to manually enter his personal data but simply pointed the camera at the document. We automated the process of comparing the photo from the ID with the image of the client in real-time. To do this, we used a facial recognition service which, after cutting out a photo from the document, had to compare it to a photo of the client (the application took selfies automatically). The facial recognition software compares and determines the similarity of the two photos better than a human. During the implementation of this stage, we discovered a vulnerability in the system. It turned out that the service can be tricked by pretending to take a photo or video of another person or by wearing a mask. To eliminate this problem, we developed a program for determining whether a person is a real one or it’s a mask/photo/video in front of the smartphone camera. The implementation of these steps made it possible to move the customer registration manager's actions online while increasing the quality of verification.
After moving the processes online, the client has opened up new horizons. Now there is no need to have dozens of branches in each city. By reducing the number of branches as much as possible, the client was able to change the format of work, turning their offices into showrooms and next-level sales points. Employee productivity increased multiple times, and the company's services became available 24/7 from anywhere in the world. The quality of service increased, as customers stopped wasting time in the offices.
It's important to note that by analyzing the incoming information and studying the cases of fraud attacks Verigram is more likely to detect fraud than the manual method of passport data verification. The program is not able to get tired and lose vigilance.
Based on three years of experience, we've come up with the right automation process for online registration and onboarding.
We can highlight three stages of customer identification:
A document recognition system scans a passport or ID card, reading all the necessary textual information (document number, full name, date of birth, etc.). The photo from the identification document is added to the database as a reference for further checks.
The client undergoes a liveness detection check for determining whether the biometric sample belongs to a real person. This allows you to determine whether the person in front of the camera is real (and not a picture).
In the last step, the facial recognition system compares photos from the ID card with the images taken during the liveness verification process. If there is a sufficient level of similarity, the identification process is considered successful and the system provides services to the client.
The automatic method of remote biometric verification of the client with the help of AI is more reliable and trustworthy than the manual method, which is carried out by the employees. The human factor in the form of mistakes of employees or collusion cannot be excluded when the questionnaire is submitted manually. The manual method forces clients to spend time waiting and answering the manager's mandatory questions. Verigram’s services allow customers to go through the authentication process comfortably, forming a positive user experience.
It is known that 84% of Internet users believe that financial services that use real-time identity verification deserve a high level of trust. At the same time, 71% of users are inclined to share additional data during identity authentication. To ensure customer security, we are constantly working to improve service quality by analyzing fraud attacks. Based on the data obtained, we improve security methods using advanced approaches in computer vision and machine learning.
Thus, thanks to the high accuracy of biometric verification, customers can be sure that fraudsters and ill-wishers will not gain access to customer data and money. This will increase the geography of business services, as users will be able to obtain the necessary service online from anywhere in the world at a time that is convenient for them.