The world's leading organizations in their decision-making and risk rating processes largely engage AI-based scoring models based on a vast database.

Plat.AI is specialized in designing AI-based scoring models in the field of finance and insurance.

During the last few years Plat.AI, in cooperation with Nork technological center, has designed a number of scoring models. On July 19 the designed products were presented to the Armenian financial sector representatives at an event entitled “Unleashing potential of income estimation through AI to prevent fraud and mitigate risks” held in the Holiday Inn hotel.

Banks.am was also present at the event and presents its details.

Introducing U.S. experience in Armenia

Plat.AI’s company originally served financial and credit organizations, offering scoring model creation, design and deployment services. Having expanded over the years, today the models offered by Plat.AI engage not only the financial but also the insurance sector.

Currently, more than 300 operating models of the company implement up to 10 million predictions for various credit and insurance organizations per day.

“Just a week after the start of the Covid-19 pandemic we began designing a new index. At that period of time, the lending market was in a limbo, as the lenders did know whether the borrowers would be employed or not in a month or two or three. So we designed the employer state stability index based on vast databases that was promptly engaged in credit companies’ real-time loan approval decision-making. Today we are implementing this and other successful experiences in Armenia,” said Alek Kotolyan, Plat.AI COO, speaking with Banks.am

He noted that in cooperation with Nork technology center, more than 7 products have been designed for Armenia’s banking and social protection sectors. They are universal AI-powered models based on vast databases.

Machine learning development in Armenia

Plat.AI Armenia has been operating since 2020. Alek Kotolyan said they planned to start operating in Armenia earlier, but it was delayed due to the lack of relevant specialists.

 
“Our principle is to give back to the community where you come from. That is why it is very important for us to operate in Armenia. I am pleased to note that the lack of personnel availability today has been filled to a certain extent, and the field is developing rapidly. Just several years ago after a hundred interviews, we finally managed to find the right candidate that met our requirements, whereas today we have Plat.AI with a full team that also works with the international market,” the company’s COO said.

To train machine learning and artificial intelligence specialists and develop the field, the company cooperates with the American University in Armenia, TUMO technological center, Children of Armenia Fund (COAF) and other institutions.

Localization of AI solutions in the Armenian market

Plat.AI models are designed based on the database of the largest credit bureaus in the USA. In cooperation with Nork technology center, the company aims to adapt them to the Armenian market with the involvement of local experience and base. To this end, Plat.AI expects active participation from the financial sector. This will help to increase the preciseness of the existing models, while the improved models, in their turn, will serve for more effective risk management in financial structures.

 
“Nowadays in the Armenian market, it is also essential to fight against fraud which will help us to cure the financial field. For this purpose, we need to work coordinated with the active engagement of financial institutions. As the types of fraud are different all around the world it is important that the models for fraud detection are designed based on the local market’s peculiarities. Hopefully, financial organizations will be more open to share information on the exposed cases of fraud with the Nork technological center. The latter will depersonalize the date and provide us with it. Thus we will be able to train our models, create more precise indexes and become able to detect and prevent fraud for the benefit of the whole market,” said Alek Kotolyan.

The advantage of the models

Lusine Sargsyan, Plat.AI’s CEO, presented the models offered by the company in detail to the participants of the event. She told about the advantages and disadvantages of both universal and domestic models.

"Each company can build its scoring model based on the internal data. On the one hand, this gives an advantage in the sense that the company is familiar with all the variables and parameters of the system. However, though there is a certain flexibility issue with the universal system, it has a significant advantage over custom systems. The latter cannot work accurately in cases where the decision is made, for example, for a person who has no credit history, or who has not previously been associated with the company, because it simply does not have the necessary information. Meanwhile, universal models are built on the basis of huge databases, in this case including data related to the entire population of Armenia, so they have all the necessary information to make a prediction with high accuracy in this case. Considering the advantages and disadvantages of these two tools, it is good if the company uses both models," said Lusine Sargsyan.

During the meeting, the head of the company also presented the study they carried out. They compared the accuracy degrees of the Plat.AI and approaches usually used in financial structures.

During the meeting, Lusine also presented the study they carried out, in which they compared the accuracy of Plat.AI models and approaches usually used in financial structures. According to the study, Plat.AI's model is on average 7-10 percentage points more accurate in all areas considered. In particular, a significant difference was seen (about 40 percentage points more accurate) if looking at the 12-month forecast for those people who actually experienced a decrease in salaries. The latter is particularly important for fraud detection.

Due to the open format of the discussion, the participants of the event had the opportunity to ask questions and get answers. After the official part was over, active networking continued in an informal environment.

Gaiane Yenokian

Photos by Emin Aristakesyan