Siyovush Muminov, AI Product Owner, Alif Bank
Alif, a fintech and banking group originally from Tajikistan with its holding company in the United Kingdom, has been successfully expanding its international market presence since being founded in 2014. Currently, Alif serves over 1.2 million clients in Tajikistan and Uzbekistan, offering a diverse range of financial services, including Point-of-Sale (POS) financing, a marketplace, money transfers, and mobile and card payments. Alif is now regarded as a market leader in many products in Tajikistan, while in Uzbekistan, it is distinguished by rapid growth and leading positions. According to the latest data, the company’s run rate revenue is $64 million.
Alif is also expanding into the Pakistani and UAE markets. In Pakistan, the company is actively working on launching BNPL in partnership with local brands. This extends Alif’s geographical presence and gives Pakistani consumers access to flexible financial services. In the UAE, under the brand Alif Payments, the company is promoting the SNPL (Send Now Pay Later) product, aimed at meeting the needs of migrants in the money transfer sector with the installment option. This solution is designed to ease the financial difficulties of migrants, allowing them to send money home with a delayed payment option.
The success of Alif can be attributed to its knowledge and use of the latest technology to transform its financial and banking services. Through an understanding of local needs, Alif has created tailored and sophisticated fintech solutions which are constantly reviewed. With AI becoming increasingly effective, Alif has applied that same line of thinking to explore and begin implementing this technology into its services.
When did the implementation of AI products begin at Alif?
Discussions started around 2019. At that time, Alif Group was not yet a bank, but rather a regulated lending and deposit-taking institution in Tajikistan. A more serious interest in this topic arose after the opening of Alif in Uzbekistan, which aimed to offer POS financing services in a new market. Notably, the Uzbekistan market is several times larger than Tajikistan’s. For comparison, Uzbekistan’s population is about 35 million, roughly 3.5 times that of Tajikistan.
The team strived to maximise the experience gained in Tajikistan to simplify the installment process in Uzbekistan. At that time, the entire process from application submission to approval of consumer credit in Uzbek banks could take several days, with one of the key points being the analysis of the borrower’s creditworthiness. Our colleagues set themselves the task of automating this process using machine learning (ML), starting in Uzbekistan.
By 2021, Alif’s market share in the Uzbekistan POS-financing market was approximately 30%, and the load on the credit scoring department increased significantly. This led to the beginning of the development of its own ML scoring. Fortunately, from the beginning, there was a well-established information storage system, and we had enough ready data for analysis.
The development of a project called “Gulchatay” began. Gulchatay is a traditional female name in Central Asia, made famous by a movie in the 1970s. At that time, colleagues did not even suspect that the project’s name would become very trendy these days. Subsequently, we decided to use a uniform naming style for all AI projects in Uzbekistan and renamed the project to GulChatAI.
How has Alif approached the integration of ML into its scoring system, and what impact has it had?
The proportion of applications processed using ML scoring in Uzbekistan increased gradually. Currently, the level of automation accounts for over 70% of all financing applications. Furthermore, the rate of overdue payments for financings issued by ML scoring is 2.4 times lower than those issued by the credit scoring department. As a result of implementing ML scoring, we were able to improve the efficiency of the installment process, reduce risks, and increase the share of automated operations, which undoubtedly contributed to the growth and development of Alif in Uzbekistan.
In Tajikistan, ML scoring was launched in November 2023. Considering our implementation experience in Uzbekistan, we expect automation in Tajikistan to progress rapidly. The most challenging part of the journey has already been completed.
In Pakistan, for the development of ML scoring, we used the services of a local consulting company specialising in AI. This was due to the lack of accumulated credit behavior data for developing our solution. However, our Uzbekistan AI team took an active part in the validation process of the proposed solution.
How else has Alif been applying AI, and what have the results been?
In Uzbekistan, Alif launched another important project at the end of 2022 — an AI chatbot to facilitate the work of the customer support service.
By mid-2022, the number of users of our products in Uzbekistan and Tajikistan exceeded 1 million people, with about 500,000 active clients, creating significant pressure on our support service. Considering the rapid development and ambitious plans, this load could increase several times in a short period. Understanding this, the AI chatbot project was the second priority in our AI implementation strategy.
Currently, the AI chatbot in Uzbekistan autonomously handles almost 60% of all daily inquiries to the support team, ensuring the accuracy of responses above 80%. As a result, the load on the support team has significantly decreased. Consequently, customer satisfaction remains high, and the efficiency of the support service improves.
The first version of the AI chatbot in Tajikistan was launched in September 2023. Now, the AI chatbot operates in all support service channels, responding to frequently asked questions and covering approximately 40% of all daily inquiries.
In Uzbekistan, in addition to the projects above, we are also actively implementing a project for optical character recognition. This significantly speeds up entering ID data for several thousand clients daily, saving valuable time and resources. In November 2023, we launched an ML recommendation system for our alifshop.uz marketplace, which will enhance the shopping experience for users.
Which types of AI technologies does Alif use at the moment and which AI technologies is it planning to use in 2024?
AI technologies for creating credit scoring are not new. Leading financial companies worldwide have long been using AI in this process. Credit scoring is a binary classification task to determine the probability of a borrower’s default. Our credit scoring model is based on a composition of multiple base models, using gradient boosting on decision trees. This significantly improves the accuracy and reliability of predictions. One of the key indicators of our model’s effectiveness is the Gini index, which reaches 0.44. This is considered a high value, given that acceptable credit scoring models usually have a Gini index starting at around 0.3.
For creating our chatbot, we chose an approach based on identifying customer intents and providing corresponding answers from a pre-prepared database. This method is one of the most popular and effective for developing goal-oriented chatbots. Even with the development and implementation of large generative models, this approach remains widespread due to its ability to provide rapid automation with limited resources and standardised customer requests. To implement this approach, we chose an open-source technology, which we adapted to meet Alif’s specific needs.
However, the chosen method for building the chatbot has an automation limit. Therefore, in 2024, we plan to expand the functionality of our chatbot by adding a generation module. As part of this update, such experiments as testing generative responses instead of standard predefined ones and developing our own LLM (large language model), specifically adapted to work with text within our specific domains are planned. This will allow the chatbot to respond more flexibly and naturally to user requests, offering a more personalised interaction experience.
The human role in selecting and integrating the right AI tool is as important as the technology itself. What has been your experience implementing AI at Alif?
In January 2023, I took full responsibility for the implementation of AI-related projects at the Alif Group level. Until then, my professional activity had been closely associated with finance. At Alif, we are not restricted in our growth — at any level, one can rotate roles, and I transitioned from the Group CFO to the AI product owner.
The transition to the new field was smooth thanks to gradual immersion in AI topics throughout 2022. During this period, I began leading the AI team in Uzbekistan, effectively holding two positions and gradually transitioning from finance to technology. My previous experience in finance at Alif proved to be useful, as I knew the internal processes of the companies and understood how best to integrate AI into the existing infrastructure.
By mid-2022, an AI implementation strategy was developed, which included creating AI teams in Tajikistan and expanding the AI team in Uzbekistan. Uzbekistan played the role of a pilot region where we planned to first implement and test our products, and then transfer successful practices to Tajikistan. This approach was due to the fact that in Uzbekistan, there was already a small, experienced team, while in Tajikistan, it was necessary to build a team from scratch.
The future of AI will be defined by companies like Alif actively exploring how the technology can be used to improve operational efficiencies, and ultimately, deliver a better service. There are exciting opportunities on the horizon, and Alif will continue to be a driver of finance and technology innovation in Central Asia and beyond.
