The automation of banking operations has moved through several well-documented stages: algorithmic credit scoring, chatbot-based customer service, AI-assisted fraud detection. The next stage autonomous voice agents conducting real-time, natural conversations with customers about financial products and obligations is no longer a projection. It is live, at scale, in Uzbekistan.
TBC Bank Uzbekistan, the leading digital banking ecosystem in Central Asia and a subsidiary of the London-listed TBC Bank Group, has deployed AI voice agents capable of handling both credit card sales and debt collection calls without human involvement. The deployment positions the institution at the frontier of conversational AI in emerging market banking.
The Technical Stack Behind the Deployment
The technical infrastructure underlying this capability has been built over several years. TBC Bank Uzbekistan developed proprietary speech recognition and synthesis models calibrated specifically for the Uzbek language, a linguistic environment that generic international AI platforms have not prioritised. A finance-specific large language model, built on Meta’s Llama architecture by specialists who previously developed Yandex’s Alice assistant, enables the agents to engage in contextually coherent conversations about loan conditions, card features, repayment schedules, and account status. All systems run on the bank’s own infrastructure, ensuring full data protection and compliance with Uzbekistan’s regulatory framework.
The operational history of AI deployment at TBC Bank Uzbekistan provides a credible baseline for assessing what the expansion into card sales and collections will deliver. Voice agents were first activated for loan repayment reminder calls, a high-volume, time-sensitive workflow where the cost advantage of automation is most immediately visible.
By early 2025, these agents were autonomously managing over 90% of all early-stage delinquency interactions, maintaining quality benchmarks that the majority of customers assess as equivalent to speaking with a human representative. The progression from reminder calls to active sales and full collections management represents a natural extension of a proven capability into more commercially complex territory.
Autonomous Agents Enter Credit Card Sales
For card sales specifically, the deployment creates a new channel that combines the reach of outbound calling with the personalisation of AI-driven profiling. The system can initiate conversations with customers who have been identified as strong candidates for card products based on their transaction behaviour, repayment history, and digital engagement patterns, present relevant product features in natural language, respond to questions and objections in real time, and progress qualified leads through the application process without handing off to a human agent.
Konstantin Kruglov, Head of AI at TBC Bank Uzbekistan, has described the efficiency differential clearly: AI-powered calls and sales chatbots are ten times more efficient than human operators, a ratio that fundamentally changes the economics of card product acquisition at scale.
Consumer Demand for Modern Card Products
The consumer context in which card products are being promoted matters for understanding the scale of the opportunity. Uzbekistan has over 45 million cards in circulation, but the product range attached to those cards has historically been limited. The population is actively seeking more sophisticated card products a pattern visible in the sustained demand for information around terms such as “банковская карта” and “online karta buyurtma berish”, which reflects a consumer base that expects to research, compare, and apply for bank cards entirely through digital channels.
For a bank whose AI infrastructure can reach this audience at scale with personalized, immediately responsive interactions, the alignment between technology capability and consumer expectation is direct and commercially significant.
Transforming the Economics of Debt Collection
For debt collection, the economic logic is equally compelling. Early-stage intervention in delinquency is consistently the most cost-effective point at which to resolve payment issues both for the lender and, often, for the borrower. The traditional constraint has been the cost of making contact with a sufficient proportion of the at-risk portfolio at sufficient frequency.
AI voice agents eliminate that constraint entirely: the bank can now reach every at-risk borrower at the optimal intervention point, personalise the conversation based on individual account details, and escalate appropriately without incurring the cost of a human call centre interaction for each contact. The impact on credit loss provisions, collection rates, and ultimately the profitability of the lending book is material.
The broader product architecture within which these AI capabilities operate amplifies their strategic value. TBC Bank Uzbekistan’s ecosystem includes Payme, a digital payments application with deep penetration among individuals and small businesses, and Payme Nasiya, a Sharia-compliant installment credit platform.
The data generated across these products feeds the AI models that power voice agents, improving both the accuracy of credit assessments and the relevance of outbound sales conversations. Each interaction whether a payment, a loan repayment, or an AI-driven card sales call becomes an input into the intelligence layer that improves every subsequent interaction.
Competitive Implications Across the Region
The competitive implications of this AI programme extend across the regional banking landscape. Uzbekistan is a market that is becoming progressively more contested: Hungary’s OTP Bank and South Korea’s Korea Development Bank are already present, and Société Générale, Kaspi, and Citibank have all been reported as evaluating entry. The institutions that arrive to find TBC Bank Uzbekistan already operating proprietary AI voice agents across sales and collections generating data, improving models, and compounding its cost advantage with every call will face a competitor whose operational infrastructure is years ahead of what can be assembled quickly.
TBC Bank Uzbekistan’s financial performance provides the commercial validation of the overall model. The ecosystem has grown its registered user base to over 17 million, recorded net profit of $27 million in the first nine months of 2024, and is targeting $75 million for the full year of 2025.
The loan book is doubling year-on-year, deposits are growing at comparable rates, and the platform is projecting a net profit of $250 million and a loan book exceeding $3 billion by 2030. These figures are being pursued while simultaneously funding one of the most ambitious proprietary AI programmes in the Central Asian technology sector, a combination that characterises an institution operating with both strategic clarity and operational confidence.
What TBC Bank Uzbekistan is demonstrating in Uzbekistan is that the frontline of banking the calls, the conversations, the sales interactions, the collections contacts is now a domain where AI can operate not as a supplement to human capability but as its replacement in the majority of use cases.
The trust question that attaches to AI-driven financial conversations is real, but it is answered by performance: when customers consistently rate AI interactions as indistinguishable from human ones, and when collection rates and conversion metrics improve, the case for continued investment becomes self-reinforcing. TBC Bank Uzbekistan is building that case interaction by interaction.
