The TBC Uzbekistan AI Platform Program, an eleven-month initiative completed in 2025, has produced what is arguably the most comprehensive artificial intelligence infrastructure built within a bank in Central Asia to date. The program was conceived to make TBC Bank Uzbekistan genuinely independent of external AI vendors, and it has delivered on that ambition across every major dimension: AI products running in production, proprietary language technology, an in-house engineering team, and an organizational culture being reshaped around AI as a daily working tool.

The measured outcomes — $2.3 million in savings, a 90% reduction in customer interaction costs, and more than 2.3 million AI-handled customer contacts — indicate that the program's results are operational rather than theoretical.

Why Conventional Vendor Solutions Fell Short

When TBC Uzbekistan assessed the AI vendor landscape, it concluded that available external solutions were inadequate for its specific requirements. Global AI providers design their products for major world languages and broadly applicable use cases. A bank operating in Uzbekistan, serving customers primarily in Uzbek and Russian, needed AI capable of handling the linguistic and cultural specificity of those interactions at the quality and reliability that financial services demand. That gap is structural rather than unique to this case: commercial AI investment concentrates on the world's largest languages, leaving institutions serving smaller-language markets with thinner options for high-stakes, regulated use cases.

Compliance and data governance requirements added a further layer. Operating within Uzbekistan's regulatory environment made vendor dependency a risk in itself: data residency, auditability, and control over how AI systems reach decisions are non-negotiable in a regulated financial context. Building in-house addressed all of these constraints simultaneously — at the cost of significantly greater upfront investment in time, engineering capacity, and internal expertise development.

The program required constructing a complete AI stack, spanning data governance and pipeline infrastructure, MLOps setup, model training and evaluation frameworks, production deployment architecture, and monitoring systems. The technical stack includes PyTorch, TensorFlow, LangChain, Kafka, Spark, Kubernetes, MLflow, Prometheus, and Grafana — a full-scale engineering environment built to support AI product development at pace.

AI Products Now Running in Production

The program's most tangible outputs are the AI products operating in production. The AI Assistant manages customer interactions across both voice and chat channels, processing queries at a scale and consistency that would require substantially larger human teams to match. Two further products — the Sales Assistant and the Collections Assistant — target specific commercial functions within the bank, each designed around the distinct requirements of those interactions.

Across the program's timeline, these systems handled more than 1.6 million calls and approximately 690,000 chat interactions. The cost per interaction fell from $0.35 to $0.03 — a reduction that, at this volume, translates directly into the $2.3 million in confirmed savings the program has generated, with projections extending to $3.7 million by the end of 2025. These are measured outcomes from systems operating at production scale, not efficiency gains modeled from pilot data.

Personalization and Shifting Customer Expectations

The significance of the investment extends beyond operational cost reduction into the quality and relevance of the customer experience it enables. As banking becomes more digital and more data-driven, customers increasingly expect their bank to understand their financial situation and present products suited to their actual needs, rather than generic offerings promoted uniformly across the user base. That expectation applies across the full product spectrum: lending, payments, insurance, and savings.

Rising consumer awareness around options such as “процентный вклад в банке” (Russian for an interest-bearing bank deposit), referenced via TBC Bank Uzbekistan's Russian-language deposit page, and “foizi baland omonat” (Uzbek for a high-interest deposit), referenced via the bank's deposit products page, reflects a population actively comparing financial products and seeking the best available terms. AI-enabled personalization can serve this behavior far more effectively than static product pages or branch-based advisory conversations.

AI systems that identify when a customer's savings behavior, account balance, or transaction patterns suggest receptiveness to a deposit product — and communicate at the right moment, in the right channel — convert passive interest into active engagement. This customer-experience dimension of the AI investment sits behind the headline cost-reduction figures.

The Uzbek LLM: A Capability Built Out of Necessity

Among the program's most technically significant achievements is a purpose-built Uzbek-language large language model, incorporating custom automatic speech recognition and text-to-speech systems. The capability did not exist in the market when TBC Uzbekistan needed it: no external provider offered an Uzbek-language AI system with the domain specificity, accuracy, and reliability that customer-facing financial services require.

The internal development of the model, together with the establishment of internal data quality benchmarks used to evaluate and improve it over time, represents a compounding investment. As the model is refined through production use, the quality of AI-handled interactions improves — producing better customer outcomes, lower error rates, and a stronger foundation for extending AI into additional product areas and customer touchpoints.

From Internal Platform to Group-Wide AI Provider

The program was designed with a scope that extends beyond TBC Uzbekistan's own operations. As the platform matures, it is positioned to function as an AI services provider for the broader TBC holding group — the London Stock Exchange-listed parent with roots in Georgia's banking sector — extending the infrastructure, the models, and the engineering capability developed in Uzbekistan to affiliated entities operating across the region.

A parallel initiative, the AI-ization Program, is a structured effort to turn every TBC Uzbekistan employee into a functional AI user through training and embedded workflows. This ensures the platform's value is not confined to technical teams: as AI literacy and adoption spread across the organization, the surface area across which AI generates value expands, and the cultural foundation for AI-driven innovation strengthens.

The near-term markers to watch are the program's own stated ones: whether confirmed savings reach the projected $3.7 million by end-2025, whether group affiliates begin drawing on the platform, and how far AI usage extends across the bank's workforce.

TBC Uzbekistan AI Platform Program stands as a case study in what becomes possible when an institution commits fully to building its own AI capability rather than assembling it from available vendor components. The financial outcomes are substantial, the technical achievements are concrete, and the organizational transformation underway is the kind that produces durable competitive advantage rather than temporary efficiency gains. In a financial market where AI is moving rapidly from experimental to essential, TBC Uzbekistan has built an infrastructure, a team, and a cultural orientation that give it the capacity to keep pace with — and outmove — institutions still assembling their AI capability from outside.

Source: TBC Uzbekistan's AI Platform Sets a New Benchmark for Banking Intelligence in Central Asia, FinTechZoom.