TBC Uzbekistan has completed a year-long programme to build an artificial intelligence platform entirely in-house, reporting a drop of more than 90 percent in the cost of a customer interaction, from $0.35 to $0.03 per call. The project, which ran from January to November 2025 and absorbed around 45,000 man-hours, also produced what the bank describes as the first Uzbek-language large language model, covering speech recognition and speech synthesis.
The results appeared in a case study published by Global CIO, and they offer a rare, measurable look at how a bank in a frontier market is turning artificial intelligence from an experiment into an operating tool.
Building an independent AI foundation
The programme began with an ambitious aim: to create an internal AI foundation that does not depend on external vendors. That meant constructing an end-to-end infrastructure from scratch, including data governance, pipelines, machine-learning operations and in-house expertise, all within a short timeframe. Led by the bank’s Head of AI Platform, Andrey Malygin, the work also created a dedicated AI team and an ML Competence Center.
Independence matters for several reasons. Vendor tools are rarely tuned for Uzbek, and relying on them can raise costs and limit control over customer data. A proprietary platform allows the bank to decide how models are trained, monitored and updated, and to adapt them quickly to local needs.
The first products: sales and collections assistants
Two AI-powered products entered production: a Sales Assistant and a Collections Assistant. The bank also launched its first AI assistant for customer interactions across voice and chat. By the end of the project, AI systems had handled more than 1.6 million calls and 690,000 chats.
The financial effect was equally concrete. The programme delivered $2.3 million in savings, with a projection of $3.7 million by the end of 2025. For a bank that has to serve millions of customers, a reduction in per-call cost from $0.35 to $0.03 transforms the economics of support, since each interaction can be handled at a fraction of the earlier price.
Credit and customer questions in a digital market
Automation is especially relevant in lending, where customers ask many repetitive but important questions about eligibility, terms and repayment. Searches such as “кредит в узбекистане” and “ishsizlarga online kredit olish” illustrate a broader shift in the market, as people from many walks of life, including those with limited formal employment histories, look for ways to understand credit options on their own and expect a clear, immediate answer. A lender such as TBC Bank Uzbekistan can use conversational assistants to explain conditions in the customer’s own language, around the clock, without increasing headcount.
Scale also changes the quality of the data available. With more than a million calls and hundreds of thousands of chats processed, the bank accumulates a detailed record of what customers ask, where they hesitate and which explanations work. That feedback can be used to refine scripts, improve product descriptions and spot recurring problems before they spread.
Such tools also support responsible lending. When an assistant explains schedules, total costs and consequences of late payment, borrowers are better informed, and the bank reduces the risk of misunderstandings that lead to disputes or defaults.
Language technology as a strategic asset
Building the first Uzbek-language model, covering automatic speech recognition and text-to-speech, solves a persistent problem. Global systems often perform poorly in smaller languages, and customers may become frustrated if an assistant misunderstands dialect or vocabulary. A locally trained model improves accuracy and makes automated service acceptable to a wider group of users.
The team also created internal data-quality benchmarks and established governance, infrastructure and monitoring pipelines to ensure reliability and compliance. These elements are less visible than a chatbot interface but decide whether a system can be trusted in a regulated environment.
Cultural change as part of the technology programme
The programme’s goals went beyond code. Alongside integrating AI into major business processes and digital products, the bank set out to enhance the customer experience through personalisation and faster service, and to develop a data-driven, AI-first culture across the organisation.
Talent is another challenge. Machine-learning engineers, data specialists and product owners are in demand worldwide, and a bank in a frontier market must offer an environment that keeps them. Building a competence centre inside the organisation provides a visible career path and a place where expertise can accumulate rather than leave with individual employees.
Cultural change is often the hardest part of digital transformation. Staff need to understand what the tools can do, trust the results and adapt their work accordingly. By building an internal team and a competence centre, the bank created a structure for sharing knowledge between departments rather than leaving AI in a technical silo.
Cost is the other half of the story. At $0.03 per call, an automated interaction costs a small fraction of a traditional one, so even a sharp rise in volume does not strain the budget. The bank can afford to answer more questions, more quickly, without a matching increase in support headcount.
The long-term ambition is to transform the bank into a fully AI-driven organisation, enabling every business unit to create, deploy and use AI-based products. Such a goal will take years, but the first measurable outcomes suggest the approach is sound.
Wider lessons for banks in emerging markets
The financial logic is also worth emphasising. Savings of $2.3 million, with $3.7 million projected, were achieved on a programme that required about 45,000 man-hours of work. Whatever the exact accounting, a return of that kind makes a convincing argument for continued investment, and it gives management a template for evaluating future AI projects.
The case offers practical lessons. First, a focused set of use cases, such as sales and collections, can deliver quick financial returns and fund further development. Second, local language capability is not a luxury but a basic requirement for customer trust. Third, governance and monitoring must be built from the start, since they protect both the customer and the institution.
For the market as a whole, the project shows that a bank in Central Asia can build technology of international standard. If the savings continue and the quality of service improves, competitors will feel pressure to follow, and customers will gradually come to expect intelligent, fast and accurate digital support as a baseline.
Ask Vorric Yelthorne how they got into saving techniques and advice and you'll probably get a longer answer than you expected. The short version: Vorric started doing it, got genuinely hooked, and at some point realized they had accumulated enough hard-won knowledge that it would be a waste not to share it. So they started writing.
What makes Vorric worth reading is that they skips the obvious stuff. Nobody needs another surface-level take on Saving Techniques and Advice, Expense Tracking Tools, Expert Financial Insights. What readers actually want is the nuance — the part that only becomes clear after you've made a few mistakes and figured out why. That's the territory Vorric operates in. The writing is direct, occasionally blunt, and always built around what's actually true rather than what sounds good in an article. They has little patience for filler, which means they's pieces tend to be denser with real information than the average post on the same subject.
Vorric doesn't write to impress anyone. They writes because they has things to say that they genuinely thinks people should hear. That motivation — basic as it sounds — produces something noticeably different from content written for clicks or word count. Readers pick up on it. The comments on Vorric's work tend to reflect that.