In Uzbekistan, banks’ assessment of borrowers may soon go beyond credit histories and formal income statements. Scoring systems are now exploring data sources such as social media activity, app usage behaviour, and even the frequency of transfers from relatives. At the “KEY TECH: AI in Banks and Fintech” conference, panellists discussed how artificial intelligence (AI) can revolutionise customer evaluation. The event was organised by Khisam Communications.
The panel session on “AI and Scoring: New Approaches to Evaluating Borrowers” was moderated by Ulugbek Tavakkalov, Deputy Chairman of the Management Board at Kapitalbank. He noted that until recently, banks relied on limited datasets when building scoring models and developing digital risk management systems.
«AI now allows us to process significantly more data,» he said.
Panellists included:
- Nuriddin Lafizov, CEO of Alif Uzbekistan
- Archil Dzodzuashvili, CEO of Wings
- Sergey Kim, Head of Scoring and Big Data Analysis at Ipoteka Bank
- Akmal Bakiev, CRO of Anorbank
- Ravshan Kadirov, CFO of InfinBank
- Nauan Sarbassov, Senior Data Analyst at Asakabank
Build or Buy? Approaches to AI in Banking
Ulugbek Tavakkalov began by asking Nuriddin Lafizov whether banks should build AI tools in-house or adopt ready-made solutions. Lafizov responded that «there is no one-size-fits-all answer — it depends on the scenario.»
He shared an example: when Alif Uzbekistan urgently needed to launch AI-based outbound call functionality, the company opted for an external solution after estimating in-house development would take six months.
Lafizov also highlighted the bank’s experience with large language models (LLMs). Initially, Alif trained its chatbots using internal data, but later recognised the value of integrating advanced language models.
«We began using the ChatGPT API to improve response quality alongside our model,» he explained.
Where Should Banks Start with AI?
Tavakkalov asked which area banks should prioritise when implementing AI. Speakers agreed: credit scoring is the logical starting point.
«In banks, AI should always begin with credit scoring — it’s simpler and less costly,» said Lafizov.
He explained that this doesn’t require huge datasets or servers; basic models can be run on regular laptops. What’s key, he stressed, is having strong risk specialists who can define the right technical objectives.
Traditional vs AI Scoring
Tavakkalov asked Sergey Kim to elaborate on the differences between traditional and AI-based scoring. Kim replied that «AI is just a tool.» He outlined three approaches:
- Traditional (logistic regression)
- Advanced (boosting techniques)
- AI-driven (neural networks and transformers like ChatGPT)
The choice depends on opportunity cost and task complexity. As tasks become more complex, they demand more sophisticated tools and expertise — along with greater time and financial investment. For banks starting from scratch, simple rules can suffice at the outset.
AI for SME Lending and Alternative Data Sources
Akmal Bakiev discussed the challenges of applying AI to SME lending, noting that corporate credit lacks sufficient statistical data. In such cases, banks must rely on business strategies, internal rules, or proprietary datasets.
«The more SMEs a bank serves, the more successful its scoring models will be,» Bakiev said.
In response to a question about creative data sources, Sergey Kim mentioned retail sales data, social media activity, and digital footprints from social connections. He believes these sources remain underutilised in Uzbekistan.
«Retail currently lags behind, but once it advances, that data will be extremely valuable. We’ll be able to assess customer reliability using it,» he noted.
Nauan Sarbassov suggested partnerships with warehouse inventory systems and e-documentation platforms, which often contain tax-verified data. These offer more timely insights than semi-annual or annual financial reports.
Lafizov shared how Alif assesses individuals who receive remittances. Initially, the team considered using paper confirmations of transfers. But after identifying card-based transfers within its own systems, the bank began marking such customers and tailoring scoring models accordingly.
He also spoke of a neural network that, with user consent, analyses installed apps on their phone. For example, gambling apps may lower a risk profile, while an app like Duolingo could improve it — suggesting a more responsible and disciplined user.
Regulation and the Future of Risk Management
Tavakkalov raised concerns about tightening scoring regulations — specifically, the requirement to assess six months of card turnover. Sarbassov noted this penalises customers without official salaries or those using cash, potentially reducing loan approval rates.
Lafizov disagreed with blanket rules, especially in economies not yet fully digital. He cited machine learning models showing that clients with unofficial income — such as rental earnings — often remain reliable payers.
Sergey Kim recommended modular scoring to ensure business continuity if certain data sources become unavailable. Archil Dzodzuashvili agreed, adding that rigid rules won’t suddenly shift people to formal income — instead, «products should be convenient enough to make digital channels the obvious choice.»
Ravshan Kadirov offered a broader perspective. While CFOs and CROs once dominated risk discussions, roles like CTO, CIO and Chief Strategy Officer are becoming more influential — and a Chief AI Officer may emerge soon.
He stressed that all businesses inherently take on risk to earn revenue — the key is how intelligently and responsibly those risks are managed.
«Risk managers shouldn’t just administer — they should lead risk strategy,» Kadirov said. «They understand which risks are worth taking and how to manage them to maximise profitability. They’re not here to slam on the brakes — they should help shape the business model.»
He added that the more seriously banks treat risk management, the more likely regulators will move towards a risk-based approach. This would establish fairer rules, accounting for asset quality, NPL levels, reserve volumes, and financial health.
«If we apply one-size-fits-all regulations, some banks with clean assets and solid returns will be unfairly held back — unable to fully leverage their capital or reach potential earnings,» he warned.
«By applying differentiated risk profiles, banks will naturally filter out untrustworthy borrowers. In business lending, an unreliable borrower is often an unreliable entrepreneur. As a result, banks will finance viable private sector players, creating jobs and bringing in foreign currency.»
In a comment to Kursiv Uzbekistan, Kadirov concluded:
«I believe the market and regulator are maturing. The Central Bank has raised this issue before. But unless banks themselves build systems that allow for transparent risk classification — covering asset quality, reserves, financial standing, and business models — the regulator will lack the tools for a comprehensive assessment. That’s the core challenge.»
