Otbasy Bank: How Kazakhstan's Housing Finance Leader Automated 40+ Processes with Agentic AI
Agentic AI and multimodal document processing — eliminating appraisal fraud, reducing loan cycle time by 30%, and freeing 13 full-time employees from manual verification.
About the Client
Otbasy Bank is one of Kazakhstan's most critical financial institutions, serving as the country's primary housing savings and mortgage lending bank. With millions of customers and a mandate spanning savings programs, mortgage origination, and government housing initiatives, the bank's operational volume is enormous — and for years, much of that volume was handled manually.
The Challenge
Manual Workflows Across Lending and Back Office
The pain was sharpest in mortgage lending. Before a loan could be approved, risk and compliance teams had to manually audit a high volume of disconnected paperwork: Property Title Deeds, Technical Passports, Purchase and Sale Agreements, and Title Search Certificates. Underwriters rarely had the bandwidth to verify that a physical asset existed or that an appraiser had actually visited the property.
The bank's Loan Origination System (LOS) and Core Banking System operated in silos, requiring extensive manual data entry and repetitive human intervention at every handoff.
Beyond lending, the same bottlenecks plagued HR, finance, accounting, and legal — departments where employees spent hours on document checks, reconciliation, and government portal queries instead of higher-value work.
The bank evaluated global enterprise automation platforms but hit two disqualifying constraints: prohibitively high per-bot licensing costs and cloud-only deployment models that conflicted with Kazakhstan's data residency requirements.
The Solution
An Agentic AI and Multimodal IDP Architecture
Otbasy Bank deployed Menelaus AI — a Python-native agentic automation platform that combines RPA, AI Agents, and intelligent document processing in a single on-premise stack. The implementation started with the mortgage lending pipeline and expanded department by department over four years.
At the core of the mortgage workflow is an end-to-end agentic pipeline with three stages:
AI-Powered Document Classification and Extraction
The lending pipeline ingests, classifies, and extracts critical data using intelligent document processing fueled by Large Language Models. The system reads and digitizes legal ownership boundaries and transaction values from Purchase and Sale Agreements, structural dimensions and construction year from Technical Passports, and active liens or encumbrances from Title Search Certificates — processing 2,000 documents per day with zero errors.
Cognitive Asset Validation via Vision Language Models
To mitigate appraisal fraud, the AI Agent leverages Vision Language Models (VLMs) to visually audit property inspection photos. The VLM analyzes visual markers and metadata to confirm the appraiser was physically present at the property, eliminating fraudulent desk-based appraisals. This is not commodity OCR — it is a cognitive validation layer that closes a compliance gap manual review cannot reliably address.
Automated Data Entry and LOS Routing via RPA
Once the AI Agent reconciles data across all documents and approves the visual inspection, it hands off execution to high-speed RPA bots. The bots map all verified applicant and property metrics directly into the bank's LOS and update the application status to "Verified," pushing the clean file to the credit committee for final approval.
Beyond Mortgage Lending
Digital employees built in Menelaus AI Studio and managed through the Orchestrator now run across HR, finance, accounting, and legal — handling employee onboarding, account creation, document completeness checks, government portal queries, and financial reconciliation. All bots operate under centralized governance with role-based access, execution logs, and alerting. The entire platform runs within the bank's own infrastructure. No customer data leaves the perimeter.
The Results
Thousands of Hours Reclaimed
In one document-intensive process alone, the platform replaced the workload equivalent of 13 full-time employees — not by eliminating jobs, but by redirecting human effort from data entry and verification toward customer-facing and analytical work.
The cost advantage was equally significant. Menelaus AI's flat-fee enterprise pricing enabled the bank to scale from a handful of pilot bots to a bank-wide deployment without the escalating per-bot costs of legacy RPA vendors.
Why It Matters
Otbasy Bank's transformation illustrates a pattern increasingly visible in banking worldwide: the convergence of RPA, AI agents, and intelligent document processing into a single enterprise AI platform. Rather than purchasing separate tools for process automation, document extraction, and fraud detection, the bank deployed one integrated stack — reducing vendor complexity, accelerating time-to-value, and maintaining full control over sensitive financial data through on-premise deployment.
For financial institutions evaluating intelligent automation — whether to accelerate loan origination, close compliance gaps, or migrate from expensive legacy platforms — this case demonstrates that enterprise-grade results don't require enterprise-grade budgets.
"It is also nice to work using the national RPA-platform, which is not inferior to global foreign platforms."
Ready to automate at this scale?
Talk to an automation specialist who knows banking — and your numbers.