Next Gen RPA Platform

Turn RPA into a strategic in-house automation platform: scale automation of banking operations without growing the back office and without vendor lock-in. Goals: complete the migration of the platform to an in-house, fault-tolerant open-source stack with zero license fees; adopt an open-source workflow orchestration engine as the target core for robots and end-to-end processes; grow the portfolio to 100+ robots in production; make routine operations 10+ times cheaper than manual processing with payback within six months; cut robot development time to 2–3 weeks; embed locally hosted AI models within the bank's perimeter so robots can handle scans, documents and free text; build the RPA developer profession inside the bank.

2026

100+ robots in production across operations, lending, finance, compliance and IT/HR; up to 7x higher throughput in automated processes; the same workload handled 10+ times cheaper than manually from the second year; payback within six months; a vendor-independent, fault-tolerant platform with an approved target orchestration architecture.

In-house RPA platform (Python); open-source workflow orchestration engine with durable execution (target core, pilot); open-source relational database in a high-availability cluster setup; CI/CD, monitoring, centralized logging and BI reporting tools; virtualized execution environment for UI robots; in-house document recognition service; locally hosted AI models on in-house GPU servers.

The project does not use solutions from the Global CIO catalog

Migrating platform data from commercial and legacy databases to an open-source DBMS without stopping running robots, while moving to a high-availability setup. Adopting a workflow orchestration core changes how robots are designed: existing robots must be re-architected as workflows without disrupting production, and the target architecture had to pass the bank's Architecture Committee. Both workstreams were delivered by a small in-house team of four engineers alongside ongoing robot support. UI robots are sensitive to changes in interfaces, external websites and legislation. AI adoption in a bank is constrained by security: models must run locally, with human oversight of high-risk decisions.

Context. The bank processes large volumes of routine operations: reconciliations, payments, repayments, document issuance, reporting, compliance checks and employee requests. Growing volumes and changing regulations were pushing the back office to expand.

Solution. Instead of buying a commercial RPA platform, the bank develops its own. Robots are written in Python and work in the bank's systems the way an employee does — on top of existing interfaces, without re-engineering the IT landscape. Code goes through CI/CD; robots are covered by monitoring, centralized logging and BI reporting.

Platform modernization (January 2025 – October 2026). Two workstreams, each staffed by two engineers. 1) Database migration (Q1 2025 – Q1 2026): platform data was moved from commercial and legacy databases to an open-source DBMS in a high-availability cluster setup — without stopping production robots. The legacy databases are fully decommissioned from the RPA stack. 2) Orchestration core (since summer 2026): an open-source workflow orchestration engine was selected as the target core for robots and end-to-end processes; the architecture was approved by the bank's Architecture Committee and a pilot is under way. Every process step is persisted, failures are retried automatically and multi-robot processes become manageable workflows instead of separate scripts.

Delivery process. Requests from business units go into a shared backlog (Demand Management), are prioritized within each business line and included in a quarterly release. Developing one robot takes 2–3 weeks on average; acceptance is done jointly with the business.

Intelligent automation. The next stage is IPA: locally hosted AI models extract data from scans and emails and classify requests, while robots execute actions in the systems; complex and high-risk cases are routed to an employee.

People. The RPA developer profession has been established: coding standards, a quick-start project template and a competency profile on the HR portal.

Kazakhstan 600+ branches