Today's corporate sector is undergoing a transformation where artificial intelligence is becoming the foundation of new solution architectures. According to Gartner, by 2028, more than 50% of generative AI models used by enterprises will be domain-specific. This trend is driven by the need for deeper automation of cognitive work. A Microsoft study shows that 49% of conversations in Microsoft 365 Copilot supported cognitive tasks. General-purpose models often fail to provide the required accuracy, shifting the focus to specialized solutions trained on industry-specific data.
Modernizing legacy systems instead of complete replacement
Many large corporations still rely on legacy ERP systems. Making changes in such monoliths takes months, which stifles innovation and leads to the emergence of "shadow IT." The main challenges of legacy systems include low flexibility, high maintenance costs, data silos, and limited integration capabilities.
Completely replacing a legacy system is risky and expensive. A more effective strategy is phased modernization: retaining critical components while gradually integrating modern microservices and domain-specific AI models. For instance, using low-code platforms allows for designing hybrid architectures where new AI services coexist with existing databases.
Practical applications of AI in the financial sector
In banking, integrating domain-specific AI models helps address several critical tasks:
- Single customer view: aggregating data from various legacy systems to identify hidden behavioral patterns.
- Personalized offers: generating hyper-personalized products based on profile analysis.
- Compliance automation: detecting suspicious transactions for AML and reducing operational risks.
- Risk optimization: more accurate credit risk forecasting based on historical data.
Risk management and AI security
Deploying AI in critical infrastructure requires systematic risk management. Key aspects include:
- Security: protection against Prompt Injection (the top risk in the OWASP LLM Top 10 2025 ranking), data poisoning, and model jailbreaking.
- Reliability: stable model performance under unpredictable inputs in line with MITRE ATLAS tactics.
- Accountability: the ability to explain the logic behind AI decision-making, which is crucial for regulatory compliance.
To build an effective risk management system, enterprises are recommended to use the NIST AI RMF 1.0 framework, which structures processes around the functions of govern, map, measure, and manage.
Market implications
The shift toward domain-specific AI models means businesses can no longer rely on rigid, siloed legacy systems if they want to remain competitive. While this transition enables deeper automation of cognitive tasks and highly personalized services, it also elevates operational risks. Organizations will face increased pressure to secure their AI pipelines against emerging threats like prompt injection and data poisoning, making robust AI governance a business necessity rather than an afterthought.
How to prepare
- Implement phased modernization: Avoid risky full-system replacements by retaining critical legacy components and gradually integrating domain-specific AI models and microservices.
- Leverage low-code platforms: Build hybrid architectures that allow new AI services to coexist and integrate seamlessly with existing databases.
- Deploy risk management frameworks: Adopt the NIST AI RMF 1.0 framework to structure AI security around the core functions of govern, map, measure, and manage, protecting systems against OWASP and MITRE-identified vulnerabilities.
Prepared by a Software Ukraine member. Original publication.