AI Summit London: Managing Legacy IT and the Pace of AI Development
At the recent AI Summit in London, a panel on AI as a Competitive Advantage highlighted a key challenge for businesses: turning AI projects from concepts into production-ready solutions. Data from the Summit’s AI at Scale stream revealed that 80% of AI proof-of-concept projects fail to reach production.
Two panelists focused on how AI fits within existing enterprise IT systems, a critical factor in moving beyond proof-of-concept. Integrating new AI technologies with legacy infrastructure and datasets remains a significant hurdle.
Technical Debt and Enterprise Complexity
Ravi Rabheru, head of Intel’s AI centre of excellence for EMEA, identified technical debt as a major obstacle. Dara Sosulski, HSBC’s head of AI and model management, emphasized that larger companies face greater complexity and accumulated technical debt. This issue is also prevalent in government, where AI adoption is a priority but often hindered by outdated IT systems.
The UK Public Accounts Committee’s March 2025 report on AI use in government highlighted poor data quality and legacy IT as barriers. The Department for Science, Innovation and Technology (DSIT) confirmed that valuable data often remains locked in obsolete systems.
Building a Suitable Infrastructure
Sosulski stressed the need for IT leaders to evaluate if their data infrastructure supports AI initiatives. She described infrastructure as the backbone that should be modular and interoperable, enabling smooth access to applications and data across platforms.
However, she acknowledged that for many organizations, it’s difficult to predict when all necessary components for AI integration will be fully in place.
Build or Buy: Choosing the Right AI Approach
The panel discussed whether to build AI tools in-house or buy existing solutions. Sosulski pointed out that the decision depends largely on the business use cases. Common tasks like software development assistance, email drafting, document translation, and Q&A are now often handled by widely available tools.
For these generalist use cases, buying fully tested, enterprise-wide AI solutions that integrate well with existing infrastructure makes sense. Companies recognize the high cost and limited benefit of trying to develop such capabilities internally.
Keeping Pace Without Chasing Every New Model
While AI advancements, including agentic AI and artificial general intelligence (AGI), continue to emerge, Sosulski advised focusing on business-relevant developments rather than chasing every new foundation model.
She noted many foundation models end up offering similar capabilities for typical enterprise needs. After gaining familiarity with these models, organizations can focus on what works best for their specific use cases.
Practical Steps for AI Adoption
- Assess and prioritize AI use cases aligned with business goals.
- Evaluate existing data infrastructure for modularity and interoperability.
- Consider buying proven AI tools for general tasks instead of building them.
- Pilot AI models through proof-of-concept projects to validate fit and performance.
- Implement control frameworks and flexible IT setups to enable quick retraining and iteration.
With this approach, companies can steadily move AI projects into production without disruptive overhauls every few months.
For IT leaders looking to deepen AI skills and understand how to apply AI effectively within their organizations, exploring practical AI training courses can provide valuable insights and strategies.
Your membership also unlocks:
