This is article five in a six-part series exploring new research findings from HCLTech and Raconteur. You can read the rest of the article series in ‘The Blueprint for AI Leadership’ here or download the full report here.
While data foundations are a critical enabler of AI, they are only part of the equation. Many organisations find that scaling AI also depends on the capacity of their architecture to support it. Unlike data initiatives, which can often be addressed incrementally, modernising enterprise architecture requires a more fundamental shift: away from infrastructure and application estates that are built for a different era of computing and towards environments designed for flexibility, interoperability and continuous change.
Who are AI Leaders & Followers?
AI Leaders, who represent 18% of the sample, are maximising ROI not simply by using AI to reduce costs, but by applying it to drive growth, innovation and improved customer experience.
AI Followers, by contrast, represent 60% of the sample. While they are realising returns from AI investments across business functions, they significantly trail Leaders in the higher-value use cases that are reshaping operating models and competitive advantage.
“Legacy technology constrains innovation,” says Mandy Lamb, head of value-added services at Visa Europe. “It limits how fast organisations can move, how much they can scale and how confidently they can experiment.”
Still, most organisations remain distant from this model. In fact, just 11% consider their architecture to be very ready for AI adoption. At the same time, only 10% say they are investing sufficiently in architectural modernisation to close that gap. This reveals a clear disconnect: While most organisations recognise their architecture is not ready for AI, very few are investing at the level required to address the shortcomings.
The challenge, therefore, lies not just in recognising the need for architectural change, but in executing it. Nearly half of organisations (47%) cite technology limitations as a reason for underinvestment in architecture. Budget constraints (31%) and, to a lesser extent, talent shortages further compound the challenge, with the latter more pronounced among Leaders than Followers.
These findings suggest that architectural modernisation is not simply underfunded but structurally difficult to deliver within the constraints of existing technology environments. Against this backdrop, organisations report that GenAI-ready architectures are further constrained by practical challenges, including integration complexity (29%), technical debt (25%) and vendor lock-in (22%). Together, these factors create a compounding effect in which the very systems organisations need to modernise are also the ones that prevent that modernisation.
The wider research reflects this tension. Organisations appear to be prioritising more visible, strategic aspects of AI adoption, such as integration and use-case development, while underinvesting in the underlying infrastructure required to support them. The result is a widening divide between organisations able to modernise their environments and those with ambitious AI strategies but limited ability to execute them.
When architecture becomes the bottleneck
For many organisations, the ambition to scale AI is running up against systems and enterprise applications that were never designed for it. But this challenge appears to intensify as organisations become more advanced in their AI journeys. AI Leaders are more likely than Followers to cite legacy systems and technical debt as a barrier to adoption and scale, at 34% compared with 20%.
This suggests a paradox of AI maturity: the further organisations progress, the more exposed they become to the limitations of their existing technology estates. Leaders may already have captured the more immediate efficiency gains from AI. As they move towards higher-value use cases – embedding AI into workflows, decision-making and enterprise transformation – legacy architectures become harder to work around.
“This paradox is pervasive,” says Chris Pickard, executive vice president, digital business services at HCLTech. “Large enterprises built their technology estates over decades to support transactional processing, reporting and operational efficiency. Those systems excel at what they were designed for but struggle with what AI demands.”
These environments were built for stability and efficiency, not adaptability. They are often tightly coupled and difficult to change, limiting how data can move across the organisation. “Many enterprises are still running systems that were never designed to interoperate. They’re dealing with inconsistent data definitions, incompatible formats and integration patterns built for a different era,” says Pickard.
Even organisations leading the way in AI adoption often find themselves constrained by their existing architecture. “The paradox exists because AI progress is often visible at the surface with pilots and proofs of concept, while the underlying foundations remain unchanged,” says Lamb.
This constraint is most visible at the application layer, where core business systems lock data and processes into rigid structures that are difficult to extend with AI. As a result, many organisations are able to generate AI-driven insights but struggle to operationalise them at scale.
This challenge is also reflected in how organisations perceive broader barriers to AI adoption. More than half (56%) cite legacy systems and technical debt as major constraints, highlighting how deeply embedded these limitations remain across the enterprise.
Pickard sees this pattern consistently across large enterprises. “Leaders who expect to layer AI onto existing estates discover that meaningful adoption requires more fundamental architectural evolution than anticipated,” he says. In effect, progress at the edge can mask stagnation at the core.
Leaders recognise that scalable AI depends on getting the basics right first
organisations overcoming this challenge take a different approach. They focus not only on building new capabilities, but on the conditions that make those capabilities sustainable. “Leaders recognise that scalable AI depends on getting the basics right first: interoperability, modularity and trust,” says Lamb. This is reflected in how they prioritise investment and sequence modernisation efforts over time.
Rethinking modernisation
Against this backdrop, the concept of AI-ready infrastructure is becoming a priority, but it is often misunderstood. Pickard describes this as a shift towards more composable architecture. “Scaling AI requires an architecture that makes compute, data and controls composable, so organisations can assemble production-grade AI solutions quickly without rebuilding everything each time,” he says.
This composability increasingly extends beyond infrastructure into the application layer, allowing organisations to reconfigure business processes and workflows as AI capabilities evolve. However, achieving this state requires more than new technology – it demands deliberate, sustained modernisation of the core.
“The balance comes from refusing a false choice,” says Pickard. “We don’t need to finish modernisation before we innovate, but we do need to modernise on purpose.”
Rather than pursuing large-scale transformation programs in a single sweep, organisations are increasingly adopting a staged approach. They decouple innovation from the legacy core, creating environments where AI can be developed and deployed quickly while modernising critical constraints in parallel. “Each modernisation increment unlocks multiple AI use cases. Those use cases then fund the next wave,” explains Pickard.
Each modernisation increment unlocks multiple AI use cases. Those use cases then fund the next wave
Lamb sees a similar trend, enabled by cloud native and API-based technologies. “Modernisation no longer has to mean a single, high-risk transformation. Leaders can prioritise the areas where change will have the biggest impact first,” she says.
However, infrastructure alone is not the full story. Technical debt is closely tied to organisational structures and ways of working. “The biggest trap is treating modernisation as a translation exercise. If we move legacy systems without redesigning them, we don’t eliminate debt, we just replatform it,” says Pickard.
Much of this technical debt sits in the seams of organisations, from fragmented ownership to enterprise applications that encode processes in ways that are difficult to adapt or scale. “AI exposes this brutally. If the enterprise can’t explain how a process works or where authoritative data lives, AI initiatives inherit that confusion,” he says.
Addressing this requires more integrated, cross-functional ways of working, with teams operating closer to both data and products. When culture and infrastructure evolve together, organisations are better positioned to reduce technical debt and unlock AI at scale.
Building from the foundations up
For organisations earlier in their AI journeys, the temptation is to layer new capabilities onto existing systems. But this approach rarely delivers sustainable results.
“The most expensive mistake is letting pilot convenience define long-term architecture,” says Pickard.
Instead, organisations need to stabilise and modernise the core. That means ensuring data is reliable and well governed, systems are modular and interoperable and that governance frameworks are established early. “Early decisions set the slope of your future,” he says.
The organisations successfully scaling AI are not those with the most advanced models, but those with the most adaptable environments. By modernising architecture, strengthening core foundations and evolving the way they work, they are removing the constraints that have historically limited progress.
For others, the challenge is becoming more urgent. As AI adoption accelerates, the gap between Leaders and Followers is widening, not just in adoption, but in the ability to scale AI into core operations. Bridging that gap will depend not on adding more technology, but on building an architecture and application environment capable of carrying AI into core business processes and operations at scale.
This is article five in a six-part series exploring new research findings from HCLTech and Raconteur. You can read the rest of the article series in ‘The Blueprint for AI Leadership’ here or download the full report here.
This is article five in a six-part series exploring new research findings from HCLTech and Raconteur. You can read the rest of the article series in 'The Blueprint for AI Leadership' here or download the full report here.
While data foundations are a critical enabler of AI, they are only part of the equation. Many organisations find that scaling AI also depends on the capacity of their architecture to support it. Unlike data initiatives, which can often be addressed incrementally, modernising enterprise architecture requires a more fundamental shift: away from infrastructure and application estates that are built for a different era of computing and towards environments designed for flexibility, interoperability and continuous change.
"Legacy technology constrains innovation," says Mandy Lamb, head of value-added services at Visa Europe. "It limits how fast organisations can move, how much they can scale and how confidently they can experiment."
