Why Enterprises Struggle to Turn AI Investments Into Business Value

While organisations continue to deploy AI tools and run pilots, many struggle to integrate the technology into core business processes, modernise legacy infrastructure and establish clear accountability. The shift towards AI-native operations also raises questions about engineering productivity, workforce readiness and governance, particularly as enterprises explore autonomous AI agents. In this interaction, Pankaj Vyas, MD & CEO, Siemens Technology and Services, discusses why AI initiatives struggle to move beyond experimentation, how organisations can redesign workflows around AI, and what it takes to build a workforce that combines human expertise with AI capabilities.

Beyond AI Experimentation McKinsey finds that AI adoption is rising faster than enterprise-level financial returns. What is the most common organisational failure preventing pilots from scaling? The most common reason AI pilots fail to scale is that organisations treat AI as a technology experiment rather than a business transformation.

Enterprises introduce AI into existing processes without rethinking how work gets done, how decisions are made or where accountability sits. This may accelerate individual activities while leaving the broader operating model unchanged. For example, AI may speed up an analysis, but if the output still passes through disconnected systems, multiple approval layers and functional silos, the overall business outcome remains constrained.

Moving beyond pilots requires organisations to start with a clearly defined business problem and establish measurable outcomes. They must also redesign workflows, make trusted data accessible, establish clear ownership and embed governance from the outset. At Siemens Technology and Services, we focus on integrating AI into engineering workflows to improve productivity and shorten development cycles while maintaining engineering rigour.

Ultimately, scaling AI presents an organisational challenge as much as a technology challenge. Sustainable value emerges when AI becomes part of how the enterprise operates rather than remaining an isolated initiative. You have warned against technical debt and fragmented infrastructure.

Which legacy-system constraint most often prevents enterprises from moving AI into production? One of the most common legacy constraints is the architectural decision to build systems as closed monoliths without open, standardised APIs. Many older operational and enterprise systems were designed to perform specific functions reliably in isolation, rather than share data across a connected ecosystem.

This creates an infrastructure of brittle, point-to-point custom integrations. When an AI application needs data from multiple sources, it must navigate a web of fragile, custom-coded connections. This goes beyond technical debt and creates a fundamental barrier to scalability and reliability.

If even one connection fails in production, it can compromise the entire AI workflow. The challenge becomes particularly pronounced in industrial environments, where enterprise IT must connect with operational technology without disrupting business continuity. Years of accumulated technical debt can make integration difficult, even when the AI model performs well.

The answer does not necessarily involve replacing every legacy system. Instead, enterprises need to modernise their architecture through secure APIs, modular platforms and consistent data standards, supported by clear data ownership. At Siemens Technology and Services, this approach drives our brownfield modernisation efforts.

We prioritise connecting existing assets to modern platforms without compromising operational reliability. Ultimately, scaling AI requires modernising the systems around it, not just improving the technology itself. Siemens Technology and Services uses generative AI for code generation, automated testing and validation.

How do you measure productivity gains without compromising engineering quality or safety? Organisations must measure productivity gains from AI-assisted engineering through outcomes, not simply the speed at which employees complete individual tasks. In engineering, faster code generation offers limited value if it creates additional work during validation, testing or release.

At Siemens Technology and Services, we apply AI to activities such as code generation, validation, testing and simulation. Our aim is to shorten engineering development and release cycles while maintaining the quality and discipline that industrial technology demands. This requires organisations to evaluate the development process end to end.

Cycle time and release speed matter, but teams must assess them alongside engineering quality, rigorous testing and operational oversight. The objective is to help engineers spend less time on repetitive activities and more time solving complex problems without weakening the checks that make engineering reliable. AI should support engineering judgement, not replace accountability for outcomes.

Organisations must retain validation and release discipline as AI becomes more integrated into development workflows. The real measure of progress is whether they can improve engineering productivity and accelerate delivery while continuing to meet the quality and reliability standards customers expect. You have described India’s Siemens centre as a driver of global product innovation.

How is AI changing the centre’s ownership of product architecture and engineering decisions? India’s role in global engineering is evolving from executing defined tasks to owning global products, platforms and long-term technology roadmaps. AI is accelerating this change.

Consider a practical example. In the past, an engineering team might have received an assignment to design a single component within a predefined product architecture. The team had limited influence over the core architecture because modelling the system-wide impact of a change required significant time and resources.

Today, AI-powered simulation and digital twins allow the same team to model the behaviour of an entire system in near real time. Engineers can use AI to explore multiple design alternatives for their component and assess their impact on overall system performance, efficiency and reliability. This changes the ownership model.

Teams can approach global stakeholders with more than a finished component. They can present data-driven evidence that a modification to the core architecture will deliver better outcomes for the entire product. They are no longer simply executing assigned tasks.

They are using AI-driven insights to challenge assumptions and influence foundational engineering decisions. For us, this creates an opportunity to deepen ownership of global platforms, influence critical engineering decisions and contribute more directly to Siemens' innovation agenda. What changes when enterprises redesign workflows around AI rather than simply add AI tools to existing processes?

When enterprises introduce AI into existing processes, they can accelerate individual tasks, but broader constraints often remain. Approvals, hand-offs, fragmented data and unclear decision rights can continue to slow down work, even when AI tools are available. The real shift happens when organisations rethink how work gets done instead of adding another tool to an existing workflow.

This requires an end-to-end view of business processes. Teams need to identify where decisions happen, which activities they can automate, where AI can support human judgement and where people must retain accountability. They must also address the foundations that enable AI to work reliably, including trusted data, clear ownership and appropriate governance.

The operating model must evolve alongside the technology. Business and technology teams need shared accountability for outcomes and the authority to redesign workflows rather than optimise isolated tasks. Organisations should measure success through improvements in cycle time, quality, productivity and customer value, not simply the number of AI tools they deploy or pilots they complete.

As AI automates more engineering tasks, how should organisations redesign roles and career paths to build a genuinely Human + AI workforce? As AI takes on more repetitive engineering tasks, organisations need to redesign roles around outcomes, judgement and problem-solving rather than simply automate activities within existing jobs. Engineers will need to understand not only their areas of specialisation but also how their work connects with the wider system, from product architecture to integration, performance and quality.

Deep specialists will remain essential, but their roles will increasingly involve collaboration across disciplines. They will use AI-enabled tools to explore options, validate solutions and make better-informed engineering decisions. This places greater emphasis on systems thinking, cross-functional collaboration and the ability to orchestrate AI-enabled workflows.

Career paths must support this evolution. Alongside deep technical expertise, engineers need opportunities to develop adjacent skills, apply AI in practical engineering contexts and learn continuously as technologies and responsibilities change. At Siemens Technology and Services, our approach centres on education, experience and exposure.

We invest in structured learning pathways and external learning platforms to help employees build these capabilities. We also host Communities of Experts across areas such as AI, cloud and cybersecurity to encourage knowledge-sharing across domains. Equally important are opportunities to apply these skills through project rotations and hackathons, which allow engineers to broaden their experience.

Our aim is to build a workforce where AI extends human capabilities while people retain engineering judgement and accountability. With AI fluency extending beyond specialist teams, how do you distinguish employees who can use AI tools from those who can identify and deliver valuable AI use cases? The ability to use an AI tool has become the baseline, but it is only a starting point.

A far greater capability lies in recognising where AI can solve a meaningful business or engineering problem and understanding what it takes to turn that opportunity into a reliable outcome. This requires domain knowledge, critical thinking and an understanding of the process being improved, not simply familiarity with the technology. I would assess whether employees can identify a genuine problem, evaluate where AI can add value and collaborate across disciplines to implement a solution.

They must also understand AI's limitations, validate its outputs and account for quality, security and governance. An interesting use case does not necessarily qualify as a valuable one. It must address a real need and deliver a meaningful outcome.

This does not mean every employee needs to become an AI specialist. The differentiator lies as much in mindset as in technical capability. Employees need the curiosity to question existing processes, the confidence to take ownership of problems and the ability to collaborate with people who bring different expertise.

Researchers, developers and data scientists all have important roles, but meaningful innovation happens when these capabilities come together around a shared purpose. The objective is not simply to use AI more effectively, but to create something of lasting value and make a meaningful impact. As enterprises move towards autonomous AI agents, what governance decisions must leaders settle before allowing these systems to act across business-critical workflows?

The first decision involves defining the boundaries of autonomy. Leaders must establish what an agent can decide and execute independently, where it needs human approval and which actions it must not take. These boundaries should reflect the risk and potential impact of each decision.

Human authority must remain indispensable for safety-critical or high-impact decisions. The second involves establishing clear ownership. Organisations need to determine who will set and enforce authorisation boundaries, including which data agents can access and which systems they can act across.

Without clear ownership, autonomous systems can quickly create gaps in accountability. The third involves ensuring explainability and traceability. Leaders need visibility into what agents do, which data they use, what decisions they make and the reasoning behind those decisions.

Organisations must make every significant action traceable, with mechanisms to monitor performance, detect unexpected behaviour and intervene when necessary. They must also establish incident-response procedures and the ability to suspend or reverse actions before deployment. Governance must form part of the operating model from the outset rather than become an afterthought.

This requires business, technology and data teams to work together to ensure responsible deployment.