Premkumar Balasubramanian, CTO at Hitachi Digital Services, highlights that successful AI strategies focus on measurable business outcomes, not experimentation. He explains that the biggest barriers to scaling AI are structural issues like poor data, unclear ownership, and weak governance, not technology. He adds that enterprises must treat AI as a core operating capability, with strong reliability, accountability, and long-term business alignment.
1. What sets successful enterprise AI strategies apart from the rest?
Successful enterprises that stand out from the rest treat AI as an outcome-driven operational capability, and not as a technology experiment. Their AI investments are anchored to derive measurable business value, such as cost reduction, cycle time improvement, reliability, risk, and revenue. Usually, they embed AI directly into end-to-end business processes where value is realized recurrently.
AI is designed to function in practical operating environments right from the start. Importantly, I think, there is a balance between the use of AI to automate decision-making on one hand, while maintaining human accountability on the other.
Enterprises that also adapt their operating models in terms of funding and governance mechanisms are the successful ones who take AI from pilot to impact.
2. What are some of the key challenges for enterprises in moving from pilot AI solutions to full implementation?
I think the most common roadblocks are structural, and not technical. Many enterprises struggle because of disconnected pilots with core business processes. They usually lack clear ownership and outcome accountability.
Underestimating AI’s operational complexity in production is another big reason. Reliability, data quality, governance, and change management are all part of it. Fragmented data, siloed teams, and funding models optimized for experiments rather than sustained operations further slow down the scale. Moreover, an unclear division of humans and AI responsibility erodes trust and adoption.
At Hitachi, we’ve addressed this through R2O2.ai, which explicitly bridges research to operations to outcomes, industrializing AI by embedding it into real business workflows with clear operational ownership, outcome metrics, and trust controls from day one.
2. With the growing significance of AI, how can enterprises re-imagine concepts of reliability, availability, and risk mitigation?
Enterprises must shift beyond traditional uptime metrics. They should rethink reliability as outcome assurance. AI systems generally fail because of degraded decisions, changing intent, or introducing silent risk into operations.
To avoid failures, enterprises should adopt these three shifts:
I. Designing reliability at the agent and workflow level
It has become imperative for enterprises to have explicit controls over agent behavior, scope, handoffs, and escalation as AI agents increasingly execute decisions autonomously. This is where our HARC Agents approach acts as a foundation, i.e., treating agents as managed operational entities, not black boxes.
II. Extending observability from infrastructure to outcomes
Critical AI systems for business operations demand observability around inputs, decision-making, actions, and impact on the business outcomes to identify any drifts, biases, or failures in advance.
III. Thinking about risk management as a constant process, not one-off events
Businesses need constant oversight via Agent Management, where guardrails, policies, auditing, and human interventions are required.
3. How should enterprises balance customized AI solutions with scalable platforms?
The two factors - customization and scalability need to be seen as synergistic rather than contradictory when applied within an organization. Scalable AI frameworks will serve as the common underlying base for such purposes as data architecture, agent management, observability, security, and governance.
As a result, customization needs to take place in the process and decision-making stages where AI technologies are fine-tuned according to each business process, regulatory environment, and metrics used to assess performance.
4. In today’s cost-conscious environment, how can organizations justify AI investments to the board?
Organizations should justify AI investments the same way they justify any major capital initiative, through clear linkage to measurable business outcomes. AI should not be positioned as an innovation spend, but as an operating lever. It should be tied to cost efficiency, productivity, risk reduction, resilience, and revenue protection or growth.
Boards respond when AI investments are framed around specific value drivers such as automation of high-cost processes, reduction in operational friction, improved reliability of critical workflows, or mitigation of regulatory and operational risk. This includes defining leading and lagging metrics upfront and committing to time-bound value realization.
Organizations should give importance to demonstrating that AI investments are industrialized, not experimental, with reuse through platforms, disciplined governance, and controlled operating costs. In uncertain environments, the strongest AI cases are those that show how AI improves the economics and resilience of the core business, not just future optionality.
In short, AI earns board confidence when it is framed as a durable, outcome-driven operating capability, with accountability, transparency, and a credible path to return.
5. What are some key changes for Enterprise AI in the next 2-3 years?
I am convinced that the transition of Enterprise AI will be from experimentation to industrialized implementation. To begin with, AI will penetrate further into the business operations process, with self-driven and assisted AI driving end-to-end workflows, rather than single actions. Secondly, concerns about reliability, visibility, and governance will take precedence because AI will play an increasingly important role in business operations, and organizations will require guarantees regarding the results, not just the model’s efficiency. Thirdly, organizations will align around AI operating models based on platforms. The transition will also be characterized by an explicit clarification of the role of people, namely, less active participation in the execution and more responsibility for goals, risks, and results.
6. What is one key mistake enterprises should avoid when adopting AI?
From my perspective, the worst possible thing that enterprises could do in this situation is to treat AI as merely another wave of technological innovation and not as a lasting change in the logic of how things operate within their field.
Approaching it from the outside only keeps them from unlocking its potential value, and what separates the companies that will progress from those who don’t is their ability to align themselves with this new paradigm.
