AI has moved quickly from experimentation to enterprise priority. But for many organizations, the harder challenge is turning AI investment into measurable business impact.

That was the central theme of a recent HCLTech LinkedIn Live discussion with Vijay Guntur, CTO and Head of Ecosystems at HCLTech, and Mark Beccue, Principal Analyst - AI at Omdia. The conversation explored findings from HCLTech’s latest enterprise AI market report, The AI Impact Imperatives, 2026, including why nearly half of major AI initiatives launched over the next 24 months are expected to fail.

For Guntur, the headline finding is not surprising, but it should focus attention on what separates successful AI programs from those that stall.

“This is not a technology problem alone,” he said. “You need to have the right foundation of the data, you have to have strong governance practices, and most importantly, you need to have the focus on creating business impact.”

That focus on business impact is critical. AI cannot be treated as a standalone technology layer or a bolt-on capability. It needs to be connected to the organization’s operating model, data environment, governance structures and business priorities.

Beccue made the same point from an analyst perspective. “AI is not a bolt-on,” he said. “Technology is not all of the issue here. It’s transformational, it’s innovation.”

In other words, enterprise AI success depends less on isolated experimentation and more on whether organizations can create the conditions for AI to scale responsibly, economically and repeatedly.

Leadership must move from reaction to readiness

One of the most striking findings from the report is that 83% of respondents said CEOs and boards do not adequately understand that underinvesting in AI may create existential risk.

Beccue sees this as a leadership and decision-making challenge. Many organizations are still trying to determine where AI should sit, who should run it and how to balance build-versus-buy decisions.

“There’s the struggle to land on centralized capabilities and investment,” he said. “How do we build this plane while we’re flying?”

Part of the challenge is that AI requires a different kind of innovation discipline. Organizations may have experience with software lifecycle management, product lifecycle management, cloud transformation or digital transformation, but AI introduces new questions around lifecycle management, risk, governance, accountability and operating model change.

Guntur warned that waiting too long could be risky because AI adoption is moving faster than previous technology transitions.

“Organizations are not very prepared, especially if you think about boards and CEOs and CXOs,” he said. “They’re not very well prepared to understand how disruptive AI can be and then take proactive steps rather than reacting.”

He added that AI transformation may not follow the slower adoption curves seen with earlier waves such as mobile, digital or internet-led transformation. “This transition is not going to take that long once it gets started,” he said. “This is going to be more rapid with AI.”

That pace also makes change management central to AI impact. Guntur emphasized that successful AI transformation requires operating model change, leadership alignment and organization-wide capability building, not just new tools.

“You need a significant amount of organization change management capability, which we don’t factor well in,” he said. “Most organizations struggle with that.”

For enterprises, this means taking people along on the AI journey, equipping teams with new skills and creating the structures needed for AI to become part of how work gets done.

The first imperative: Build the right foundation

The first imperative identified in the report is the right foundation. For enterprises, this means confronting the legacy applications, technical debt, fragmented data estates, process debt and skill gaps that limit AI impact.

Guntur described the problem through what he called “PTSD”: process debt, tech debt, skill debt and data debt. These forms of debt have accumulated in many organizations because modernization has often been delayed.

“All of these plague organizations that have been not modernizing for the fear of either lack of investment, or ‘if it’s not broken, why touch it?’” he said.

That becomes a major issue when organizations try to scale AI. AI depends on high-quality data, strong architecture, modern systems and the skills needed to operate them. If those foundations are weak, AI initiatives are more likely to struggle.

“If you don’t have the right foundation, AI is going to fail miserably, because AI feeds on high-quality data,” Guntur said.

At the same time, he argued that AI can help organizations address some of the debt that is holding them back. AI-enabled tools can accelerate software development, data lifecycle management and technical debt reduction. Guntur said he has seen technical debt reduction programs that once took multiple years accelerate by 20% to 25%.

For Beccue, the difficulty is that most enterprise infrastructure was not originally built for AI.

“Most of this infrastructure and systems are not purpose built for AI,” he said. “You’re making some very difficult decisions about what you can keep, what you can’t keep, what’s going to work and what’s not.”

Fragmented data remains a central obstacle. Enterprises have been discussing big data for years, but GenAI and Agentic AI are now increasing pressure to make that data usable, connected and governed. As Beccue noted, organizations still need to address silos, legacy systems, unstructured data, routing and latency before AI can deliver at scale.

Model choice, sovereign AI and token economics

As AI moves into production, the foundation question becomes more complex. Enterprises need to decide where models run, which models to use, how workloads are routed and how to balance performance, latency, security, sovereignty and cost.

Guntur pointed to the growing number of available AI models as evidence of the complexity ahead. Enterprises may need to manage frontier models, private AI, small language models, guardian models and routing systems that dynamically decide where workloads should run.

“Organizations, just like in the cloud era, had decisions to make on what’s private, what’s public, what can be optimized,” he said. “All of those similar things will be required with AI, but the complexity is a little more because you also have this innovation going at high speed.”

Sovereign AI adds another layer. Beccue described it as a systems-level challenge rather than a point solution, requiring coordinated decisions across public cloud, sovereign cloud, private cloud and on-premises environments.

“Sovereign AI is inherently multi-location,” he said. “Workloads are distributed across public cloud, sovereign cloud, private cloud and on-premises environments based on where the data residency, performance and regulatory needs are.”

AI economics will also become a board-level and operating model question. Guntur expects per-token costs to fall over time, but total token consumption to rise significantly as AI adoption expands across digital and physical environments.

“Per-token costs are going to drop,” he said. “But you should also think about total costs.”

That means enterprises need to understand how model choice, workload design, prompts, agents, compute environments and routing decisions influence the full cost of AI. Guntur compared this to the emergence of FinOps in cloud, suggesting that similar disciplines will be needed for AI and token consumption.

For Beccue, token economics has helped surface a broader issue: how to run AI more efficiently.

“Tokenomics is really about running AI more efficiently,” he said. “What’s my cost-benefit analysis to driving AI where it makes sense?”

From efficiency to business transformation

Many organizations are still prioritizing operational outcomes such as productivity, process automation and workflow efficiency. Beccue argued that this is understandable because many early AI use cases naturally lend themselves to internal productivity gains.

“AI is freeing up the humans to go after the revenue generation stuff,” he said.

Guntur agreed that efficiency is an important starting point but emphasized that enterprises also need to identify use cases that create top-line impact. He gave the example of cargo space in aircraft, where unused space loses value once the plane is in the air. AI can help optimize order intake, space utilization and dynamic pricing, creating revenue impact as well as operational efficiency.

He also pointed to Physical AI use cases in areas such as worker safety, public safety, production line simulation and plant configuration. In these environments, AI can help organizations reduce risk, improve throughput and identify better operating models before changes are made in the physical world.

Beccue added that retail provides another useful example, particularly around restocking and supply chain optimization. If a store is out of stock, it misses the opportunity to sell. AI can help identify availability issues in real time and support faster operational response.

The second imperative: Implement the right AI governance

The second imperative is the right AI governance. The report found that 76% of organizations have delayed AI deployments because of Responsible AI concerns.

For Guntur, the answer is to make governance systematic from the start.

“You need to really build that into your organization muscle,” he said. “As an afterthought, it’s very difficult to do.”

This includes policies, procedures, cross-functional teams, AI risk assessment, mitigation, testing, monitoring and recovery plans. Organizations need to test for security, data issues, bias, industry-specific requirements and failure scenarios before systems move into production.

Beccue connected this directly to trust.

“How do we trust an automated system?” he asked. “The number one concern of enterprises about AI is security, and that goes down to a deeper level to protecting data.”

Governance also needs to account for shadow AI. As employees and business teams experiment with tools independently, organizations need ways to scan, monitor and govern AI systems. Guntur compared this to citizen development: if governed well, it can create value; if not, it can create risk.

The third imperative: Select the right partners

The third imperative is selecting the right partners. The report shows that organizations working with expert partners see stronger outcomes in ramp-up time, skills coverage, business impact and AI-related cost management, with 90% saying partners are accelerating time to value.

For Guntur, the reason is simple: no enterprise can build the full AI ecosystem alone.

“No one company can do all this,” he said. “You need help from experts in this space to build organization strength, skill and capability.”

Enterprises know their customers, markets and business problems. But AI requires capabilities across chips, models, hyperscalers, infrastructure, governance, skilling, systems integration and technology delivery. Partners bring complementary expertise that can help organizations move faster and reduce execution risk.

Beccue described partners as the innovation engine of the ecosystem.

“The vendor community are the innovators,” he said. “Their experience is typically a breadth of experience.”

That breadth matters because partners can draw lessons from multiple implementations across industries and apply those learnings to new enterprise contexts.

Building AI as an operating capability

The discussion ended with a question on what separates AI programs that become operating capabilities from those that remain experiments.

For Guntur, the difference comes down to scale, foundation and business impact.

“Isolated experiments are unlikely to succeed,” he said. “They will cause a little bit of mistrust and frustration if you’re not seeing results.”

Beccue emphasized the need to iterate. AI programs cannot be set once and left unchanged. Organizations need the flexibility to refine, test, measure and adjust.

“You’re not going to get AI programs that become operating capability if you just set and go,” he said. “You have to be flexible.”

For business and technology leaders, the message from the discussion was clear. AI impact depends on more than ambition. It requires business discipline, strong foundations, scalable governance, operating model change, organization-wide capability building and the right ecosystem of partners.

As Guntur put it in his closing remarks, enterprises need to think beyond data and governance to the full transformation required.

“You need to think about your operating model change, you need to think about organization change management, and then building organization strength for the future,” he said.

The winners in enterprise AI will be the organizations that treat AI not as a project, but as a new operating capability built for scale.