From AI pilot to business impact: Five ways to turn AI pilots into business results

Ask any board member whether their company is “using AI” and the answer will be yes. Ask whether it is working, and the confidence drains fast. Enterprises have committed US$30bn to US$40bn to generative AI so far. Ninety-five percent of the resulting pilots show no measurable return on the balance sheet.¹ The most pertinent ask from leaders now is how to turn AI pilots into measurable business results. 

Recently, Emeritus Enterprise conducted a workshop in Hyderabad on what leaders must get right as AI moves from experimentation to enterprise-wide transformation. Herman Singh, CEO, NED, and expert in Digital Strategy, Exponential Leadership and Organizational Agility, led the session. Many stalled AI projects, never escape what he calls pilot purgatory, cycling through pilot after pilot without ever scaling AI across the business. Five decisions determine whether a company escapes it. 

Measure AI ROI by what improves, not what’s automated 

Start with how enterprises define success, because AI ROI depends on measuring the right thing. Tokens processed, chats automated and “AI-enabled” workflows are the metrics that make it into board decks. They are also, on their own, meaningless. The theory of constraints explains why. Speeding up any step in a process that was not the actual bottleneck produces zero net improvement, no matter how much faster that one step now runs.

“Vanity metrics create the illusion of progress. Outcome metrics, such as how many bottlenecks have been eliminated, show whether anything has actually improved.” 

Herman Singh, CEO, NED, and expert in Digital Strategy, Exponential Leadership and Organizational Agility 

The data suggests this confusion is now systemic, not anecdotal. Gartner’s most recent CIO survey found that 58% of CIOs face growing pressure to deliver AI-driven cost savings, yet only 36% believe they will actually meet those targets, and just 13% report significant value from AI tools so far. Worse, 51% expect AI to increase total cost of ownership over its lifecycle, not reduce it. The metric being demanded of technology leaders and the metric AI is actually producing are moving in opposite directions.

Choose the right AI use case 

Another common mistake is using AI for problems that do not need it. Rule-based, low-variance decisions are better served by deterministic software that behaves identically every time. AI’s comparative advantage lies in judgment-heavy, probabilistic territory. Using AI where simpler software would work costs organizations more than necessary. 

MIT’s research shows exactly this misallocation in the budget data. More than half of enterprise AI spending in 2025 went to sales and marketing pilots, the most visible and board-friendly use case, yet these delivered the lowest returns of any category. The same research found a stark split in execution quality. Pilots that paired internal specialists with external partners succeeded 67% of the time, versus 22% for organizations that tried to build everything in-house with IT alone.² 

Plan for AI failure before you scale

AI can give a convincing answer that is wrong. A system may appear reliable in testing and fail when conditions change, or more people use it. This calls for what can be described as responsible speed, a core principle of responsible AI. Leaders should put checks and rollback options in place before a deployment reaches customers, rather than after the first failure, and decide in advance how they will detect an error, who can intervene, and when they will pause the system.

Assign AI accountability and governance 

When an AI-assisted decision goes wrong, people and organizations remain responsible for the result. AI itself cannot be disciplined, sued, fired, or held responsible when it makes an error; a human always inherits the consequence. Effective AI governance means deciding, deliberately, whether a human sits in the loop, on the loop, or in control for each AI-assisted decision.

“Everybody talks about human in the loop. Nobody talks about human on the loop, and very few people talk about human in control.” 

Herman Singh, CEO, NED, and expert in Digital Strategy, Exponential Leadership and Organizational Agility 

Gartner warns that CHROs may have to manage the effects of AI systems they did not help design. CIOs may face workforce problems caused by tools developed without HR input. Gartner predicts that by 2029, 30% of organizations will have joint HR and IT teams to oversee AI ethics, skills-based talent management and human-AI performance.³ Accountability, in short, is a staffing decision, and most enterprises have not made it yet. 

Drive AI adoption by understanding how employees really use it 

AI can do more than most firms currently ask of it, yet many struggle to bring it into everyday work. Leaders often assume that a powerful system will draw employees in. In practice, people choose what is easy to access, helps with pressing tasks or is required for their job. When the approved option falls short, they may turn to personal accounts instead. MIT research found that employees at more than 90% of firms do so, even after their company’s official pilot has stalled.⁴

Gartner’s parallel research on change management reinforces the fix. It highlights that organizations that continuously adjust their rollout based on real employee response are four times more likely to achieve a successful change outcome than those that stick to a fixed plan.⁵ 

The gap between AI investment and business returns will persist if companies treat each pilot as a separate technology project. Leaders need to choose a problem worth solving, measure the result, plan for errors, assign responsibility and give employees a system they can use. These decisions cut across business, technology and HR, so CHROs, CIOs and CTOs must make them together. A pilot can show what AI is capable of. The harder test is whether a company can use it reliably across the business and show what has improved. 

(This article draws on an Emeritus Enterprise workshop, Emeritus Exchange in Hyderabad led by Herman Singh. To learn how Emeritus Enterprise helps organizations move from AI pilots to measurable business results, get in touch with our team, here) 

Also read: More insights on building AI-first organizations, skills, and HR strategy.  

1. 5 Things That Separate AI-First Organizations From Everyone Else 
2. Top 5 AI-Era Upskilling Trends for 2026 
3. The AI Workforce Roadmap: What HR Must Prioritize