AI Is Already Inside the Business. Why Is Measurable Value Still Hard to Scale? 

What business and technology leaders revealed about the gap between AI adoption and measurable impact 

At a recent dinner hosted by Zallpy and Plug and Play, business and technology leaders came together for a candid conversation about AI. 

The discussion moved beyond tools, demos, and predictions. Around the table, leaders shared what is already working inside their organizations, where initiatives are losing momentum, and what prevents promising experiments from becoming dependable business capabilities. 

Throughout the evening, participants responded anonymously to a series of questions, with up to 13 responses per question. 

This was not designed to be a broad market study. It offers something more specific: a snapshot of how leaders are experiencing AI adoption inside real organizations, with real operational constraints. 

The quick answer 

AI adoption is moving faster than companies can redesign their operations around it. 

The tools are available. Employees are using them. Pilots are underway. 

The harder work begins when AI needs to fit into real workflows, connect with existing systems, earn people’s trust, and improve an outcome the business can measure. 

The room reflected a larger market pattern 

What we heard during dinner was not an isolated experience. 

McKinsey’s 2025 global survey found that 88% of respondents said their organizations were regularly using AI in at least one business function. Yet nearly two-thirds had not begun scaling it across the enterprise, and only 39% reported any enterprise-level impact on EBIT. (mckinsey.com) 

Deloitte reported a similar contrast in 2026. Worker access to authorized AI tools increased by 50% in one year, growing from fewer than 40% to around 60%. Still, only 34% of organizations said they were truly reimagining the business with AI. (Deloitte Italia) 

Adoption is spreading quickly. 

Operational transformation is not keeping pace. 

Most companies have already started 

The participants at the dinner were not waiting for AI to arrive. 

Of the 13 people who responded: 

• Eight said AI was already part of their operations. 

• Four were running pilots. 

• Only one was still at the beginning of the journey. 

That means 92% were already piloting AI or using it in their operations. 

The challenge was no longer gaining access to the technology. It was moving from individual use and isolated experiments to capabilities the organization could trust, measure, and scale. 

The technology works. The organization resists. 

When participants were asked where AI initiatives usually stalled, six out of nine selected the same answer: 

The technology works, but the organization resists. 

A successful demonstration proves that a model can complete a task. It does not prove that the company is prepared to use it as part of a real operation. 

That requires answering a different set of questions: 

• Who owns the business outcome? 

• Who validates the output? 

• How does AI fit into the existing workflow? 

• What happens when the system is wrong? 

• Which business measure should improve? 

Without those answers, a pilot can remain a technical success with limited operational value. 

The bottleneck is often not the model. 

It is the operating model around it. 

AI exposes what the company has never formalized 

Another question focused on institutional knowledge. 

Ten of the 12 respondents said that at least some critical knowledge remained concentrated in individuals. Four said most of it still lived in people’s heads. 

This knowledge goes beyond documented procedures. It includes exceptions, workarounds, historical decisions, informal dependencies, and the judgment people use to keep operations moving. 

An AI assistant cannot retrieve tacit knowledge the organization has never captured or made accessible. 

Before applying AI to a process, companies need to understand how the work actually happens, not only how it is described in official documentation. 

In many cases, the right starting point is not a model. 

It is an operational diagnosis. 

Faster tasks do not always mean better operations 

Participants were also asked where skilled employees lost the most time. 

Six out of 11 selected repetitive analysis and reporting. Other responses included reconciling data, routing requests, and searching for information across documents. 

Consider a recurring executive report. 

The final output may be a presentation, but the work behind it can involve collecting data from several systems, resolving inconsistencies, updating spreadsheets, identifying changes, creating charts, and writing explanations. 

Generative AI may produce the final slides faster. But that only improves the last step. 

The larger opportunity is to understand why the data requires manual collection, why the sources disagree, and whether the information could be continuously available to decision-makers. 

AI should not simply help skilled people perform inefficient work faster. 

It should help remove unnecessary work from the operation and improve the decisions that remain. 

Leaders in the room were looking beyond cost reduction 

The final responses revealed an important difference between AI activity and business value. 

When asked what mattered most when bringing AI into the organization: 

• Seven chose moving faster and making decisions sooner. 

• Four chose reducing risk and error. 

• One chose creating a new competitive advantage. 

• No one selected cost reduction as the primary objective. 

Efficiency still matters. But the leaders in the room were not focused only on doing the same work with fewer resources. 

They were looking for faster responses, better decisions, lower operational risk, and new ways to compete. 

Recent BCG research points in the same direction. In an analysis of more than 600 large US public companies, only 6% of the sample qualified as AI leaders. Those companies were more likely to reinvest productivity gains to scale the business and create new opportunities than to focus primarily on cost cutting. (BCG Global) 

The real advantage is not simply doing less. 

It is increasing what the business is capable of doing. 

Three shifts that move AI forward 

1. Start with the operation 

Map the people, data, systems, decisions, handoffs, exceptions, and informal workarounds behind the process. 

AI is unlikely to improve a workflow sustainably when the company does not understand how it really operates. 

2. Define the business result 

Choose a measurable outcome before selecting the technology. 

That outcome might be a faster decision, fewer errors, a shorter cycle time, improved forecast accuracy, reduced exposure, or a better customer experience. 

3. Engineer the path into production 

Moving beyond a pilot requires ownership, integration, governance, monitoring, training, and continuous improvement. 

The solution may involve generative AI, machine learning, automation, data engineering, software integration, or a combination of these capabilities. 

The problem should determine the technology, not the other way around. 

What leaders need now 

At the end of the evening, participants were asked what would make the gathering worthwhile. 

Their answers focused on honest conversations, practical use cases, meaningful connections, and relationships that continued after the event. 

They did not need another demonstration of what AI might eventually do. 

They wanted to understand where it works, why initiatives stall, and how other leaders are turning experiments into dependable capabilities. 

AI is already inside the business. 

The next advantage will not come from adding more tools or accumulating more pilots alone. It will come from understanding the operation, choosing the right problems, and connecting technical execution to outcomes the business can measure. 

Business outcomes do not emerge from access to AI. They have to be engineered.