Companies struggle to see AI benefits despite heavy investment, citing data and expectation gaps

Many companies are investing heavily in "AI transformation" (AX), adopting AI models and creating dedicated teams to encourage their use. However, the reality on the ground often shows little change from existing work practices. A report by MIT's NANDA highlighted that 95% of organizations failed to achieve measurable financial gains from generative AI, despite substantial spending.

“

If a company's unique domain knowledge, work manuals, and project histories are not organized in a format AI can understand, its responses will remain general or even hallucinate.

The primary obstacle identified by experts is the "data" problem. "If a company's unique domain knowledge, work manuals, and project histories are not organized in a format AI can understand, its responses will remain general or even hallucinate," explained Kim Dong-min, a team leader at KT's AX business division. Even sophisticated AI models struggle without specific business context, providing useless answers.

This data is often scattered across internal systems and not properly organized for AI use. "Most data input into AI within companies is unstructured, like PDFs and Word documents," noted a team leader at a major domestic corporation involved in AX. "This data requires refinement, but many companies don't collect or refine it sufficiently." The process of collecting, refining, and converting data for AI use also creates human bottlenecks, as employees may view requests to change document formats or notations as burdensome.

“

Most data input into AI within companies is unstructured, like PDFs and Word documents. This data requires refinement, but many companies don't collect or refine it sufficiently.

Another significant issue is "unrealistic expectations." While AI's benefits are most apparent in repetitive tasks with sufficient data, attempting to implement it in difficult areas leads to wasted costs and minimal returns. For instance, applying AI to infrequent tasks like fault diagnosis (occurring every 2-3 years) can consume excessive time and resources due to insufficient data and high validation standards. "Companies have very high expectations for AI, but some areas are genuinely difficult to implement," said Heo Young-shin, Chief Business Officer at MachinaLabs. "The likelihood of successful transformation is higher in areas with specific AI adoption goals and measurable effects. Otherwise, internal consensus on the cost-effectiveness of AI adoption weakens."

“

Companies have very high expectations for AI, but some areas are genuinely difficult to implement. The likelihood of successful transformation is higher in areas with specific AI adoption goals and measurable effects. Otherwise, internal consensus on the cost-effectiveness of AI adoption weakens.

Experts emphasize that successful AI transformation goes beyond mere technology adoption; it requires a thorough review and redesign of business processes. "Innovation presupposes changing the way we work," stated Heo Young-shin. "We need to redefine tasks from an AI perspective, considering where it can be used." Kim Dong-min added, "AX is a process of redefining work using AI as a tool. Company-wide agreement on performance metrics like reduced work hours and cost efficiency, along with verification, is necessary for pilot projects to translate into actual results."

“

Innovation presupposes changing the way we work. We need to redefine tasks from an AI perspective, considering where it can be used.