Companies Pour Money Into AI, but Just 5% Deliver Results as AX Falls Into the ‘DX Failure Trap’

Companies rush to adopt AI solutions without redesigning workflows
Clashes between legacy reporting and approval structures and new systems deepen operational confusion
Despite massive investment, both profitability and quality deteriorate alongside efficiency

The number of companies adopting artificial intelligence (AI) is surging, yet only a tiny fraction have managed to translate it into higher revenue and lower costs. With companies layering new solutions onto existing workflows and reporting, approval, and evaluation structures, automation is increasingly backfiring in the form of deteriorating service quality and heavier workloads for frontline employees. In South Korea, the spread of vendor-driven implementation models tailored to government subsidies is also pushing company-specific system design and post-deployment support to the sidelines. This has prompted warnings that the failures seen during the Digital Transformation (DX) boom five years ago—when companies portrayed system launches as achievements while neglecting actual utilization and profitability—could be repeated in the era of AI Transformation (AX).

Corporate AI Stalls at the Adoption Stage

According to “The Widening AI Value Gap,” a Boston Consulting Group (BCG) report released on the 20th based on a survey of senior executives and AI decision-makers at 1,250 companies worldwide, “future-built” companies that deploy AI enterprise-wide and generate substantial value accounted for just 5% of the total. Another 60% had made significant investments but failed to achieve tangible revenue or cost gains. The remaining 35% had generated some returns and entered the scaling phase, although many executives said progress remained limited in both speed and scope.

A McKinsey survey of 1,491 respondents across 101 countries likewise found that more than three-quarters were using AI at work within their organizations, but more than 80% said it had produced no discernible impact on company-wide operating profit. Among the 25 factors reviewed by McKinsey, workflow redesign showed the strongest correlation with operating-profit contributions, yet only 21% of respondents said their organizations had fundamentally overhauled at least some workflows. Fewer than one in five organizations tracked key performance indicators for AI.

Table 1. Outcomes and Limitations of AI Adoption at Klarna and Starbucks

Rushed AI Adoption Only Deepens Confusion

Behind the failure of AI adoption to generate higher profits lies the way companies are pursuing AX, rushing to deploy solutions while postponing the overhaul of their operating systems. Swedish fintech company Klarna is a prime example. In February 2024, Klarna announced that its AI assistant had handled 2.3 million customer inquiries within a month of its introduction, managing two-thirds of all customer-service chats and performing work equivalent to that of 700 agents. The company placed the resulting cost savings front and center, saying average resolution time had fallen from 11 minutes to 2 minutes. The following year, however, CEO Sebastian Siemiatkowski acknowledged that an excessive emphasis on cost had degraded service quality. It was the result of aggressively pursuing automation without sufficiently overhauling workflows and quality-control practices.

While Klarna suffered a decline in customer-service quality, Starbucks repeated the same mistake in store operations. In September last year, Starbucks rolled out “Automated Counting,” an AI system designed to automatically tally inventories of milk, syrup, and other items, across its North American stores. It discontinued the system in May this year, just nine months after its introduction. The initiative was intended to reduce stockouts and ease the burden of manual inventory checks on baristas, but the system repeatedly confused similar milk products and failed to detect items displayed in stores. Starbucks subsequently returned to manual processes, saying it would standardize inventory-counting procedures across locations. By deploying AI across all stores while leaving its legacy IT infrastructure and fragmented supply chain intact, the company ultimately increased the burden on frontline employees.

Subsidy-Dependent AX Sidelines Operationally Tailored Design

As companies continue to deploy AI solutions without overhauling their operating systems, South Korean businesses are showing a growing tendency to pursue AX with government subsidies. A survey conducted last year by the Korea Chamber of Commerce and Industry and the Korea Institute for Industrial Economics and Trade among IT and strategic-planning personnel at 500 domestic companies found that the most frequently cited reasons for not using AI were inadequate technology and IT infrastructure, at 34.6%, and cost burdens, at 23.1%. Asked what policies they wanted from the government, 51.4% cited investment and research and development support, the largest share, followed by AI infrastructure development at 25%. The findings indicate that companies view government assistance as a major source of financing when considering AX.

The intense competition for recent support programs further demonstrates this demand. The “2026 Smart Service Program for Small and Medium-Sized Enterprises,” announced by the Ministry of SMEs and Startups on the 18th, drew applications from 545 companies, of which 175 were selected. Each of the 150 companies developing new solutions will receive approximately $36,000, while 25 companies upgrading existing services will receive as much as approximately $72,000 each. In the manufacturing sector, the “POSCO AI Track Smart Factory Program,” announced by the Korea Federation of SMEs in May, was structured so that the government would cover 50% of implementation costs, POSCO 30%, and the adopting company 20%. For businesses facing heavy upfront investment requirements, government funding offers a practical means of lowering the barriers to AX adoption.

In government-supported AX, however, investment decisions and responsibility for actual system use can easily become disconnected. A study published last year in the Journal of the Korea Contents Association found that system utilization depended on whether the support provided matched a company’s circumstances, whether the vendor properly understood frontline operations, and whether management and maintenance continued after deployment. This is because companies differ in their production sequences, approval structures, and data-processing practices. Common specifications may be applied to servers and sensors, but business software must be redesigned to fit each company’s actual operating environment. When vendors are preoccupied with meeting project schedules and inspection criteria, however, they are likely to make minor adjustments to preexisting solutions and deploy them as-is. Systems built in this manner fail to integrate properly into existing operations, forcing employees to use legacy processes and new software simultaneously.

Repeating the DX Failures of Five Years Ago

Similar side effects emerged during the DX boom that swept through the corporate sector roughly five years ago. Companies introduced enterprise resource planning (ERP), customer relationship management (CRM), and mobile work systems in quick succession, but made virtually no changes to their existing reporting procedures or approval practices. As the launch of a system itself was presented as an achievement for senior management, actual adoption rates and contributions to profitability were relegated to secondary concerns. This is why new software hardened into a separate layer of procedures imposed on top of existing work.

Recent assessments of Hyundai Capital’s digital platform illustrate the same failure formula. Hyundai Capital invested heavily in an AI-based credit assessment system and upgrades to its proprietary application, but the platform was ultimately regarded as little more than a channel through which customers purchasing Hyundai Motor and Kia vehicles could check and apply for installment financing. Without proprietary services and revenue models capable of attracting customers, its IT investments failed to translate into greater sales competitiveness. The disconnect between DX spending and business performance persists to this day.

The situation overseas is no different. In Vietnam, companies that installed software with various forms of DX support encountered operational confusion after failing to integrate their internal systems. Sales, accounting, and inventory-management systems operated separately, while data-entry standards differed by department. Employees were forced to enter the same information repeatedly into multiple programs, while companies also incurred additional system-maintenance costs.

German discount retailer Lidl’s ERP replacement project likewise ran into a clash between existing business practices and a new system. Lidl began building an ERP system based on software from German enterprise software provider SAP in 2011, but the company’s practice of calculating inventory value at purchase prices was incompatible with SAP’s standard methodology, which used selling prices. Rather than changing its operating procedures, Lidl chose to extensively customize the system, causing the project’s scope and costs to balloon. After investing approximately $585 million over seven years, the company abandoned the project in 2018 and returned to its previous system. By attempting to force new software onto an unchanged operating model, it incurred massive costs with little to show for them.

Will the Management-Frontline Disconnect That Derailed DX Recur in AX?

Such missteps were widespread across DX projects at the time. BCG’s 2020 analysis of 895 DX initiatives found that only 30% achieved their targeted value and delivered sustained change. Another 44% generated some results but fell short of their original objectives, while 26% created less than half of the targeted value. BCG concluded that success or failure depended less on the technology itself than on organizational structures, operating models, business processes, and corporate culture. Companies that introduced new systems without changing how they worked failed to achieve the results they had anticipated.

A McKinsey survey released the following year reached a similar conclusion. Among 1,034 respondents who had participated in corporate transformation initiatives over the preceding five years, fewer than one-third said those efforts had improved organizational performance and sustained the gains over the long term. Even companies deemed successful captured only 67% of the financial value they could have expected on average. Notably, 20% of the total value lost occurred after system implementation had been completed. Although companies launched new systems, they failed to embed them in everyday operating structures such as management meetings, business planning, and employee evaluations, allowing the returns on their investments to dissipate.

Senior management stood at the center of that divide. In the same McKinsey survey, senior executives were nearly 20% more likely than employees at other levels to believe that transformation objectives had been translated into concrete terms aligned with employees’ actual work. Senior leaders believed they had provided sufficient direction, while frontline employees did not even know what they were expected to change or how. If executives continue to demand in-person briefings, approve decisions using legacy forms, and conduct performance evaluations based on outdated metrics, employees have little incentive to alter the sequence of their work on their own initiative. If an unauthorized experiment fails, the employee bears the responsibility. This accountability structure is highly likely to recur in AX.