AI Transformation That Accelerates Value Creation [Transformation Execution] ─ Moving Beyond PoCs and Connecting AI to Decision-Making and Financial Outcomes

The use of generative AI is rapidly expanding. Individual productivity is steadily improving, and PoCs have increased. However, executives do not feel optimistic. Although AI utilization is progressing, workload has not decreased, and lead time, quality costs, inventory, profit and loss, and cash flow have not significantly changed.
According to the Information-technology Promotion Agency (IPA) “DX Trends 2025” (comparison of Japan, the United States, and Germany; 1,535 Japanese companies), approximately 80% of Japanese companies are engaged in digital transformation initiatives, reaching the same level as the United States. However, less than 60% reported “achieving results,” compared with over 80% in the United States and Germany^*1. In addition, a global 2025 survey (105 countries, approximately 2,000 respondents) found that less than 40% of companies that introduced AI could report enterprise-wide impact on earnings before interest and taxes (EBIT)^*2.
What should be questioned here is not the presence or absence of digital transformation or AI adoption, but the unit of design for initiatives and the connection to outcomes. If efforts remain limited to efficiency improvements in individual operations, they will not lead to overall results. If outcome indicators are not designed, improvements at the operational level will not be captured as financial outcomes. Furthermore, unless decision-making authority and criteria are changed, the acceleration of output generation by AI will not lead to workload reduction.
In this insight, this structure is organized into three disconnects (Figure 1).

Applying AI to individual workflows produces localized efficiency improvements. However, their sum is not equal to overall optimization.
Consider an example from the supply chain management (SCM) domain in manufacturing. A company introduced AI in order processing, inventory inquiry, and billing processing, achieving efficiency improvements of 10–20% in each. However, the cash conversion cycle did not improve. The cause of inventory stagnation lies in multiple factors such as demand forecasting, production planning, procurement lead time, and special order responses in sales. Even if inventory inquiry is accelerated, cash will not move unless the decision-making structure changes. Therefore, AI initiatives must be prioritized not only by ease of implementation but also by their impact on overall constraints.
AI accelerates each process. However, it does not necessarily accelerate the entire enterprise.
The same applies to the quality domain. Even if the creation of quality records is automated with AI, quality costs will not decline unless these are connected to deviation prevention, root cause analysis, and shipment decisions. In the sales domain as well, even if proposal creation is accelerated, gross margin will not change unless pricing decisions and deal profitability change.
The sum of workflow-level optimizations does not match optimization of the entire business process, from order intake to shipping, billing, and payment. The same IPA survey shows that while Japanese companies have a high proportion of initiatives focused on optimizing individual operations, companies in the United States and Germany have a higher proportion focused on enterprise-wide process optimization^*1. The structure in which the accumulation of partial optimizations does not lead to overall results is particularly evident in Japanese companies. In AX, the question should not be “which operations to apply AI to,” but rather “which business events, if changed, will remove overall constraints and connect to financial outcomes.”

The CFO function has a clear view of financial figures but has not fully visualized the causal structure behind them. On the other hand, frontline operations understand actual conditions but cannot design connections to profit and loss and cash flow.
As a result, the effect of AI utilization remains limited to activity indicators such as “it was convenient” or “usage increased” and is not converted into enterprise value. In other words, the issue is not that results are not being achieved, but that the structure for linking outcomes to measurable enterprise value has not been designed.

Even if AI accelerates output generation, workload does not necessarily decrease. Much of the workload in companies arises not from tasks themselves but from decision-making processes such as confirmation, approval, meetings, alignment, and exception handling. Therefore, reviewing organizations and systems alone is insufficient; unless the hierarchy, criteria, and authority of decision-making are redesigned, workload will structurally persist.
In companies where PoCs are proliferating, the issue is not that there are too many AI initiatives. Rather, criteria and authority for stopping them have not been designed.
These three disconnects occur regardless of the domain of AI utilization.
AI utilization needs to be designed separately in two domains: horizontal AI domains that have broad, shallow effects across the enterprise, and vertical AI domains that deeply penetrate specific operations, functions, or industry processes such as sales, SCM, quality, development, and regulatory compliance.
Horizontal AI targets common tasks such as document creation, search, summarization, and internal inquiries, generating return on investment through workload reduction, outsourcing cost reduction, hiring suppression, and reflection in workforce planning. In contrast, vertical AI moves gross profit, cash flow, and quality costs only when connected to decision-making such as pricing, inventory and production decisions, and deviation prevention.
In both domains, without a design that connects AI utilization to workload, decision-making, and financial outcomes, return on investment dissipates.

*1 Information-technology Promotion Agency (IPA), "DX Trends 2025," June 2025
*2 McKinsey & Company, "The State of AI: Global Survey," November 2025

The design structure for translating AX into execution consists of the following five layers:

• L1 Strategy (enterprise value targets and North Star)
• L2 Governance and decision-making (authority and decision criteria)
• L3 Business processes (business redesign and elimination)
• L4 Technology and data (EA, data, ontology)
• L5 Organization and talent (roles, required workforce, evaluation systems)

These five layers are characterized by encompassing not only business processes and technology addressed by traditional target operating models (TOM), but also decision-making structures, governance, workforce allocation, and connection to financial outcomes.
However, the five-layer model only indicates design targets; it alone cannot drive AX. Each layer must be traversed through the processes of visualization, semantic structuring, decision and control, operational deployment, and harvesting results. Figure 4 shows these concretely for each of the five layers.

*ROIC = Return on Invested Capital
*CCC = Cash Conversion Cycle
*RACI = Responsible / Accountable / Consulted / Informed responsibility assignment framework
*Value certification gate: criteria for investment decisions and continuation decisions for AI initiatives
*SOP = Standard Operating Procedure
*UAT = User Acceptance Test

For example, in the SCM domain, business events from demand forecasting to payment collection are visualized, and how each affects CCC and gross profit is structured. Based on this, authority and automation conditions for inventory, production, and delivery decisions, as well as financial reflection, are designed. Only after this design does AI become not merely automation of inventory inquiry, but a transformation tool that simultaneously affects inventory, throughput, and cash flow.
Why are efficiency gains not visualized and returned across the entire company? The answer is that only L3 business processes were changed without addressing L2 decision-making structures and L5 organization and talent. In AX, it is insufficient to change only one of the five layers. Only when all five layers are traversed through the five processes and business events, decision-making, data, EA, and workforce planning are connected through causal relationships can PoCs be converted into enterprise value.
In this context, proceeding to implementation without “visualization” and “control” is the structural cause of the proliferation of PoCs.
The challenge of AX is not limited to technical implementation and automation. It lies in designing authority regarding who decides what, what to stop, and how far AI should be entrusted. In decision-making design for “control (Govern),” responsibility allocation (RACI) is defined for each AI initiative. Entering implementation without “control” undermines the stability of EA. Implementation without control is not transformation; it is disorder.
It is also important to note that the “information asymmetry” identified in Chapter 1 is resolved within this structure. The CFO function is responsible for value certification of AI investments and resource allocation, but the design of financial outcomes cannot be completed independently. It is necessary to jointly design the causal structure that connects frontline events and financial key performance indicators with business, operations, and data functions.
Based on this premise, ABeam Consulting advocates an approach called “AI Native,” which reconstructs operations, decision-making, and workforce allocation on the assumption of AI. While an AI center of excellence is essential for standardization and control, it cannot independently handle business elimination, decision-making authority, workforce reallocation, and harvesting results.
The value realization layer integrates multiple functions, including the center of excellence, and connects AI to operational deployment and enterprise value.

Moving beyond PoCs does not simply mean transferring AI to a production environment. It means making definitive decisions on what to stop, what to delegate, how to allocate people, and which outcomes to reflect in financials. Even if mechanisms are in place, AX will not move without management’s will to make these decisions.
Leadership in AX is not about encouraging AI utilization. It is about stopping initiatives that should be stopped and redesigning operations and decision-making. Decisions to withdraw from PoCs, eliminate operations, and update decision criteria cannot be achieved through frontline consensus alone. These are areas that management must take responsibility for. Particularly critical decisions concern talent allocation. Top talent must be allocated to challenging areas that AI cannot replace, such as withdrawal or turnaround of low-profit businesses, SCM reform, quality and regulatory compliance, overseas subsidiary management, and post-merger integration after mergers and acquisitions. Even if AI can present options, the final decision-making and responsibility for mobilizing stakeholders rest with humans. The question is not the number of top talent, but whether they are placed in critical areas with sufficient authority and responsibility.

In other words, the success or failure of AX is not a matter of technology adoption but of how thoroughly management decision-making can be executed. Exercising authority to stop initiatives, placing top talent in critical areas, and choosing enterprise-wide optimization based on structure and numbers—only when these are executed does AI become a core function for driving transformation.
In this series, AX has been organized from three perspectives: conceptualization (Part 1), financial discipline (Part 2), and execution of transformation (this article). Part 1 diagnosed “why and what should be changed” as structural bottlenecks, Part 2 organized “where to invest and how to recover” in terms of financial discipline and asset creation mechanisms, and this article discussed “how to execute and what to stop” in terms of the value realization layer and management decision-making. The central question throughout this series is not how to introduce AI, but how to redesign management on the premise of AI. Investment without a concept lacks focus, execution without financial discipline dissipates return on investment, and concept without execution sinks into PoCs. Concept, financial discipline, and execution—if even one is missing, AI will not connect to enterprise value. The essence of management capability in the AX era lies in whether all three can be integrated. This is not a challenge for AI or digital transformation departments but a management agenda that must be directly undertaken by the executive leadership.
Going forward, ABeam Consulting will continue to provide hands-on support by engaging deeply with both management and frontline operations in complex decision-making associated with AX. As a “Real Partner” that supports everything from conceptualization to business and data design, AI implementation, connection to EA and operational deployment, talent reallocation, and realization of financial outcomes in an integrated manner, we will connect companies’ AI transformation to the enhancement of enterprise value.