[Interview] A&M Partner Kim Myeong-gu: AX Results Must Be Measured by P&L, Not Time Savings

In an interview, Kim Myeong-gu, a partner at A&M, explained the firm’s corporate AX strategy. The photo was credited to reporter Yang Seung-gap. A&M said it is expanding its business scope into the AI transformation (AX) market, building on its existing strengths in corporate restructuring and turnaround management.

Kim said A&M’s strategic direction is to combine management innovation and an understanding of industrial sites with AI and data technologies. He said the goal is to create tangible management results for companies. A&M said it is also expanding its business areas into management innovation, strategy, digital, and tax.

Kim said A&M’s strength lies in execution-focused consulting that goes beyond strategy formulation. He explained that the firm improves cost structures, organizations, and work processes by directly participating in corporate operations. A&M said it operates a One Firm model in which about 14,000 people around the world work together, and that the number of employees in Korea is about 100.

In an interview with this paper, Kim Myeong-gu, a partner at A&M, pointed to management innovation as one of the firm’s traditional strengths. He referred to people’s experience and insights as the basis for problem-solving in the past, and described AI adoption as a way to advance services in the present.

Kim opposed limiting the criteria for evaluating AX results to the number of AI deployments and the reduction rate in work hours. While he said it is meaningful in the early stage of generative AI to confirm the technology’s potential uses, he also said that at this stage companies need proof of changes AI has made to actual profits and losses, including revenue, costs, and productivity.

Kim said reduced work hours are only part of the process. He cited improvements in management indicators such as revenue, productivity, and defect rates as the true measures of success, and said that measuring only AI model accuracy and work-hour reduction has limits when it comes to explaining the value a company has actually gained.

Kim said companies need to move beyond repeated proof-of-concept (PoC) stages in their AX efforts. He proposed improving the profit and loss statement (P&L) as the next stage for corporate AX, saying AX should shift from intermediate indicators such as the number of deployments or time savings to a stage that demonstrates management outcomes linked to profit and loss.

It was emphasized that AX should not be seen as a simple extension of existing digital transformation (DX). DX is characterized by digitizing and automating individual tasks previously performed by people. By contrast, AX is more akin to redesigning multiple work processes and employee roles around AI agents.

The subtitle is a change in AX evaluation criteria, moving beyond PoC toward P&L improvement. Partner Kim assessed domestic companies’ AI investment not as overheating, but as evidence that actual investment and companywide change are still not sufficient.

Kim said it is inappropriate to equate the AI boom and rising stock prices of related companies with excessive AI investment at corporate sites. He said there is much discussion about the need for AI.

Kim said he has rarely seen domestic companies suffer management problems because of excessive AI investment costs. He added, however, that there may be companies that misunderstood how to proceed. He said he doubts whether domestic companies’ AX investment is realistically excessive.

Caution is needed when judging the usefulness of AI based only on early PoC results for generative AI, since the performance and scope of use for generative AI have improved rapidly. Accordingly, companies that failed to achieve the results they expected from PoCs conducted several years ago need to review them again.

At the same time, companies should avoid staying focused only on repeatedly adding new PoCs. Now is the time to move into the stage of applying AI to actual work. They also need to connect AI to management performance.

Companies generally cite time savings and improved model accuracy as the effects of AI adoption. But if labor is kept unchanged after time is saved and no additional revenue or cost reduction occurs, AI adoption may actually increase costs from the company’s perspective.

Kim said companies should not stop at reducing a 10-hour sales task to 3 hours. The saved time needs to be used for customer targeting and higher-value work. He also said companies need to see how much revenue rises from an existing KRW 1 billion level.

In manufacturing, AI application criteria focus more on actual reductions in defect rates, downtime, productivity gains, and yield improvements than on AI model accuracy. Qualitative effects are also needed, but ultimately they must be linked to corporate management indicators.

In a recent interview with this paper, Kim Myeong-gu, a partner at A&M, said the reason A&M is pushing AX is that it judges outcomes primarily on management performance achieved through AI, rather than AI itself.

Kim said that in traditional industries such as manufacturing, construction, heavy industry, engineering, procurement, and construction (EPC), the ability to understand the business and the work site is just as important as AI technology.

He then pointed to tacit knowledge as the starting point for manufacturing AX. Tacit knowledge refers to knowledge and know-how acquired through long experience but difficult to express clearly in documents or numbers.

Examples of tacit knowledge in manufacturing sites include sensing subtle sounds, vibrations, and raw material changes in equipment, detecting abnormal signs, and expert judgment about the best corrective action. The photo was credited to reporter Yang Seung-gap.

One of the areas A&M has recently been focusing on is traditional industries such as manufacturing, construction, heavy industry, engineering, procurement, and construction (EPC). The early AI market centered on software companies and office-work efficiency. Recently, AX demand has been expanding into “heavy industries,” where field experience and expertise matter.

Kim stressed that in these industries, understanding the business and the field is as important as AI technology. The premise for introducing AI agents is that they perform a specific task from start to finish. To do this, companies need both personnel who understand the technology and field experts who know what needs to change and what results must be produced.

A&M’s manufacturing AX service takes a different approach from existing smart factories. Traditional smart factories focused on digitizing equipment and processes. By contrast, manufacturing AX focuses on turning skilled workers’ on-site “judgment” into digital assets.

It has been pointed out that skilled workers’ judgment processes face limits in transmission and preservation because there is not enough accumulated documentation and data. Experience-based judgment is difficult to fully pass on to other employees, and if skilled workers retire, companies may lose the capabilities that have solved long-term accumulation problems. He said there are cases where skilled workers do not want to share their know-how, but more often they simply do not know how to record and datafy the judgment process.

He said experience left on paper and experience kept only in the head need to be structured. If skilled workers’ tacit knowledge is reflected in AI, the work level of less experienced employees can improve, and existing skilled workers can also use cases from areas they have not personally experienced. In this way, experience tied to individuals can be transformed into a shared organizational asset.

To do this, interviewing frontline workers alone is not enough, he said. It is necessary to structure information about skilled workers’ judgment criteria and response criteria through observation of actual field work, and it is also necessary to convert structured field knowledge into a format that AI can process.

A&M is running projects in this way. It deploys business consultants and data scientists at the same time, carrying out field knowledge structuring and AI model application in parallel.

Kim compared the process of datafying tacit knowledge to turning a master chef’s touch into a recipe. He said a master chef checks ingredient conditions through experience and adjusts ingredient quantities and cooking times, but may have limits in quantitatively explaining the basis for those judgments. Accordingly, consultants and data specialists take on the role of exploring the variables that affect the artisan’s judgment and structuring the related process.

He also noted that datafying tacit knowledge is not something that ends after one round. As equipment, raw materials, and production environments change, judgment criteria need to be continuously updated. He added that AI-generated results need to be reflected in the field, and in that process skilled workers must judge the appropriateness of AI results and feed that judgment back into the model.

For this, Kim proposed a system called Human in the Loop. He said internalizing tacit knowledge in AI does not mean skilled workers are no longer needed. Rather, he explained, the role of skilled workers shifts from directly performing repetitive tasks to evaluating AI results and making decisions.

Kim described the difference between generative AI and AI agents based on the scope of work they perform. Generative AI has the character of a copilot or assistant that responds to human questions and supports work, whereas AI agents are characterized by carrying out a company’s specific work modules from start to finish.

According to this distinction, AX in an AI agent adoption scenario does not end at the level of simply adding one AI function to an existing process. The reason given is that AI carries out work across multiple departments and systems.

Accordingly, AX requires redesigning work flows, redesigning authority, and redesigning roles and responsibilities (R&R) among members of the organization. Unlike the auxiliary functions of generative AI, AI agents carry out specific tasks from start to finish, so AX is not simply about adding functions but about moving into a redesign phase.

Kim said DX work redesign was centered on automating tasks previously done by people. By contrast, AX requires a reexamination of the work itself and an examination of the possibility of integrating digitized individual processes through AI, as well as the possibility of transcending existing departmental boundaries.

When deciding what scope of work to assign to AI, companies need to consider the impact of errors and the possibility of recovery. Tasks that are easy to roll back and have little business impact can be given greater AI autonomy. By contrast, tasks where misjudgment has a large impact, such as manufacturing safety or major financial losses, require stronger human involvement.

In manufacturing in particular, reliability and whether the model produces stable results under the same conditions are valued more than simple model accuracy. If an AI judgment error could lead to production stoppages or safety incidents, it is necessary to keep review and approval procedures by skilled workers.

Kim Myeong-gu, a partner at A&M, stressed that the market is moving toward a situation in which people with AI skills, rather than AI itself, replace those without them. The photo was credited to reporter Yang Seung-gap.

As AI agents spread, it is expected that the intermediate stages of work handoff and consolidation will shrink within corporate organizations. He said this reduction in middle steps applies not only to middle managers’ roles but to the entire work process.

In particular, the stages of reporting, handoff, and simple document preparation between customer requirements and final deliverables are expected to be replaced by AI. He cited the example of a process in which, in the past, a department head gave instructions, a section manager relayed them, and a new employee carried out the research.

But managers with extensive work experience can use AI to quickly obtain the necessary materials and analysis results themselves. As a result, roles that simply receive instructions, prepare materials, and pass them on are expected to shrink significantly.

On the other hand, the importance of personnel with extensive industry and work experience may rise. That is because judging whether AI-generated results fit reality and correcting the direction of errors requires long-accumulated experience and contextual understanding. In this context, the point is that people who use AI well will replace people who cannot use AI.

Kim said the value of experience and know-how increases in the AI era. He explained that people who have both the ability to use AI and extensive experience will draw attention. He added that this does not refer to experienced people who cannot use AI.

Kim said the structure is not one in which AI itself replaces people. Instead, he forecast a market in which people skilled at using AI replace those who do not use AI.

The key condition for AX success was presented as decisive action by top management.

Companies’ common concerns about advancing AX were summed up in two questions. One is whether AX is possible when DX is not yet complete, and the other is whether outside AI experts can understand the company’s industry.

On this point, Kim said DX completion is not a prerequisite for AX. He explained that the more sufficiently digitized the data is, the more favorable it is for AX efforts. He also said that full completion of DX is not a necessary condition for starting AX.

As a solution to the issue of industry expertise, he proposed collaboration by a single team of frontline workers and AI experts. In past ERP implementation projects, IT staff handled system development while frontline workers conveyed requirements, with roles separated. By contrast, AX assumes joint participation by frontline experts, business consultants, and AI and data specialists.

In AX work methods, results must be continually checked and revised, and this requires the task of redesigning work together. In this process, a bottom-up approach helps collect field ideas and identify use cases. But a bottom-up approach alone makes it difficult to drive organizational and work changes or decisions on investment and staffing.

For this kind of collaborative structure to work in practice, strong leadership from top management is needed. Kim said AX will not work if it is left only to a specific executive or an outside hire because it is fashionable. He added that the owner or CEO must personally set the direction and lead decision-making and implementation.

He pointed out that in AX efforts, the determining factor is not where a person is placed, but the status and real authority given to that person. He said the rank and authority of the executive in charge of AI are important, and that in traditional industries it is not enough simply to hire a young AI expert as an executive. If that person does not have the authority to adjust existing business units or change how work is done, it is hard to overcome organizational resistance. Accordingly, he said the AI lead needs a status close to that of top management, along with authority to lead companywide work redesign and investment decisions.

He also pointed out that it is risky to start AX only after another company has produced a success story. That is because most companies are still in the process of pushing AX forward, making it difficult to find a completed success case in the same industry. He explained that if companies follow only after a case has been fully verified, they are likely to fall behind in competition.

Kim said what matters is not whether a company adopts AI first, but who actually creates management results with AI. He said the stage is now beyond exploring ideas and into a phase where work redesign, investment, and staffing must be tied to results.

Source: IT DAILY · Lee Jae-young
Original: https://www.itdaily.kr/news/articleView.html?idxno=241521

References

• https://www.itdaily.kr/news/articleView.html?idxno=241521