Why Organizations Without Principles Become Overrun with Exceptions

The Moment a Trade Show Becomes an “Implementation Venue”

The Japan DX Week Spring 2026 “9th AI & Business Automation Exhibition” will be held. This news itself is merely an event announcement. However, the evolution of a recurring exhibition serves as a mirror reflecting market maturity. If the first event was a place to explain “What is AI?”, the ninth has transformed into a venue to demonstrate “How to use it.” Exhibitor Virtualex is a company providing business automation solutions utilizing AI. The very raison d’être of the exhibition is shifting from introducing technology to consulting on implementation.

This change indicates that AI adoption is transitioning from an “experimental phase” to a “practical phase.” The purpose for executives and CTOs visiting the exhibition is no longer just gathering information on the latest technology. They come seeking judgment criteria: for their specific challenges, which vendor’s solution can be introduced, at what cost, and over what timeline. The exhibition is strengthening its function as a “full-scale showroom” for comparison and evaluation.

The Reality Highlighted by the “90% Revision” Survey Findings

Meanwhile, a reality check survey by Thanks Lab Career conveys a harsh reality. It reports that 90% of tasks utilizing generative AI involve post-output revision work. Moreover, the “accumulation of waiting time” associated with these revisions is emerging as a new challenge. The process of humans checking, revising, and approving documents or code generated by AI often takes unexpectedly long, leading to inefficiency in many cases.

This “90% revision” wall is a phenomenon that can occur as a result of many companies simply implementing generic SaaS-based AI tools. AI not optimized for a company’s specific workflows and knowledge does produce “somewhat usable” output. However, the task of polishing that output to a level usable in practical work ultimately falls to humans. This becomes a project bottleneck in the form of “waiting time.” The survey results vividly illustrate the limits of SaaS dependency.

Learning from Municipal Case Studies: Designing with “On-the-Ground Voices” Incorporated

An instructive insight here comes from the generative AI adoption case in a government agency reported by abeam.com. What they emphasize is building a “safe and convenient AI environment that reflects on-the-ground voices.” Administrative work involves unique rules, formats, and review criteria. A generic AI that ignores these is almost useless.

Their approach begins by identifying the repetitive routine tasks and the manuals/ordinances staff frequently reference. Then, they train AI models on this data to build models specialized for administrative work. Alternatively, they develop “wrappers” that automatically correct and reformat the output of generic AI based on their own rule sets. This “specialization” is the key to dramatically reducing the proportion of revision work and achieving true efficiency.

Three Concrete Actions Executives Should Take Now

The evolution of exhibitions, the reality of survey results, and lessons from pioneering cases. Synthesizing these reveals the actions executives and CTOs must take right now: a phased departure from SaaS dependency and a careful, sure shift toward in-house development. Below are three concrete steps.

1. Diagnosing “AI Suitability” of Tasks and Prioritizing

First, classify all company tasks by their “suitability for automation/enhancement by AI.” In my consulting, I sometimes use the following four-quadrant matrix.

• High Frequency & Routine (Top Priority): Daily/weekly report creation, email response template generation, simple data aggregation. Combining Claude or ChatGPT API with Google Apps Script or Zapier can automate these for a monthly cost of tens of thousands of yen (a few hundred USD).
• High Frequency & Non-Routine (Requires Customization): Sales proposal creation, contract review, customer support. Requires developing a “wrapper” to convert generic AI output to company-specific formats. Initial investment of ~$3,150 to $12,600 (¥0.5-2M), with monthly operational costs of ~$315 to $945 (¥50k-150k) as a guideline.
• Low Frequency & Routine (High Efficiency Impact): Assistance with closing operations, support for personnel evaluations. Effective integration of RPA tools and AI. Implementation requires significant time for task analysis.
• Low Frequency & Non-Routine (Maintain Status Quo): Advanced strategy formulation, creative problem-solving. Currently, keep these human-led, with AI limited to assisting in information gathering and organization.

Without this classification, it is impossible to allocate budget and resources effectively.

2. Establish “Criteria for In-House Development Decisions”

Not everything needs to be developed in-house. Decision criteria are needed. I advise that tasks meeting two or more of the following three conditions should be seriously considered for in-house development (or building specialized models).

1. Directly linked to core business competencies: Involves unique processes or knowledge that creates the company’s competitive advantage.
2. Revision costs become enormous with SaaS: The “90% revision” from the survey occurs, and waiting time becomes a drag on business speed.
3. Medium-to-long-term use is expected, leading to high cumulative costs: Using a SaaS costing ~$630 (¥100k) per month for 3 years totals ~$22,680 (¥3.6M). This could exceed the initial investment for in-house development.

For example, the case of SweetLeap Corporation (Mapion article), which provides product development support for companies in the Northern Osaka area, can be interpreted as a move to realize the “core” aspect of region/industry-specific support not with generic tools, but with AI customized in-house.

3. Building a “Co-Creation Model” Involving Diverse Talent

The case of AKKODiS Business Support is highly instructive. They are building an “AI utilization model created together with employees with disabilities.” This is not merely a diversity initiative. It embodies a powerful competitive principle in the AI era.

AI design and prompt development require diverse perspectives and deep “insight” into the work. Veteran employees long engaged in routine tasks, digitally native younger staff, and employees with disabilities who have often been placed outside the “digitalization” of work. The unique sensibilities and needs each holds are themselves valuable requirements for elevating AI from a “usable tool” to a “revolutionary support means.”

What executives should do is create a mechanism to collect such diverse “on-the-ground voices” and reflect them in AI development requirements. This could be through internal workshops or collaboration with partner companies. What’s important is the perspective of transforming the AI design process itself into something open and co-creative.

The Transition Period: From “Making Deals at Exhibitions” to “Implementing In-House”

The 2026 AI & Business Automation Exhibition will likely be an event announcing the market’s entry into a maturity phase. The solutions exhibited will likely be more modularized, with integration (via API) with in-house systems as a premise. Vendors are shifting from selling “magic tools” to providing “building materials to strengthen your castle (in-house systems).”

The role of executives and CTOs is to scrutinize those materials, select those suitable for their company’s castle, and draw up the blueprint for assembly. At the core of that blueprint must be a deep understanding of the company’s core tasks and the voices of the diverse talent supporting them.

The “90% revision” reality shown by the survey is the pitfall awaiting at the end of the “easy path” of directly implementing another company’s solution. On the other hand, cases from government agencies and pioneering companies teach us that at the end of the “difficult path” of customizing and specializing with one’s own hands lies sustainable efficiency and differentiation. From now on, the exhibition will be a venue that shows the first step on that “difficult path,” along with the concrete tools for it.

Now is the time to re-examine your company’s tasks and begin drawing a strategic blueprint for what, to what extent, and how to “AI-ize.” Only with that blueprint in hand will the deals struck on the exhibition floor become not mere purchases, but solid investments in the future.