Factory Digitalization Ver11. “Why AI Prescriptions Fail: The Importance of Operational Continuity in Batch Production”

shigenoritanaka3
5月1日
読了時間: 3分

　　　　　　　　　　　　　　　　　　　　　　　　　　　　　May 01, 2026

Hello everyone.

Today, I would like to discuss a practical issue often observed in manufacturing sites: AI prescriptions do not always lead to the expected improvement in quality.

While factories are expected to move toward autonomous, unmanned operation powered by AI in the future, it will still take time before such a state becomes reality.

For the foreseeable future, we will continue to rely on a hybrid approach in which:

AI provides the prescription, and humans adjust the parameters during production.

AI-recommended optimal conditions are not “set and forget.” They assume continuous monitoring and adjustment from the start to the end of each batch.

However, in many factories, this operational premise is not consistently maintained.

🟦 Two Typical Patterns Observed on the Shop Floor

① Ignoring the AI prescription altogether (worst case)

Some operators prioritize past successful patterns or equipment habits and choose not to adopt the AI-recommended settings. In such cases, it is natural that no improvement occurs. This is not “AI not working”—it is simply “AI not being used.”

② Following the prescription only at the beginning, then stopping adjustments (very common)

This pattern is extremely common in batch production environments:

Parameters are set according to the prescription at the start
But managers and operators are pulled away by phone calls, urgent requests, or other line support
No one monitors the parameters during the mid-to-late stages of the batch
As a result, defects increase toward the end of the batch

This is not a limitation of AI. It is an issue of operational continuity.

Unless individual products carry unique IDs, it is impossible to pinpoint exactly which part of the batch produced the defects. However, when the root cause is investigated afterward, it often becomes clear that:

The actual operating conditions gradually drifted away from the prescription as the batch progressed.

🟦 Why “strict adherence for every batch” is unrealistic

Given the realities of factory operations, it is not feasible to fully adhere to AI prescriptions for every single batch.

Managers and operators often handle multiple lines, respond to unexpected issues, and cannot continuously monitor deviations between prescribed and actual values throughout the entire batch.

However, the opposite perspective is also true:

If we narrow the focus to specific products, strict adherence becomes entirely achievable.

🟦 Which products should be prioritized?

As discussed in a previous article (Factory Digitalization Ver08. “Pareto Chart × Defect Reduction”— Maximizing Improvement Impact Through Focus and Prioritization), defects and cost impact are concentrated in a limited number of products:

The 20% of products that generate 80% of defects (high volume × high defect rate)
High-cost-impact items (e.g., heavy-weight products)

These “priority items” are where prescription adherence yields the greatest benefit.

🟦 “Focus and Concentration” on priority items is the most rational approach

Rather than attempting to enforce prescription adherence across all batches, a more realistic and effective approach is:

Focus on high‑impact priority items, and ensure consistent prescription adherence from start to finish.

This approach:

Minimizes operational burden
Maximizes defect reduction and cost improvement
Delivers the highest ROI for AI implementation

In short, it is the most rational way to convert AI prescriptions into measurable results.

🟦 Summary

AI is progressing toward autonomous operation, but full unmanned production will take time
For now, we need a two-step model: AI prescribes → humans adjust
Two patterns commonly occur: ignoring the prescription / stopping adjustments mid-batch
These are not AI limitations —they are issues of operational continuity
Full adherence for all batches is unrealistic
But adherence is achievable —and highly effective —for priority items
As discussed in the previous article: