ROI Case File No.346 | 'Tech Innovators' Loss Through Rework'

Chapter 1: A Treasure Trove of Past Information—Yet No One Can Mine It

The day after solving GlobeTech's blueprint creation case, we received a consultation regarding development workflow efficiency. Volume 28, "The Pursuit of Reproducibility," Episode 346 tells the story of optimizing an entire business process.

"Detective, we have over 10 years of development information. Blueprints, trouble cases, design points. A vast amount of knowledge. However, we cannot leverage it. And we keep encountering the same problems, resulting in rework."

Yusuke Suzuki, Development Director from Kawasaki at Tech Innovators, visited 221B Baker Street with an exhausted expression. In his hands were a project Gantt chart and, in stark contrast, an analysis report noting "Average rework hours: 180 hours/project."

"We specialize in industrial robot development. Transport robots, assembly robots, inspection robots. About 45 development projects annually. Average development period is 6 months. However, we have a problem. We cannot utilize past information, and rework keeps occurring."

Tech Innovators' Development Structure: - Established: 2010 (Industrial robot development) - Employees: 120 (Development department: 55) - Annual projects: Approximately 45 - Average development period: 6 months - Average rework hours: 180 hours/project - Problems: Insufficient utilization of past information, RPA's lack of versatility

There was deep anxiety in Suzuki's voice.

"The development process is divided into five phases: requirements definition, basic design, detailed design, implementation, and testing. We should reference past project information at each phase, but in reality, we don't.

For example, during the detailed design phase, a concern arises: 'This motor placement might cause cooling problems.' But no one knows that the same issue occurred in the past. As a result, the problem is discovered during the testing phase after implementation, and we have to redo the design. 180 hours of rework."

Typical Rework Cases:

Case 1: Motor Placement Problem (Rework hours: 220 hours) - Phase: Detailed design - Problem: Motor placed in enclosed space - Discovery: Overheating during testing phase - Cause: Insufficient cooling - Rework: Revised design, added cooling fan - Reality: The same problem occurred in a different project 3 years ago, and the solution was documented - However: No one referenced it

Case 2: Control Software Bug (Rework hours: 180 hours) - Phase: Implementation - Problem: Software freezes with specific operation patterns - Discovery: Testing phase - Cause: Memory leak - Rework: Complete code review - Reality: A similar bug occurred 2 years ago, and the fix pattern was shared - However: The developer didn't know

Case 3: Parts Procurement Delay (Rework hours: 120 hours) - Phase: Implementation - Problem: Discovered that special parts require 3 months delivery - Discovery: During ordering - Cause: Insufficient preliminary research - Rework: Find alternative parts, adjust design - Reality: The same parts caused a delay 1 year ago, and an alternative parts list was created - However: No one accessed it

Suzuki sighed deeply.

"Furthermore, we have another problem. We've already implemented RPA. We use it to automate routine tasks. However, there's a versatility issue. When the work method changes slightly, the RPA cannot adapt.

For example, we created an RPA to 'copy blueprints from Folder A to Folder B.' But if the folder name changes, it doesn't work. We have to modify the RPA each time. We want AI agents that can respond to flexible instructions."

Chapter 2: The Promise of AI Agents—But How Should They Be Utilized?

"Mr. Suzuki, do you think implementing AI agents will solve all your problems?"

At my question, Suzuki showed confusion.

"Yes... I expect so. I hear they can search past information, replace RPA, edit presentation materials. Various things. But I don't know specifically how to utilize them."

Current Understanding (AI Omnipotence Expectation Type): - Expectation: AI agents can do various things - Problem: Can't see where in the development process they should be applied

I explained the importance of visualizing the entire business process and optimizing the value creation flow.

"The problem is that 'where in the development process are the issues' is not visible. Value chain analysis. We decompose the business process and visualize how added value is created at each step. Then we determine where placing AI agents would be most effective."

⬜️ ChatGPT | Catalyst of Conception

"Look at the process. Where value is created and where it's lost. Visualize it with the value chain."

🟧 Claude | Alchemist of Narrative

"Past knowledge is always needed at 'key points in the process.' Find those key points."

🟦 Gemini | Compass of Reason

"Value chain is a technology of optimization. Decompose tasks and design the flow of value creation."

The three members began their analysis. Gemini deployed the "Value Chain Framework" on the whiteboard.

Value Chain Structure: 1. Primary Activities: Activities that directly create value 2. Support Activities: Activities that support primary activities

"Mr. Suzuki, let's first decompose Tech Innovators' development process using the value chain."

Chapter 3: Discovery Through Decomposition—Three Processes Where Value Is Lost

Phase 1: Value Chain Analysis of Development Process (3 weeks)

Primary Activities (Development Process):

1. Requirements Definition (Average: 120 hours) - Activity: Customer needs interviews, specification documentation - Added value: Clarify customer requirements - Issue: Cannot reference similar past projects, estimation accuracy is low - Rework occurrence rate: 15%

2. Basic Design (Average: 200 hours) - Activity: Overall system design, architecture design - Added value: Create feasible design - Issue: Cannot reference past trouble cases, overlook risks - Rework occurrence rate: 25%

3. Detailed Design (Average: 280 hours) - Activity: Determine detailed specifications for each component - Added value: Create manufacturable design - Issue: Cannot reference past design points, cannot optimize - Rework occurrence rate: 35% (highest)

4. Implementation (Average: 320 hours) - Activity: Parts procurement, assembly, software development - Added value: Create actual product - Issue: Parts procurement delays, software bugs - Rework occurrence rate: 20%

5. Testing (Average: 160 hours) - Activity: Operation verification, performance evaluation - Added value: Guarantee quality - Issue: When problems are discovered, return to previous process - Rework occurrence rate: 5%

Support Activities:

1. Information Management - Activity: Store past blueprints, trouble cases, design points - Problem: Stored in dispersed locations, difficult to search - Location: Shared folders, SharePoint, individual PCs

2. Routine Tasks (RPA-automated) - Activity: Blueprint copying, data transcription - Problem: RPA cannot adapt when work methods change

3. Document Creation - Activity: Proposals, presentation materials - Problem: Time-consuming to create

Phase 2: Issue Identification (1 week)

Value chain analysis revealed that value was being lost in the following three processes.

Value Loss Point 1: Detailed Design (Rework occurrence rate 35%) - Cause: Cannot reference past design points - Impact: 45 projects/year × 35% = 15.75 projects with rework - Loss: 15.75 projects × 180 hours = 2,835 hours/year

Value Loss Point 2: Basic Design (Rework occurrence rate 25%) - Cause: Cannot reference past trouble cases - Impact: 45 projects/year × 25% = 11.25 projects with rework - Loss: 11.25 projects × 180 hours = 2,025 hours/year

Value Loss Point 3: Implementation (Parts Procurement Delay) - Cause: Cannot reference alternative parts list - Impact: 45 projects/year × 20% = 9 projects with rework - Loss: 9 projects × 120 hours = 1,080 hours/year

Total Loss: 5,940 hours/year

Phase 3: AI Agent Deployment Strategy (1 week)

Strategy: "Deploy AI agents at value loss points to automatically provide past information"

AI Agent 1: Design Support Agent (Detailed Design) - Role: Automatically search and present past design points - Example: "With this motor placement, cooling problems have occurred in the past. We recommend adding a cooling fan."

AI Agent 2: Risk Prediction Agent (Basic Design) - Role: Automatically search and present past trouble cases - Example: "With this control software pattern, memory leaks have occurred in the past. Please refer to the fix pattern."

AI Agent 3: Parts Procurement Support Agent (Implementation) - Role: Automatically check parts delivery times, propose alternative parts - Example: "This part has a 3-month delivery time. Alternative part A has a 1-month delivery and specifications also match."

Chapter 4: Optimization Through Deployment—Results After 6 Months

Phase 4: AI Agent System Construction (4 months)

System Specifications:

Foundation: GPT-4 Based Enterprise AI Agent - Target data: 10 years of past development information (blueprints, trouble cases, design points) - Integration: Links with existing information management systems (SharePoint, shared folders)

Functions:

1. Design Support Agent - Search for similar past designs in real-time during design - Warn "This design has these risks" - Present recommended design proposals

2. Risk Prediction Agent - Predict potential risks at the basic design stage - Automatically search past trouble cases - "With this architecture, XX problems have occurred in the past"

3. Parts Procurement Support Agent - When parts list is entered, automatically check delivery times - Propose alternatives for parts with long delivery times

4. General Task Agent (RPA Replacement) - Natural language instructions: "Copy blueprints from Folder A to Folder B" - Can respond even when folder names change

5. Document Creation Support Agent - Edit presentation materials, create documents

Development Period: 4 months Development Cost: 22 million yen

Phase 5: Operation Start (Month 4-10)

Operation Flow:

Detailed Design Phase: - Designer designs with CAD - Design support agent presents past information in real-time - "With this motor placement, cooling problems may occur. Would you like to check past solutions?" - Designer references and takes preventive measures

Basic Design Phase: - Risk prediction agent automatically analyzes - "With this system configuration, XX troubles have occurred in the past" - Designer takes preventive risk measures

Implementation Phase: - Parts procurement support agent automatically checks delivery times - "This part has a 3-month delivery. We propose alternative parts" - Designer adopts alternative parts

Results After 6 Months:

Rework Hours Reduction:

Detailed Design: - Before: 15.75 projects/year × 180 hours = 2,835 hours/year - After: 3.15 projects/year × 180 hours = 567 hours/year - Reduction: 2,268 hours/year (80% reduction)

Basic Design: - Before: 11.25 projects/year × 180 hours = 2,025 hours/year - After: 3.38 projects/year × 180 hours = 608 hours/year - Reduction: 1,417 hours/year (70% reduction)

Implementation: - Before: 9 projects/year × 120 hours = 1,080 hours/year - After: 1.8 projects/year × 120 hours = 216 hours/year - Reduction: 864 hours/year (80% reduction)

Total Reduction: 4,549 hours/year (77% reduction)

Development Period Reduction: - Before: Average 6 months/project - After: Average 4.8 months/project - Reduction: 1.2 months (20% reduction)

Annual Project Number Increase: - Before: 45 projects/year - After: 56 projects/year (+24%) - Reason: Development period reduction allows handling more projects with same resources

Revenue Increase: - Before: 45 projects/year × average 18 million yen = 810 million yen/year - After: 56 projects/year × average 18 million yen = 1.008 billion yen/year - Increase: 198 million yen/year (+24%)

Investment Recovery: - Initial investment: 22 million yen - Annual reduction effect (labor costs): 4,549 hours × 4,500 yen = 20.47 million yen - Annual revenue increase: 198 million yen - Total effect: 202.47 million yen - ROI: 820% (first year) - Investment recovery period: 1.3 months

Organizational Change:

Developer A's Voice: "Previously, during detailed design, I was anxious, thinking 'Is this design okay?' Problems were discovered during the testing phase, and I had to redo it about 3 times a year.

But since AI agents were introduced, they warn me in real-time during design. 'This motor placement has a risk of cooling problems.' They also present past solutions, so I can take preventive measures. Rework has decreased dramatically."

Suzuki's Voice: "Until we conducted the value chain analysis, we didn't know 'where AI agents should be used.' However, by decomposing the development process into five steps and visualizing where value was being lost, the places to deploy them became clear.

Detailed design, basic design, implementation. By deploying AI agents in these three processes, rework hours were reduced by 77%. Development period was shortened by 20%, and annual project numbers increased by 24%. Revenue increased by 198 million yen."

Chapter 5: The Detective's Diagnosis—Look at the Process, Where Value Is Created

That night, I contemplated the essence of value chain.

Tech Innovators didn't know "how AI agents should be utilized." Without viewing the whole, there was a danger of ending with partial optimization.

By decomposing the development process into five steps with value chain and visualizing where value was being lost, the places to deploy AI agents became visible. Detailed design, basic design, implementation. By concentrating on these three, they achieved 77% rework reduction and 24% revenue increase.

"Look at the process. Where value is created and where it's lost. Visualize with value chain. Then deploy AI where value is lost."

The next case will also depict a moment of optimizing the entire business process.

"Decompose primary and support activities. Discern where value is created and where it's lost. Value chain analysis leads to the optimal solution for AI agent deployment"—From the Detective's Notes

Related Files