ROI Case File No.343 | 'TechNova's Scattered Knowledge'

Chapter 1: The Loss of Dispersion—Needed Information Cannot Be Found

The day after resolving AllureSign's correction work efficiency case, a consultation arrived regarding internal knowledge management. Volume 28 "The Pursuit of Reproducibility," Story 343, is a tale of discerning information value.

"Detective, we have vast knowledge within our company. However, we cannot access that knowledge. Past troubleshooting information for similar issues should exist, but no one can find it. And we repeatedly spend time on the same problems."

TechNova's Information Systems Director, Kenji Sato from Shibuya, visited 221B Baker Street with an exhausted expression. In his hands were internal system configuration diagrams and, in stark contrast, a survey report marked "Information search time: Average 18 minutes/search."

"We develop and sell industrial machinery for manufacturing. 280 employees. Technical department 120, sales department 80, administration 80. Due to handling advanced technology, accumulated internal knowledge is vast."

TechNova's Information Management Structure: - Established: 2005 (Industrial machinery manufacturer) - Employees: 280 - Information storage locations: - Shared folders (Word, Excel, PDF) - Email (past exchanges) - Personal PCs (individual notes) - Microsoft 365 (SharePoint, Teams) - Problem: Information scattered, low searchability, cannot utilize

Sato's voice carried deep concern.

"Currently, we've implemented Microsoft 365 with several paid Copilot licenses. However, we cannot utilize them. Because information is scattered.

In the technical department, past troubleshooting information is saved in shared folders. However, file names aren't standardized, and searches don't find them. 'Motor_failure_2023.docx,' 'Error_response_memo.xlsx,' 'Trouble_cases_Tanaka.pdf.' This is the state."

Typical Problem Cases:

Case 1: Technical Department New Engineer A (1 year employed) - Situation: Machine stops at customer site. Error code "E-4503" displayed - Response: Search internally for error code → Not found - Result: Consult senior, learn handling method (2 hours downtime) - Actually: Same error occurred 5 times previously. Handling method documented, but file name was "December_2021_trouble_response.docx," couldn't search by error code

Case 2: Sales Department Staff B (5 years experience) - Situation: Customer requests "want to compare with competitor products" - Response: Search for competitive analysis materials → Not found - Result: Create materials myself (3 hours) - Actually: Same materials created six months ago, but saved in deep SharePoint hierarchy, didn't appear in search

Case 3: Administration Department HR Staff C (3 years experience) - Situation: Employee asks "childcare leave procedure" - Response: Check employment rules → Description ambiguous - Result: Search past cases but not found, confirm with labor consultant (takes 1 day) - Actually: Same question occurs 3 times monthly. Past responses buried in email

Sato sighed deeply.

"We feel the need for AI. However, we don't know specifically what to do. Knowledge sharing tools? Internal FAQ? Generative AI agents? Cannot judge which is optimal."

Chapter 2: The Misconception of AI Omnipotence—Tools Don't Solve Everything

"Mr. Sato, do you believe implementing AI technology will solve all problems?"

My question left Sato with a bewildered expression.

"Yes... we expect so. But actually, we don't know what can be done. Received vendor proposals, but cannot judge if they fit our challenges."

Current Understanding (AI Implementation Hopeful Type): - Expectation: AI tool implementation will solve everything - Problem: Cannot see which information should be prioritized for organization

I explained the importance of quantitatively evaluating information value and establishing priorities.

"The problem is not knowing 'which information has the most value.' RFM analysis—Recency, Frequency, Monetary. Customer analysis method, but this time we'll apply it to internal information. Recency, usage frequency, business contribution. Evaluate information with these three and identify what should be prioritized for organization."

⬜️ ChatGPT | Catalyst of Conception

"Don't organize everything. Discern value. Prioritize with RFM and organize high-value information first"

🟧 Claude | Alchemist of Narrative

"Past troubleshooting always conceals 'future solutions.' Measure that value"

🟦 Gemini | Compass of Reason

"RFM is value evaluation technology. Score information by recency, frequency, contribution and maximize investment effectiveness"

The three members began analysis. Gemini deployed "RFM Framework" on the whiteboard.

RFM's 3 Elements (Information Version): 1. Recency: When was information created/updated 2. Frequency: How often is information accessed 3. Monetary (Contribution): How much does information contribute to business

"Mr. Sato, let's first evaluate internal information with RFM."

Chapter 3: Discovery Through Evaluation—High-Value Information is Only 12% of Total

Phase 1: Information Inventory and RFM Analysis (4 weeks)

Surveyed all internal information.

Survey Targets: - Shared folders: Approximately 18,000 files - SharePoint: Approximately 5,000 files - Email: Outside survey scope (too vast, established new rule to transfer important emails to shared folders) - Total: Approximately 23,000 files

RFM Score Calculation Method:

Recency: 1-5 points - 5 points: Created/updated within past 3 months - 4 points: Within past 6 months - 3 points: Within past 1 year - 2 points: Within past 2 years - 1 point: Over 2 years ago

Frequency: 1-5 points - Aggregated past 6 months' access count from access logs - 5 points: 10+ times monthly - 4 points: 5-9 times monthly - 3 points: 1-4 times monthly - 2 points: 1-5 times in 6 months - 1 point: 0 times in 6 months

Monetary (Contribution): 1-5 points - Evaluated by information type - 5 points: Troubleshooting information (prevents business stoppage) - 4 points: Sales materials (directly linked to revenue) - 3 points: Technical specifications (essential for product development) - 2 points: Internal procedure FAQ (contributes to efficiency) - 1 point: Other (archive materials, etc.)

RFM Total Score: 3-15 points

Phase 2: Analysis Results (4 weeks)

High-Value Information (RFM Score 12-15 points): 2,760 files (12%) - Recency: 4-5 points (recently created/updated) - Frequency: 4-5 points (frequently accessed) - Monetary: 4-5 points (large business contribution)

Breakdown: - Troubleshooting information: 680 files - Sales materials (proposals, competitive analysis): 520 files - Technical specifications (latest versions): 880 files - Internal FAQ: 680 files

Strategy: Top priority for organization and AI agent support

Medium-Value Information (RFM Score 8-11 points): 6,900 files (30%) - Occasionally accessed but not urgent

Breakdown: - Old version technical specifications - Past sales materials - Training materials

Strategy: Gradual organization, searchability improvement

Low-Value Information (RFM Score 3-7 points): 13,340 files (58%) - Rarely accessed - Old information

Strategy: Archive, exclude from AI agent scope

Phase 3: High-Value Information Organization (6 weeks)

Organization Content:

1. File Name Standardization (680 troubleshooting information files) - Before: "December_2021_trouble_response.docx" - After: "[E-4503]Motor_failure_cooling_fan_abnormal_2021-12.docx" - Rule: [Error code]Equipment_name_Cause_Date

2. Metadata Addition - Tags: "Error code," "Equipment name," "Response time," "Person in charge" - Keywords: Extract important terms from text

3. Consolidation to SharePoint - Migrate high-value information from shared folders to SharePoint - Redesign folder structure: By category, by department

Chapter 4: Evolution Through AI Agents—Results After 6 Months

Phase 4: Generative AI Agent Implementation (3 months)

System Specifications: - Base AI: GPT-4-based enterprise AI agent - Target data: 2,760 high-value information files - Functions: - Natural language search ("E-4503 error occurred. What's the handling method?") - Similar case presentation ("3 past cases responded to same error") - Auto-summarization (Summarize long technical specifications in 3 minutes) - Internal FAQ auto-generation (Auto-detect frequent questions, generate answers)

Implementation Cost: - Initial deployment: 4.8 million yen - Monthly operation: 500,000 yen

Phase 5: Operations Launch (Month 3-9)

Information Search Time Reduction:

Before: - Search time: Average 18 minutes/search - Search count: 280 people × monthly average 5 searches = monthly 1,400 searches - Total: 1,400 searches × 18 minutes = 25,200 minutes (420 hours/month)

After: - Search time: Average 2 minutes/search (AI agent responds immediately) - Search count: Monthly 1,400 searches - Total: 1,400 searches × 2 minutes = 2,800 minutes (46.7 hours/month)

Reduction: 373.3 hours/month (89% reduction)

Troubleshooting Time Reduction:

Case 1 Reverification: Error Code "E-4503" - Before: Consult senior, 2 hours downtime - After: Ask AI agent, immediately retrieve handling method, restore in 15 minutes - Reduction: 1 hour 45 minutes/case

Similar error occurrence frequency: 12 monthly cases - Reduction: 12 cases × 1.75 hours = 21 hours/month

Sales Material Creation Time Reduction:

Case 2 Reverification: Competitive Analysis Materials - Before: Create myself, 3 hours - After: AI agent presents past materials, update in 30 minutes - Reduction: 2.5 hours/case

Similar request frequency: 8 monthly cases - Reduction: 8 cases × 2.5 hours = 20 hours/month

Internal FAQ Response Time Reduction:

Case 3 Reverification: Childcare Leave Procedures - Before: Search past cases, confirm with labor consultant, 1 day - After: AI agent responds immediately, 5 minutes - Reduction: 7 hours 55 minutes/case

Similar question frequency: 90 monthly cases (childcare leave, various procedures, etc.) - Reduction: 90 cases × 7.92 hours = 712.8 hours/month

Results After 6 Months:

Work Time Reduction: - Information search: 373.3 hours/month - Troubleshooting: 21 hours/month - Sales material creation: 20 hours/month - Internal FAQ response: 712.8 hours/month - Total: 1,127.1 hours/month

Annual Reduction Time: - 1,127.1 hours/month × 12 months = 13,525.2 hours/year

Monetary Effect: - Average hourly wage: 3,500 yen (all employees average) - 13,525.2 hours × 3,500 yen = 47.38 million yen/year

Investment Recovery: - Initial investment: 4.8 million yen + 500,000 yen × 6 months = 7.8 million yen - Reduction effect: 47.38 million yen ÷ 12 months × 6 months = 23.69 million yen - ROI: 204% (6 months) - Investment recovery period: 2.0 months

Employee Satisfaction Improvement: - Before: Stress from information searches (NPS 45) - After: Immediate resolution with AI agent (NPS 72) - Improvement: +27 points

Organizational Changes:

Technical Department Engineer A's Voice: "Previously, when errors occurred, I'd ask seniors or search shared folders for tens of minutes. But when I ask the AI agent 'E-4503 error occurred. What's the handling method?', past response cases immediately appear. Can restore in 15 minutes now."

Sales Staff B's Voice: "Creating competitive analysis materials previously took 3 hours. But when I tell the AI agent 'want to create comparison materials with Company A,' past materials appear. Just update them in 30 minutes. Customer proposal speed increased."

HR Staff C's Voice: "Employment rule inquiries come almost daily. Previously, searching past responses took time. But the AI agent auto-generated internal FAQ. When asked 'childcare leave procedures?', just guide them to 'search with AI agent.'"

Sato's Reflection:

"Until conducting RFM analysis, we thought 'we must organize all information.' Organizing all 23,000 files is impossible.

However, evaluating with RFM for recency, frequency, contribution revealed priorities. High-value information was only 12% of total (2,760 files). First, we organized this 12%.

By implementing the generative AI agent, information search time reduced 89%. Troubleshooting, sales material creation, internal FAQ response times also drastically shortened. Annual 47.38 million yen reduction effect.

Past troubleshooting information is no longer 'buried treasure.' The AI agent retrieves it anytime."

Chapter 5: Detective's Diagnosis—Focus on the 12% of Value

That evening, I contemplated information value evaluation.

TechNova was trapped in the illusion that "we must organize all information." However, organizing all 23,000 files was inefficient.

Evaluating with RFM for recency, frequency, contribution revealed high-value information was only 12% of total. Focusing on this 12% achieved annual 47.38 million yen reduction effect.

"Don't organize everything. Discern value. Evaluate with RFM for recency, frequency, contribution and focus on high-value information. 12% surpasses the remaining 88%."

The next case will also depict the moment of discerning information value.

"Recency, Frequency, Monetary. Evaluate information by recency, frequency, contribution. Don't organize everything. Focus on the 12% of value. Future solutions exist there"—From the Detective's Notes

Related Files