The AI Triage Center

Lean Six Sigma + AI Routing Workshop
Budget Used
$0.00
Accuracy
--
Tickets
0/15
Score
0
01 Current State
02 Identify Waste
03 Route Tickets
04 Results
05 Debrief

Scenario: MegaMart Customer Service

MegaMart, a major e-commerce retailer, handles 10,000+ customer service requests per day. Currently, every single request — from "Where's my order?" to complex complaint resolution — gets routed to the same expensive GenAI model costing $0.10 per request.

Your consulting team has been hired to analyze this process, identify waste, and design a smarter AI routing system that improves speed and quality while dramatically cutting costs.

Step 1: Review the current-state SIPOC diagram below. Notice how every request flows through a single, undifferentiated pipeline — the same bottleneck regardless of complexity.

Current-State SIPOC Diagram
Suppliers → Inputs → Process → Outputs → Customers
Suppliers
Customers (via web, app, email, phone)
Order Management System
Product Catalog Database
CRM System
Inputs
Customer inquiry text
Order history & tracking data
Account information
Product specs & policies
Process
1. ALL requests enter single queue
2. ALL routed to GenAI ($0.10/each)
3. GenAI generates response
4. Response sent to customer
Outputs
Customer response / resolution
Updated order/account records
Service interaction log
Customer satisfaction signal
Customers
End consumers
Internal CS managers
Finance (cost tracking)
Product team (feedback loop)
⚠ Current State Metrics (Daily Average)
Cost per Request: $0.10
Daily Cost (10K req): $1,000
Avg Response Time: 8.2 sec
Error Rate: 12%
CSAT Score: 72/100
GenAI Hallucination Rate: 8%

Discussion: Where do you see waste? Why is this process expensive, slow, and error-prone?

Identify the 8 Wastes (DOWNTIME)

Using the Lean DOWNTIME framework, click each waste category below to reveal how it manifests in MegaMart's current "route everything to GenAI" process. Your goal: identify all 8 wastes before designing the fix.

0/8 wastes identified

Design Your AI Router

Now route each incoming customer request to the optimal AI lane. Match the complexity of the tool to the complexity of the task — the same principle as right-sizing in Lean.

Budget constraint: $0.45 for 15 requests. Quality target: ≥80% correct routing. Route wisely!

Budget Remaining
$0.00 spent$0.45 remaining

Route to Lane

Performance Scorecard

Your routing decisions have been scored. Compare your optimized process against MegaMart's original approach.

Routing Breakdown

Framework Connections

The AI router you just designed isn't a new concept — it's a direct application of Lean, Six Sigma, and Theory of Constraints principles. Here's how each connects.

🏭 Lean Connection

You eliminated waste by right-sizing the tool to the task — the same principle as not using a CNC machine to hammer a nail. Simple lookups go to cheap, fast rules engines (eliminating overprocessing). Parallel lanes eliminate the single-queue bottleneck (reducing waiting). Each request now flows through the minimum viable process (one-piece flow thinking).

📊 Six Sigma Connection

You just executed a DMAIC cycle. Define: Excessive cost and poor accuracy. Measure: $0.10/request, 12% error rate. Analyze: Root cause is undifferentiated routing. Improve: Three-lane classification system. Control: The router's decision rules are your control plan. You also reduced variation — GenAI hallucinating on simple lookup tasks was assignable cause variation that your routing rules eliminated.

🔗 Theory of Constraints

The original system had one bottleneck: the single GenAI queue processing all request types at the same speed and cost. You broke the constraint by creating parallel lanes with different capacities and costs — exactly what Goldratt would prescribe. The constraint was "exploited" (only the requests that truly need GenAI go there) and "elevated" (simple requests bypass it entirely).

💰 Cost of Quality (COPQ)

Two types of misrouting cost: Overprocessing cost — sending a simple lookup to GenAI wastes $0.099 per request. Under-processing cost — sending a nuanced complaint to a rules engine produces a robotic, inadequate response that damages customer satisfaction and may lead to escalation. Your router minimizes both.

🚀 Real-World Application

This is exactly what companies like Anthropic, OpenAI, Google, and Amazon are building right now — AI orchestration layers that decide which model (or whether AI at all) should handle each request. If you can walk into an interview and say, "I designed an AI routing system using process mapping and cost-of-quality analysis to optimize which AI model handles which request type," that's a memorable, differentiated answer.