The ROI Equation

Every leader has 50 ideas for AI. "Let's summarize emails!" "Let's write code!" "Let's predict the stock market!" Your job is to say "No" to 47 of them.

Implementing GenAI is expensive.

• Token Costs: You pay for every question and answer.
• Talent Costs: AI engineers are expensive.
• Risk Costs: What if the AI hallucinates a promise to a customer?

In this lesson, we define a framework for Prioritization.

---

1. The Impact / Effort Matrix

Sort your use cases onto a 2x2 grid.

graph TD
    subgraph "High Impact"
    A[Transformative Bets]
    B[Low Hanging Fruit (Do This First)]
    end
    
    subgraph "Low Impact"
    C[Money Pits]
    D[Incremental Improvements]
    end
    
    style B fill:#34A853,stroke:#fff,stroke-width:2px,color:#fff
    style C fill:#EA4335,stroke:#fff,stroke-width:2px,color:#fff

Quadrant 1: Low Effort, High Impact (Low Hanging Fruit)

The Sweet Spot. These utilize "off the shelf" models for common tasks.

• Example: "Summarize Customer Support Tickets."
• Why: Use Gemini Flash (cheap). No fine-tuning needed. Data is already text. Huge time savings.

Quadrant 2: High Effort, High Impact (Transformative Bets)

The Strategic Differentiators. These require custom data, fine-tuning, and complex agents.

• Example: "Generative Design for Car Parts."
• Why: Requires proprietary physics data and 3D models. Hard to build, but gives you a massive competitive edge.

Quadrant 3: High Effort, Low Impact (Money Pits)

The Trap. Avoid these.

• Example: "AI to write birthday emails to employees."
• Why: It takes engineering time to connect to HR systems, but the business value is close to zero.

---

2. Buy vs. Build (Revisited for Strategy)

In Module 2, we looked at the technical layers. Now let's look at the financial decision.

The "Buy" Case (SaaS)

• Use: Gemini for Workspace, Salesforce Einstein, Microsoft Copilot.
• Cost: Licensed per user ($20-$30/month).
• ROI: Immediate but capped. You get the same advantage as your competitors who also buy it.

The "Build" Case (Vertex AI)

• Use: Custom Agent for your innovative product.
• Cost: Usage-based (Tokens) + Development (Salary).
• ROI: Slow to start, but potentially uncapped if it fundamentally changes your business model.

Rule of Thumb: Buy for Internal Productivity (Back Office). Build for Customer Experience (Front Office).

---

3. Calculating Feasibility

Before greenlighting a project, ask the Technical Feasibility Questions:

1. Data Availability: "Do we strictly possess the data needed to answer these questions?" (If the data is in people's heads, AI fails).
2. Latency Tolerance: "Can the user wait 10 seconds for an answer?" (Chatbots: Yes. High-speed trading: No).
3. Error Tolerance: "What happens if the AI is wrong?" (Marketing copy: Minor edit. Medical diagnosis: Lawsuit).

The "Human-in-the-Loop" Cost

If the Error Tolerance is low, you must pay a human to review the AI output.

• equation: Cost = (AI Cost) + (Human Review Time * Hourly Rate)
• If Human Review Time is same as Doing it manually, the ROI is zero.

---

4. Summary of Module 4

We have categorized value and strategy.

• Lesson 4.1: Value comes from Creation, Summarization, and Discovery.
• Lesson 4.2: Agents unlock transactional value (Doing > Reading).
• Lesson 4.3: Prioritize "Low Hanging Fruit" (High Impact, Low Effort). Avoid high-risk projects where error tolerance is zero.

Next Steps: In Module 5, we address the elephant in the room. What about Bias? What about Copyright? What about Data Leaks? We will cover Responsible AI & Governance.

---

Knowledge Check