← Mission 4 Teach Your Workflow to Decide 20 slides

1 / 20

Agentic AI for Business — Mission 4

Title Slide

1Agentic AI Essentials
2Teach Your Workflow to Decide
3Mission 3
4Fast isn't enough. We need fast and right.
5#AIOfficerCertification | Dave Hajdu | dave@ai-officer.com

2 / 20

Continuity Bridge

1Mission 1: You packaged your AI.
2Mission 2: You wired it into a workflow.
3Mission 3: You teach it to make decisions.

3 / 20

The Problem

1Your workflow is fast.
2But it treats everyone the same.

4 / 20

Today's Agenda

11. When one path isn't enough
22. The decision pattern: Classify -- Route -- Respond
33. Build the logic layer

5 / 20

The Linear Workflow Problem

1New feedback comes in.
2AI writes one generic reply.
3You send it, log it, move on.

6 / 20

The Fix -- Mental Model

1First listen. Then decide. Then respond.

7 / 20

Exercise -- Classify These

1Exercise: What type is each one?
2A: "Your team was so helpful on my last order!"
3B: "My package is late again and I still don't have tracking."
4C: "Have you ever thought about adding a subscription option?"
5Share your labels in the chat.

8 / 20

The Decision Pattern

1Classify -- Route -- Respond.

9 / 20

The Pattern Works Everywhere

1Customer feedback -- Revenue Office.
2Employee surveys -- Talent Office.
3Time-off requests -- Operations Office.
4Focus group notes -- Innovation Office.

10 / 20

Workflow Structure

1New feedback -- Classify -- Route -- Respond -- Log.

11 / 20

Exercise -- Tiny Decision Canvas

1Exercise: Design your logic on paper before touching any tool.
2For each input type, define: what it means, and what happens next.
3PRAISE -- Customer is happy -- Thank them, ask for review.
4COMPLAINT -- Something went wrong -- Apologize, help fix.
5SUGGESTION -- Idea for improvement -- Thank them, confirm captured.

12 / 20

Building the Logic Layer

1Before: Trigger -- AI -- Action -- Log.
2After: Trigger -- Classifier -- Router -- AI -- Action -- Log.

13 / 20

Exercise -- Step 1: Build the Classifier

1Exercise: Build your classifier.
2System prompt: "Classify this feedback. Output ONLY one word: PRAISE, COMPLAINT, or SUGGESTION. No explanation."
3Input: feedback text. Output: one label.

14 / 20

Step 2: Build the Router

1Router reads the label. Selects the path.
2PRAISE -- Praise path.
3COMPLAINT -- Complaint path.
4SUGGESTION -- Suggestion path.
5The router doesn't think. It just directs.

15 / 20

Step 3: Build the Responses

1Exercise: Build a response for each path.
2PRAISE -- Thank them, optional review ask.
3COMPLAINT -- Acknowledge, apologize, help fix.
4SUGGESTION -- Thank them, confirm captured.
5All paths end with: send reply + log.

16 / 20

Exercise -- Pick Your Building Blocks

1Exercise: Map your components in your automation platform.
2You don't need to connect everything yet. Just identify: trigger, classifier, router, response paths, log.

17 / 20

Outcome

1You designed a workflow that classifies, routes, responds differently, and logs everything.
2This is personalization at scale.

18 / 20

What's Next

1Today you designed every path.
2What happens when you can't?

19 / 20

Your Commitment

1What is one process in your work where different inputs should get different treatment -- and currently don't?

20 / 20

Thank You

1Thank you.
2[QR Code Placeholder -- Session Feedback]
3#AIOfficerCertification | Dave Hajdu | dave@ai-officer.com | ai-officer.com