Today’s outcome

• Build a reviewable AI workflow
• Preserve sources and formulas
• Separate decisions from AI suggestions
• Choose the right human/tool/automation boundary

The business workflow canvas

Source → transform → review → decision → record

AI can assist at every stage. It should not erase the source or silently make the decision.

AI across the business

Demo: from messy CSV to decision memo

1. Preserve the raw export
2. Create a working copy
3. Ask AI for anomalies
4. Verify calculations
5. Draft a cited recommendation
6. Ask a human to approve the action

Do not automate past the risk boundary

AI may draft, classify, summarize, and queue.

Human approval is required for external sends, publication, payments, access changes, production changes, and unresolved ambiguity.

Mini-practice: chat to action list

Extract only:

• agreed decisions
• named actions and owners
• assumptions
• unresolved questions

Do not invent agreement from tone.

Lab: Operations Intelligence Sprint

Use synthetic CSV, spreadsheet, research pages, GitHub issues, and chat.

Deliver: a sourced one-page decision memo and an automation boundary.

Review question

“Where did this number or claim come from?”

If the answer is “the AI,” the work is not ready.

The real skill

AI literacy is operational judgment:

Know what to automate, what to review, what to preserve, and when to stop.

Exit ticket

Name one workflow AI can accelerate and the evidence/approval it still needs.