Where AI Belongs
AI opportunity · 01
AI does not belong everywhere.
It belongs where better judgment changes an outcome people care about.
AI
01 / 08
The starting point · 02
Begin with the decision—not the technology.
Name who is deciding, what they are trying to improve, and what evidence they use today.
02 / 08
Choose a role · 03
Give AI the smallest useful job first.
Low permission
Retrieve
Bring the right context forward.Recommend
Suggest an answer or next step.Prepare
Draft work for human approval.High permission
Act
Complete work inside defined limits.
03 / 08
Opportunity map · 04
Prioritize high value with bounded downside.
High value · low downside
Start here
Fast learning with a credible path to impact.High value · high downside
De-risk first
Use review, limits, and staged exposure.Low value · low downside
Automate later
Convenient, but unlikely to change the business.Low value · high downside
Do not build
Complexity without a compelling return.04 / 08
A good candidate · 05
The strongest opportunities create their own learning loop.
01
Frequent
Enough repetitions to learn quickly.02
Context-rich
Relevant data can improve the judgment.03
Observable
Good and bad outcomes can be recognized.04
Recoverable
A weak answer can be corrected safely.
05 / 08
The first experiment · 06
Test the riskiest assumption before building the system.
01
Collect
50 realistic cases02
Compare
Baseline versus AI-assisted03
Review
Quality, time, and failure04
Decide
Stop, revise, or expand
06 / 08
Product discipline · 07
A serious AI idea includes a reason to stop.
Keep going when
Evidence improves the decision.
Quality, speed, adoption, and safe recovery move together.Stop when
The workflow absorbs the value.
Review burden, edge cases, or low trust erase the promised gain.07 / 08
The output · 08
Leave with one decision—not an AI backlog.
Choose one opportunity, one role, and one test that can change your mind.
- The decision
- The smallest useful AI role
- The evidence required to continue
08 / 08
Slide 1: AI does not belong everywhere.