The AI Decision Matrix

 

Who decided what your AI is allowed to decide?
⠀
In many companies, nobody did.
⠀
Researchers at MIT's Center for Information Systems Research built a tool for this.
They interviewed 30 executives and created the AI Decision Matrix below.
⠀
It sorts every decision by two questions:
→ Ambiguity: how clearly does the data point to one answer?
→ Risk: how much damage if it's wrong, and how hard is it to undo?
⠀
Then it splits each decision into three parts:
→ Frame: define the problem.
→ Act: recommend or take the action.
→ Learn: own the outcome and improve it.
⠀
THE FOUR DECISION TYPES
⠀
◆ Exploratory (high ambiguity, low risk)
New questions where learning matters more than precision.
Humans and AI frame the problem together. AI acts. A human owns the learning.
⠀
◆ Strategic (high ambiguity, high risk)
Decisions with organization-wide consequences.
Humans lead. AI supports the framing and action. A human owns the learning.
⠀
◆ Routine (low ambiguity, low risk)
Well-defined decisions with little downside.
Automate and monitor. Even here, autonomy is earned.
At One New Zealand, marketers still review AI-built campaign segments before launch.
⠀
◆ Consequential (low ambiguity, high risk)
Well-defined decisions that affect many people or significant money.
AI frames and acts. Humans manage exceptions and own the learning.
One New Zealand drew a hard line: every pricing decision keeps a human in the loop.
⠀
WHAT'S EASY TO MISS
⠀
Look at the Learn column.
Only in Routine decisions does AI own the outcome.
In the other three, a person does.
⠀
AI can do the work. A person still answers for it.
At One New Zealand, every deployed agent has a named human owner.
⠀
I learned why in pharma.
I once pulled a batch our quality system had cleared.
The equipment was slowly drifting out of calibration.
The data said fine. The person who owned the outcome said stop.
⠀
Second, decisions move.
A routine decision can turn consequential when prices, volumes or rules change.
So the quadrant you pick today needs a review date.
⠀
RUN THIS WITH YOUR TEAM THIS WEEK
⠀
1. List every AI use case you have live.
2. Write the business decision each one shapes.
3. Place each decision in a quadrant.
4. Name the person who owns Learn.
5. Set a date to recheck its risk and ambiguity.
⠀
The researchers' core point is the one to keep:
stop asking where AI can be used, and start managing the decisions it shapes.

https://lnkd.in/p/dhQ-599E