# AI suitability checklist
Work through the factors that make a task a good or poor fit for AI before you commit.
## The task
- [ ] The task involves interpreting language, documents or variable input.
- [ ] It is not easily captured as a stable `if → then` rule.
- [ ] It happens often enough for the effort to pay back.
- [ ] It is currently slow or costly to do by hand.
## The data
- [ ] The information the task needs is available to the workflow.
- [ ] The input is reliable, or can be validated.
- [ ] There is a clear [source of truth](/knowledge/what-is-a-source-of-truth) for the facts involved.
## The errors
- [ ] A wrong result can be detected.
- [ ] Someone can judge whether a result is good.
- [ ] The cost of a wrong result is understood.
- [ ] Consequential actions are reversible or gated by approval.
## The workflow and ownership
- [ ] The underlying workflow is understood and mapped.
- [ ] Exceptions have been identified and have owners.
- [ ] One person can own the workflow after launch.
- [ ] Responsibility for the result can be assigned.
## The alternative
- [ ] A simpler mechanism has been considered and ruled out.
- [ ] You can explain why AI is justified over that simpler option.
In 30 seconds
Use this before committing to an AI approach. It is a structured judgment, not a score. If most answers point the wrong way, prefer a simpler mechanism or more preparation first.
The task
The task involves interpreting language, documents or variable input.
It is not easily captured as a stable if → then rule.
It happens often enough for the effort to pay back.
It is currently slow or costly to do by hand.
The data
The information the task needs is available to the workflow.
Consequential actions are reversible or gated by approval.
The workflow and ownership
The underlying workflow is understood and mapped.
Exceptions have been identified and have owners.
One person can own the workflow after launch.
Responsibility for the result can be assigned.
The alternative
A simpler mechanism has been considered and ruled out.
You can explain why AI is justified over that simpler option.
What this checklist does not prove
A good result here means the task is worth designing for AI, not that the build will succeed. Suitability is the entry condition. Decomposition, authority design and testing still decide the outcome.
