When AI Is the Wrong Tool
Why this matters
Most AI failures start with the wrong problem being handed to the model. The model is not at fault. The filter was never run. Before you reach for any tool, run it.
What you will be able to decide after this
- Whether a problem is an AI problem, a process problem, a data problem, or an ownership problem.
- When a simple rule or checklist beats a model.
- When to say "fix something else first" instead of building.
Core lesson
The five-question filter:
- Can a competent human already do this reliably with the data you have?
- Does this decision or task happen often enough to justify the cost of setup, review, and maintenance?
- What happens when the output is wrong?
- Would a simple rule or checklist solve most of the cases?
- Who owns the result when it fails?
If the answers are weak, AI is not the solution. Process, data quality, or clarity of ownership usually is.
Note what the filter does not include: how exciting the tool is, what the vendor demoed, what the benchmark said, or whether everyone else is doing it. None of those survive contact with question three.
Worked example
A team wants AI to route customer emails to departments. Run the filter: a competent human can already do it (question one: weak for AI). The volume justifies the cost (question two: maybe). When routing is wrong, a customer waits or gets the wrong queue (question three: visible, moderate). A simple rules filter on subject lines handles most cases (question four: yes — weak for AI). The team owns the result (question five: unclear).
The honest call is not "use AI." It is "fix ownership first, and start with rules." The AI can wait.
Practice
Take three tasks from your own work and run all five questions on each, in writing. Resist the urge to pre-decide. Write the answers, then the call.
Apply — produce the artifact
For each of three real tasks or decisions from your work, answer the five questions in writing. End with a clear call: Use AI / Do not use AI / Fix something else first.
Verify
Have someone who knows the work challenge one of your "Use AI" calls. If you cannot defend the volume, the failure mode, and the ownership, change the call.
Sources
This module is original practice guidance based on the authoring standard and does not depend on a specific external factual claim. Editorial review is still required.