Do not automate unclear work
If the team cannot explain how the work happens today, AI will not fix the problem. It may only make the confusion faster and harder to inspect.
Before automation, the business should know who starts the workflow, what data or documents are used, what output is expected, who reviews it, and what happens when something is wrong.
Unclear work needs mapping before automation. Otherwise the project becomes a technology-shaped argument.
Do not use AI when the output cannot be verified
AI-assisted systems should produce outputs that a person can check. If nobody can tell whether the answer is right, the system is not safe enough for business use.
This is especially important for summaries, recommendations, classifications, and answers from internal documents. The system should show sources, assumptions, and confidence limits wherever possible.
Avoid high-risk autonomous decisions
Early AI projects should not approve payments, reject customers, make legal commitments, change official records, hire or fire people, or silently send sensitive external responses.
That does not mean AI cannot support those workflows at all. It means AI should help prepare information for a qualified human reviewer, not replace the approval step.
- No silent payment approvals.
- No black-box employee ranking.
- No unsupervised customer commitments.
- No final legal, financial, medical, or compliance judgment.
- No sensitive data workflow without access controls and review rules.
Do not use AI to hide bad data
A model can make bad data sound coherent. That is dangerous. If reports disagree, definitions are unclear, or source systems are unreliable, the first fix may be data cleanup, validation, ownership, or plain reporting discipline.
AI can help explain exceptions or summarize messy text, but it should not become a decorative layer over numbers nobody trusts.
The no-build decision is still valuable
A good triage may conclude that a workflow is not ready for AI. That is not failure. It is useful judgment. The company avoids wasted money, protects trust, and can focus on a smaller workflow that is actually ready.
What to remember
- Do not use AI where the workflow is undefined or the output cannot be checked.
- Avoid autonomous approvals, financial actions, and high-risk decisions in early projects.
- Fix source data, ownership, and process gaps before adding AI.
- Saying no to bad AI ideas builds trust and protects the business.