What should a manager ask before approving an AI project?
Ask seven things: what task it improves and for whom, how success will be measured, what data it needs, what happens when it is wrong, what it costs to run, who owns it after launch, and what the smallest useful version is. Clear answers matter more than the technology chosen.
What task does it improve, and for whom?
A good answer names a person, a task and what is slow, costly or error-prone about that task today. A warning sign is a goal that only says the company should use AI.
Ask the team to finish this sentence: "When this is live, this person will be able to do this, which today takes this long or goes wrong this often." If they cannot, the goal is not clear yet.
How will we know it works?
Ask for the measure before anything is built. A team that is ready can say which past cases it will test on and what result counts as good enough. "We will know when we see it" means the team is not ready.
What data does it need, and may we use it?
A good answer says where the data is, who owns it, what state it is in and whether this use is allowed. A warning sign is a plan that assumes the data is clean and ready.
Check three things early. Can the team get access? Is the data good enough? Do privacy rules, contracts or company policy limit how it may be used? These three stop many projects before the technology is ever the problem.
What happens when it is wrong?
Every AI system is wrong some of the time. A good answer says what a wrong answer costs, who catches it and how. A warning sign is a plan with no check between the system and the customer.
What will it cost to run?
Ask for the cost of running the system, not only of building it. That includes the cost of each use, the monitoring, and the people who keep it working as things change.
Who owns it after launch?
A good answer is a name. Someone must watch quality, handle complaints and update the system. If no one is named, the project ends on launch day.
What is the smallest version that would be useful?
Ask for the smallest version that a real user would use in real work. It shows whether the idea works before the larger budget is spent.
If the team can put these seven answers on one page, the thinking has been done. If not, the next step is that page, not a bigger budget.