The 12 Questions Every Organization Must Ask Before Deploying AI
After working with hundreds of organizations on their AI journeys, I've identified the questions that separate responsible deployment from reckless adoption. None of them are technical.
When I work with organizations on AI implementation, the conversation almost never starts where it should. Teams are eager to talk about which models to use, which vendors to trust, which workflows to automate first. The technical questions get all the airtime. The human questions get pushed to 'we'll figure that out later.'
Later, in my experience, never comes — or it comes in the form of a crisis.
Here are the 12 questions I've come to believe every organization must answer honestly before any AI is deployed in a people-facing context. I'm not a philosopher. I'm a technology executive who spent 30 years shipping products to millions of users and learned, sometimes the hard way, what the questions are that you really should have asked.
1. Who benefits from this deployment — and who bears the risk? These are almost never the same people. When a hospital deploys AI for triage decisions, the efficiency gains accrue to the institution. The risks — misdiagnosis, bias in prioritization, data exposure — are borne by patients. Be honest about this asymmetry.
2. What does failure look like, and who suffers it? Every AI system fails. The question is how and for whom. When your AI recommendation engine fails, does a user get a slightly worse product suggestion, or does someone lose their housing assistance application?
3. Can the people affected by this system contest its decisions? If not, you haven't built a system — you've built a judgment machine with no appeal process. Every meaningful deployment needs a human override layer.
These questions aren't obstacles to AI adoption. They're the scaffolding that makes adoption durable.