The Questions to Ask Your Technical Team
Deliberate Academy Editorial Team
Reviewed for accuracy and professional relevance
You're 3 lessons in — don't lose your progress.
Sign up free to save where you are and earn a verified certificate when you pass.
- Identify the five questions that surface the most common causes of AI project failure, before a project is approved rather than after it stalls
- Distinguish a technical team's confident-sounding answer from one that actually addresses risk
- Explain why "how will we know if this is working" is a more useful question than "will this work"
- Apply these five questions to a described AI initiative and identify which answer reveals a genuine gap
A technical lead presents an AI project plan to leadership: three months to build, positive early results in testing, ready to scale. The leader in the room asks, "will this work?" The technical lead says yes, confidently, and means it. Six months later the project is quietly shelved — not because the AI model failed, but because nobody had defined what "working" meant well enough to notice, early, that the system's outputs were drifting away from what the business actually needed. The technical lead was not wrong. The question was.
This lesson is not about becoming technical. It is about the five questions that consistently surface risk a confident "yes" will otherwise hide.
The Five Questions
"What decision does this system make or influence, and who is accountable if it is wrong?" This question forces clarity on stakes before anything else. A system that drafts a first-pass email carries very different accountability than one that approves a loan or screens a job candidate. If nobody can answer who is accountable, that is the finding — not a detail to sort out later.
"How will we know if this is working, and how often will we check?" Not "will this work" — that invites a confident yes. This question forces a concrete, measurable answer: a metric, a review cadence, a named owner. A team that has genuinely thought through the deployment will have this ready. A team that has not will describe the model's capabilities instead of answering the question.
"What happens when it is wrong, and who notices first?" Every AI system will be wrong sometimes. The question is whether a human notices before or after the error reaches a customer, a regulator, or a board deck — and whether that human has the authority and information to actually intervene, not just a nominal review step.
"What data is this trained or operating on, and do we have the rights and quality to use it this way?" Data quality and rights issues are the single most common reason AI projects stall mid-implementation. Asking this question at approval time, not after a contract is signed, is what separates leaders who catch this early from those who fund a project twice.
"What is the plan if this underperforms, and at what point do we decide to stop?" A credible technical team will have a defined off-ramp — a specific point at which the project is paused or redirected if results do not materialize. A team without one is asking for open-ended investment against an undefined bar.
Ask these five questions in a follow-up email after the meeting, not just verbally in the room. A team that answers precisely in writing has actually thought it through. A team that answers vaguely, or takes a long time to respond, is telling you something about how far the plan has actually been developed.
A leader asks a technical team 'will this AI system work?' and receives a confident 'yes.' What is the strongest critique of this exchange, based on this lesson?
Select one answer.
The Off-Ramp Question That Surfaced a Missing Plan — Manufacturing Firm
Context
A VP was asked to approve an AI-powered predictive maintenance system projected to reduce unplanned downtime. The technical team's presentation was confident and well-produced, covering the model's approach in detail.
Action
The VP asked the fifth question from this lesson directly: what is the plan if this underperforms, and at what point do we decide to stop? The technical lead had not defined a stopping point — the plan assumed continued investment and iteration until the system worked, with no defined threshold for cutting losses.
Outcome
The VP required the team to define a concrete underperformance threshold — a specific false-alarm rate and a missed-failure rate, reviewed at a 90-day checkpoint — before approving the budget. At the 90-day review, the system was underperforming on one of the two defined metrics. Because the threshold had been agreed in advance, the team redirected the project to a narrower use case rather than continuing to invest against an undefined bar, a redirection the VP credited directly to having forced the question at approval time rather than discovering the gap eighteen months in.
Why does this lesson recommend asking the five questions in a written follow-up rather than relying only on the verbal answers given in a meeting?
Select one answer.
Exercise
Your Task
Take an AI initiative currently being discussed in your organization (or a plausible one). Draft the five questions from this lesson as you would actually send them to the responsible technical lead. For each question, note what a strong answer would specifically include, so you can recognize a weak or evasive answer when it comes back.
Your reflection
Did you complete this exercise? What did you find? (Saved locally in your browser)
- The five questions — accountability, measurement, failure detection, data rights, and off-ramp — surface the most common causes of AI project failure before approval, not after months of investment.
- "How will we know if this is working" forces a measurable answer; "will this work" invites an unfalsifiable, confident one.
- Data quality and rights issues are the single most common reason AI projects stall mid-implementation — asking about them at approval time catches the problem before a contract is signed.
- A credible technical team will have a defined off-ramp — a specific point at which the project is paused if results do not materialize. The absence of one is itself the finding.
- Requesting written answers to these questions, not just verbal ones in a meeting, is a reliable way to distinguish a genuinely developed plan from a confident but underdeveloped one.