1. What's the one thing that fails on Monday morning if this project disappears?
The honest answer tells us the project's actual critical path. It's usually a specific report, a specific automation, or a specific person who can finally stop running a Friday-evening spreadsheet.
The dishonest answer — "a lot of things, really" — tells us the project doesn't have a real owner yet. Both responses are useful. The second one means our first deliverable is helping the client find the owner.
2. Who can decide a thing on the spot, and who do they have to call first?
Decision latency is the single biggest predictor of how a project will go. If "we'll need to check with finance / legal / the parent company" shows up in the answer, multiply your timeline by 1.5 and put the meeting on the calendar before the kick-off ends.
3. What's the existing thing we should not touch?
There's always one. A legacy service. A spreadsheet someone runs on the 15th of every month. A poorly documented Lambda that fires on a Friday and absolutely nobody knows why.
Ask directly and the team will tell you. They will not volunteer this in a kickoff deck. The follow-up is the question that matters: "And what would it take to retire that thing?" Sometimes the answer is "two days and a small bet." Sometimes it's the whole project.
4. What happens to the data if we shut this down tomorrow?
A few clients answer immediately. Most don't.
The question is a gentle way of forcing them to describe data ownership, retention windows, and contractual obligations in concrete terms. Whatever the answer, it tells us how the data should be modelled, who needs an export feature, and which fields will need an audit trail by the time we ship.
5. What's the success metric you'd defend in a board meeting?
"Ship by Q3" is a deadline, not a metric. Push back gently.
The real metric is a number that moves: revenue uplift, churn reduction, time-to-fill, fraud rate per 10k transactions, ops cost per onboarded customer. If we can't agree on the metric in the first conversation, we can't agree on what "done" means six months in. And if we can't agree on "done," every conversation in week 22 is going to be the same conversation as the kickoff — only louder.
None of these are about the code
By the time you look at the code, the architectural decisions are already made. They were made when somebody answered question one with the wrong name, or when nobody could answer question five.
Spend the 30 minutes. It pays back in months.