I’ve sat through enough kickoffs that start with timelines and tools and somehow never get to outcomes, and by the time someone finally asks what “done” means the budget is already half-committed. That isn’t a process problem so much as a courage problem dressed up as project management.

Whether you’re looking at cloud migration, a security overhaul, an ERP change, a custom web portal, or an AI agent program, the same questions keep people out of ditches. None of them are clever, and that’s intentional.

1. What has to get better, and how will we know?

If success is “go live,” you have a theater metric. Prefer outcomes you can argue about with a straight face: fewer tickets of a certain type, faster close, restores that actually work, phishing that doesn’t end in a scramble, one workflow that stops bouncing between three systems, a public site that converts without a page-builder hostage situation. AI projects are especially bad at this because demos look successful even when operations don’t change. If you can’t name a measure you’d bet a week of payroll on, you’re not ready to buy.

2. Who owns day two?

Every project has a hangover. Name the internal owner before the vendor’s project manager disappears into the next logo slide. If nobody owns it, you’re buying permanent dependency. Sometimes that’s fine, but it should be a choice rather than a surprise at renewal. I once asked “who owns this after go-live?” and three people pointed at each other, so we paused the project. Best meeting of the quarter.

3. What are we not doing because we’re doing this?

Opportunity cost is real and people hate talking about it. Funding a new platform while backups are unverified, identities are messy, or monitoring is a firehose is a classic failure pattern. We push that trade-off into the open because someone has to, and the same goes for custom development when nobody has budgeted for support after launch.

4. What’s the reversible path if we’re wrong?

Some decisions are glue: multi-year licenses, proprietary data models, network redesigns, process changes driven by agents that rewire how work gets done, web platforms you can’t export cleanly. Map reversibility. If the answer is “we’re committed,” raise the evidence bar, because commitment without evidence is just stubbornness with a purchase order.

5. What does “no” look like in this process?

A healthy process can reject a proposal. If the only acceptable answer is approval, you’re in a sales process. We’ve recommended wait, cut scope, or keep the money. Clients don’t always like hearing it. They’re usually glad later.

6. How do security, compliance, and support change on day one?

Pretty architecture diagrams are not operational design. Ask about identity, logging, HIPAA/PCI constraints if they apply, and who answers the phone when the new thing breaks at 7 a.m. Projects that ignore handoffs create silent debt, and silent debt collects interest.

Use these as a gate, not a worksheet

If leadership can’t answer most of these in one meeting, don’t sign. That’s not bureaucracy. That’s risk management. We often turn the answers into a one-page decision memo: outcome, owner, cost of delay, reversibility, vendor incentives. When AI is involved, add one more: what happens when the agent is wrong? If the answer is “a human will notice,” define how. If the answer is “it won’t be wrong,” you’re listening to marketing.

Print the questions, put them in the meeting invite, and when a vendor tries to skip to architecture, point at the list. It feels awkward the first time and professional the second. By the third project, people wonder why they ever approved work without it. Nearly three decades of watching people skip this list has made me boring on purpose, because boring saves money, and asking dull questions while walking away is still an option is how adults buy technology.