Discovery at Island Mountain runs under six clauses, and we hold every engagement to them. A sales process that can’t fail can’t be trusted, so these exist to make failure possible.
One. The pain must be frequent and owned. A problem that appears twice a year, or belongs to nobody who feels it weekly, doesn’t justify infrastructure. It justifies a calendar reminder.
Two. The data must be available, governed, and lawful to use. “The data exists in theory” means a governance project’s hiding inside the AI project. It goes first, or everything waits.
Three. Success gets measured in numbers the institution already respects. Enthusiasm isn’t a metric. A baseline you can’t state before the build is an improvement you can’t claim after it.
Four. The human keeps the verdict. When a business case only works by removing a consequential decision from a qualified person, the engagement stops there. That’s the line I don’t move for any budget.
Five. The facility and the operating team must be able to carry the system. Power, cooling, administration, recovery. A deployment the institution can’t sustain is a countdown, not a capability.
Six. Risk stays proportionate to the measured gain, with blast radius sized honestly, including what one bad output does downstream on a crowded day.
Three verdicts come out of that gate, and all three carry value. “No” protects staff from another pilot that arrives with executive attention and departs leaving cleanup. “Not yet” names the dependency worth solving on its own schedule. “Yes” produces an owned workflow, a baseline, authority boundaries, and acceptance tests written before procurement.
A sequencing corollary I’d defend anywhere: make the first deployment dull, frequent, measurable, and small in consequence. Draft the report. Assemble the record. Early credibility funds the harder second problem.
NIST’s AI RMF fronts context for the same reason this gate exists. A model can’t be judged apart from its task, its setting, and what its failures cost.
Run one diagnostic on your process: name the finding that would stop the project. If nobody can, the decision was made before the meeting, and the meeting’s theater.