What lets a firm run AI in production, at scale, with its reputation on the line. The work starts with one real workflow, not a slide deck.
Most AI programs produce a faster version of the same work, because they automate the container and keep the sequence that was built around a person.
Ask a firm which of its processes could be automated and you will get a list of processes. That list is the wrong unit. A process is a container, and inside it sit a dozen distinct cognitive acts that have nothing in common except that one person currently performs them in sequence.
Batch sizes, handoffs, the monthly close, the four-eyes rule, the exception report, the standard template: none of these are laws of nature. They are artifacts of human throughput and human error, and they are load-bearing only for as long as production is the expensive part.
How the cognitive work is allocated today, what the control model silently assumes, and which of those assumptions AI has already invalidated. That produces a map and a sequence, not a deck.
From there the work is designing how AI actually runs in your operation: what gets built, how it is validated, who signs, and how the people who spent careers performing the work become the people who stand behind it.
How cognitive work gets decomposed, repackaged for an AI era, and allocated across humans, agents, and classical software. Most firms skip the repackaging step, which is why most programs deliver speed rather than change.
What makes deployed AI trustworthy. When machines produce the work, verification and accountability become the scarce commodities, and the maker-checker pyramid was built for a failure mode that no longer applies.
The migration path. The people who performed the work must become the people who stand behind it. That transition is a program rather than a memo, and it is where most transformations will actually fail.
Monthly rather than hourly. Early results come fast. The change that counts takes quarters, and I will say so.
Most firms cannot state their current error rate or cost per unit, which makes every later claim unfalsifiable. We fix that first.
Not delivered to them. What we build together is owned outright and runs without me.
I resell nothing and take nothing from vendors. Where the right answer is a tool you already own, that is the answer.
Transformation functions from the ground up at Gen II and at SEI, and transformation run at JPMorgan and Citi.
I helped found Booz Allen's Value Engineering practice, built on reengineering companies for growth rather than for cost.
Most of the business at a fund administrator, the largest hedge fund and private equity relationships at a global provider, and a P&L as a general manager.
Including the one I run my own week on, and machine reading of documents brought into fund operations before it was fashionable.
When there is a concrete operation to fix, or when the leadership team already has firsthand command of the tools. If neither is true yet, coaching is usually the faster start, because an operational program without fluent sponsors tends to stall at the first hard decision.
The framework is set out in Trust Is the Bottleneck and The Validator Operating Model. The complete version, including how confidence scoring is constructed, is available as a briefing, by email.
Leaders who cannot sponsor the change. A transformation group that is busy without producing results. Nobody who owns AI. Those look like three problems and they are one, which is the distance between what AI can do and what your organization will let it do. Coaching, transformation advisory, and the Operating Partner seat are three doors into the same work, and the engagements that work usually use more than one.
The first conversation is thirty minutes and free. You will leave with an honest read on whether there is anything worth doing.