The Return Is a Person, Not a Platform

The wealth firms showing a top-line productivity gap from AI aren't using it to pick investments. Ninety-six percent of them aren't. They're using it on meeting notes, documents, and CRM updates — the unglamorous work that eats the advisor's calendar. The revenue shows up because the advisor got hours back, not because the model got smarter about markets.

Short post for wealth executives: "The 4% and the 96%." Contrast the 4% using AI for investment decisions with the 96% getting returns from administrative work, and ask which one their own AI budget is actually pointed at.

Most AI ROI conversations start with "what can we cut." For a company that doesn't have revenue yet, the ROI question is inverted: what capital can we win that we couldn't have competed for before? A $12 million award pursued with AI-built materials is a revenue case, and the math on it is simpler than any productivity claim.

A piece for founders and CFOs on the three AI use cases that show up before the first dollar of revenue: grant and award applications, investor materials, and competitive intelligence. Frame each as capital won, not cost avoided.

The tool cost in these businesses is almost trivially small — pennies per output, tens of dollars a month for hosting. That is the same pattern we saw in the manufacturing room last month: seats at $18 to $25, a company-wide cap under $2,000. Nobody's ROI problem is the subscription. The cost that actually matters is the time of the one person who builds the workflow. If your AI business case is arguing over license fees, it's arguing about the wrong line.

A one-graphic post: a two-column ledger, "What the tools cost" versus "What the person cost," with real figures from three operators. The visual contrast does the work; the caption asks which column their own AI budget is tracking.

Here's the line every executive should take from this study: the driver of AI success wasn't firm size and it wasn't spending. It was whether someone owned it, whether people were trained, and whether results were measured. Those are choices, not resources. A $500 million firm can make every one of them this quarter. Most $5 billion firms haven't.

Workshop exercise: hand a leadership team the four leader traits — governance model, named owner, formal training, formal measurement — and have them score their own firm one to five on each before anyone mentions a tool. The scores are the readiness assessment; the discussion is the workshop.

Notice what's missing from that list: nobody claimed a percentage of revenue and nobody cited a vendor benchmark. Every number was a task the speaker used to do by hand, with a before and an after. That is what credible AI ROI looks like at this size of company right now. If a proposal in front of you has a bigger number with less specificity, ask who measured it.

A short video walking one of the six ledger rows end to end — the wiring schematic to digital twin is the most visual — ending on the weeks-to-build, minutes-to-run ratio as the shape of every real return.

The gap between "24 hours to 2" and "20% no-touch" is the whole compliance story. The clock got compressed without removing the human from the loop, which means for the next several years the binding constraint isn't model quality — it's whether the people still in the loop can make defensible decisions at eight times the previous speed. Regulators won't ask how fast you got. They'll ask what stayed human, and why.

Practical piece: "Judgment mapping" — instead of ranking what to automate, rank what must not be, document why, and treat that document as the artifact you hand to audit, risk, and regulators. Most AI programs can't produce it when asked. Use the carrier's numbers as the setup, not the subject.

This is the IP story hiding inside a cost story. When a third of companies build instead of buy, the workflow they built is now theirs — tuned to their shop, owned outright, and no longer a line in a vendor's renewal. The manufacturers in our room did exactly this: a 3D-printing bureau built its own execution system because the commercial one wasn't tuned to its shop. The question for every executive is which of their next three software renewals is actually a build decision in disguise.

Post for CFOs and CIOs: "Add one question to every software renewal." The question is whether agents could build it safely, with a run-cost estimate and an ownership plan alongside the vendor price. Use the 32% figure as the hook and the manufacturing example as proof it's happening at mid-market scale.