Choosing an AI Platform for an Endowment's Investment Office
Turned 20 loosely written use cases into specified workflows, held six AI vendors to the same pass/fail gates, and laid out two defensible paths. The executives picked the other one. The follow-on work was built to make it succeed.
The ask
The investment office of one of the world's largest philanthropic endowments wanted to know which AI to bet on. About 35 people. A list of 20 use cases. A prototype agent their tech team had already built. An open debate about whether they needed a data warehouse first.
Their technical side was about two and a half full-time people. They didn't need a build. They needed a decision they could defend.
What I did
I led it end to end: discovery, analysis, vendor diligence, the scorecard, the contracts and the relationship.
- Use cases into workflows. Many of the 20 use cases were two or three builds sharing one name. I broke them into single workflows. Each one got a trigger, inputs, outputs, the steps between, must-have and should-have data, where that data lives, impact and priority.
- A map of the data. A second sheet listed every source they used or wanted, what it was for, and how we could reach it. An API, an MCP connector, or nothing yet.
- Requirements from the gaps. The workflows surfaced the non-negotiables. Email access. A connection to their CRM, ideally without building one. Security. Getting their data back out.
- Gates, then differentiators. Six vendors went through seven pass/fail gates and 15 nice-to-haves. Claude went through the same gates as everyone else. Every vendor claim was labeled demoed, claimed or inferred, and 37 open questions lived in one place.
Their tech team told me more than once that the workflow sheet made it obvious what needed to get built.
The recommendation
Two legitimate paths, matched to what they cared about.
- Path A: a finance-specific platform as the primary tool, plus Claude. Less to build and maintain. Value sooner.
- Path B: all in on Claude, across chat, Cowork, Claude Code and the API. Full control. More to build.
If speed to value mattered most, A. If long-term control mattered most, B. The investment team leaned A.
What happened
The executives went the other way. The wider organization was going all in on Claude, and they wanted the investment office on the same platform.
Nothing had to be redone. The trade-offs were already on the table, so the follow-on work was scoped to make Path B succeed: advisory, solution design and Claude training for the investment and technical teams, plus organization-wide training for AI ambassadors and executives. Two follow-on contracts signed in August 2026, about $170K together.
I wrote the handover guide for the team taking the account, and stayed on their first calls to help them start.
What I learned
The recommendation isn't the deliverable. The decision framework is. When a client picks the path you didn't lean toward, a good framework means you're already holding the plan for the one they chose.