How Leading Family Offices Are Turning AI Investment Into Real Impact

Four Reasons an AI License Hasn’t Produced an AI Result 

Half of family offices have invested in AI, but less than 15% are using it daily, according to Citi Private Bank’s Global Family Office Survey. 

Licenses are bought. A few people are getting real results. Nothing has changed how the office works.

Some of that is a function of how this cycle arrived. Cloud and ERP landed at the executive level and were pushed down from there. AI landed on individual desktops. Someone runs Copilot in Outlook, someone else drops a pair of contracts into a chat tool to spot where the language diverges, and the capability belongs to that person rather than the firm. It leaves when they do.

Greg Tselikis and Nick Scott of SC&H spent the hour on what closes that distance. Four points stood out:

1. The assessment is the step that pays for itself 

An honest read on data, process, and people gives you an inventory of everywhere AI could help. That inventory is what makes the first project a decision instead of a reflex. 

Working with a multi-state nonprofit, Greg and Nick came back with about 20 places AI could plausibly earn its keep. Only a handful made the first cut, chosen for clean data and a team ready to pick something up. The earliest projects weren’t the biggest earners on the list, because the biggest earners needed data nobody had captured yet. 

Skipping this usually costs more than running it. The build gets a quarter finished before someone finds the gap.

In connection with taking control, lenders need to set up acquisition vehicles, negotiate governance rights, and figure out what to do with the old company. Whether it’s a formal dissolution, an assignment for the benefit of creditors (ABC), or simply letting the entity fade out, the goal is to minimize risk and avoid unnecessary litigation.

2. Most of the available value is in tier-one work 

SC&H scores candidate projects across three tiers, usually a crawl, walk, run approach. Tier one has a benefit you can put a number on and a scope small enough to pilot without a budget fight. Tier three is the cross-system agentic work that fills most vendor decks. 

Once an office has been assessed, tier one is usually where the backlog sits, and it’s the cheapest tier to deliver. Moving up is optional. There’s no schedule but yours. 

3. Your AI policy doesn’t tell you what’s actually happening 

A document naming which tools are allowed and which aren’t is a starting point. You should also be able to put a direct question to IT and get a direct answer: what’s in use across the office, and what data is going into it. Shadow AI is almost always bigger than leadership assumes. 

Paying for a subscription protects nothing on its own. On the major platforms, a personal or consumer plan can feed your chats into model training unless someone switches that off. A business or enterprise agreement rules it out in writing. If your office handles family financial records, confirm which plan you’re on before anything else, and pair that with a real vendor review covering SOC 2 and where your data physically lives.

4. Training is what turns AI spend into a result 

Greg has sat with CFOs who can point to twelve months of AI invoices and nothing on the other side of them. The pattern repeats: the office bought licenses, sent out logins, and left the rest to chance. 

What changes it: A handful of people trained deeply enough to teach everyone else. Sessions built around the work your team already does. Shared workspaces, so context lives somewhere other than private chat histories. And a plain rule about ownership, which is the one item here with no software in it. If you send it, you own it. AI drafting doesn’t move responsibility for accuracy off your desk.