Let’s talk dollars.
In the last few posts we’ve described the hidden labor problem in abstract. This post puts numbers on it, because we think the industry talks around the economics rather than through them.
The numbers below are drawn from Deloitte’s Family Office Insights series, J.P. Morgan’s Global Family Office Report, Botoff Consulting’s compensation benchmarks, and the operational benchmarks Aleta has published — all 2024 vintage (Fortune).
The cost structure of a single-family office
A full-service single-family office can run over $1M a year, and personnel is 60–70% of that cost. The median chief investment officer is making around $500K, per Botoff. Below the CIO sits the operating layer — the controller, the accountants, the analysts, the bookkeepers. That layer typically runs:
Bucket one— controller / head of operations: $200K–$350K all-in.
Bucket two— staff accountant or operations analyst: $90K–$180K all-in, often two or three of them.
Bucket three— outsourced fund admin, CPA, and audit fees: $75K–$300K depending on complexity.
Add it up and the structured-data-and-reporting layer of a single family office is comfortably $300K–$700K a year. For a family with $500M AUM, that’s 6–14 basis points of assets, every year, in perpetuity, just to know what they own.
For a multi-family office, the layer scales but doesn’t get cheap — it gets distributed across more relationships, with comparable per-family economics.
Why “just hire another analyst” stopped working
For a decade, the answer to growing operational complexity in family offices was to add headcount. It worked because the analyst supply was deep, the assets were less weird (more publics, fewer private vehicles), and the technology was a complement rather than a substitute.
That formula has broken. Three reasons.
One— the alts mix has roughly doubled. UBS’s Global Family Office Report shows family offices now allocating 42% to alternatives, up substantially over five years (PR Newswire). Each alt position is roughly an order of magnitude more operationally intensive than a public position.
Two — the talent supply has tightened. McKinsey projects an advisor shortfall of 90,000–110,000 by 2034, and the operations talent pipeline that feeds into family offices and RIAs is comparably stressed. Hiring another senior accountant is harder and more expensive than it was in 2019.
Three— the consolidator math. RIA consolidators now control roughly $1.5T in AUM, per Cerulli, and they’re bidding aggressively for back-office talent. Family offices that used to pull from regional CPA firms now compete with PE-funded RIA platforms for the same hires.
So the cost line goes up, the supply line goes down, and the complexity line goes vertical. You can see why principals are open to having a conversation they wouldn’t have had three years ago.
Where AI does and doesn’t change the math
This is where it gets interesting, and where we want to be careful, because there is more hype than truth in the wealth-AI conversation right now.
McKinsey’s research suggests gen AI can reorient 20–30% of an advisor’s time toward growth and away from low-value work. Their asset-management research projects 25–40% of total cost base potentially capturable through AI-driven workflow redesign at scale. Real numbers. Real opportunity.
Where AI works in family-office operations, today, in production:
(1) Document classification and extraction.A capital-call PDF, a K-1, a quarterly statement — modern LLMs paired with structured extraction pipelines handle these well, with human review for exceptions.
(2) Reconciliation flag generation.AI is good at saying “this position doesn’t tie; look here.” Better than rules, faster than people.
(3) Summarization and first-draft client communication. Strong, with human sign-off.
The beauty of a mixed solution — AI × White Glove humans — is also that each of these errors is iterative. You teach the model back on what the human fixed, and next time it learns it a little better. You magnify that over a wide range of clients — not just your cute little Claude workflow you built in-house and now think can power everything for your family office, now and forever. (By the way — nice job forgetting to respond to that new potential investment while you were heads-down in Claude. This is one of the many reasons why outsourcing exists.)
(4) Tax document organization.Sorting, classifying, tying-out — strong.
Where AI does not work, today, in production:
(1) Unsupervised judgment on ambiguous data.The K-1 that ties to two possible entities. The transaction that could be a capital call or a fee. The trust that holds the LLC that holds the partnership interest. An AI will give you an answer. It will sometimes be wrong. In financial data, “sometimes wrong” is unacceptable.
(2) Net new structural work.Setting up a new entity in the data model. Mapping a complicated ownership graph. The first-time work, the unprecedented work — AI assists, humans decide.
(3) Anything that needs to be defensible to a CPA, an auditor, or the IRS. AI output without a human-reviewed audit trail is not a deliverable.
The pure-AI vendor problem
There is a wave of pure-AI vendors pitching family offices right now with some version of “fire your back office, we’ll do it with agents.” We’d be cautious about that pitch, and not just for the obvious reason that we run a company that takes the opposite view.
Deloitte’s 2024 Global Outsourcing Survey found that 83% of executives are already using AI in outsourced services, but only 25% report meaningful cost or quality benefits so far (Copernican Shift). That gap — between adoption and realized benefit — is exactly where the pure-AI pitches fail in production. The technology can extract a number from a document. It cannot, today, defend that number to your auditor, your CPA, or your principal at 9pm when something doesn’t tie.
The math that actually works is not “AI replaces the analyst.” The math that works is “AI replaces 70% of the analyst’s hours and the remaining 30% of human time gets reallocated to the judgment work and the exceptions.” That’s the White Glove × AI model. We’ll get into it in detail next post.
What this looks like on a P&L
If a family office is spending $400K a year on back-office accounting and operations today, the realistic AI-plus-human production model brings that to something like $120K–$180K — not zero, but materially lower — while improving accuracy and reducing key-person risk.
Like anything else, let’s think critically. Can we all go in together and share this $120K–$180K across a bunch of us who have the exact same AI × White Glove issue — and how low can we get that cost, then? We’ve done a lot of math. It’s really low.
The savings come from labor reallocation, not labor elimination, and they only show up when the AI is supervised by people who actually understand a partnership agreement.
The reframe
The economics of the family-office back office are real and getting worse, not better, under the current model. AI changes the math. It does not eliminate it. The combination of human judgment and AI throughput is what produces the actual operational improvement. The vendors that understand that distinction will win the next decade. The vendors that don’t will produce a lot of demos and very few production deployments.