Start with the work, not the model
The conversation about AI in fund operations usually starts with a tool. A better place to start is a workflow: who does it, how often, from which inputs, and what happens when it goes wrong. Capital calls, distribution notices, investor statements, reconciliations and quarterly reporting packs are all candidates — and so are the dozens of smaller tasks that quietly consume an operations team’s week.
Mapping that work first does two things. It shows where time actually goes, and it reveals which steps depend on judgment rather than repetition — a distinction that matters more than any feature list.
Three tests for any workflow
Before automating a step, we ask three questions:
- Are the inputs consistent? Structured data and standard documents automate well. One-off negotiations and bespoke side letters do not.
- Can a person review the output quickly? If checking the result takes as long as producing it, automation has moved the work rather than removed it.
- Is the cost of an error contained? Drafting a notice for review is low-risk. Posting an allocation directly to an investor’s capital account is not.
If a reviewer cannot check the output faster than they could produce it, automation is not saving time. It is moving it.
Workflows that pass all three are strong candidates. Workflows that fail any one of them may still benefit from AI — but as an assistant to a person, not as an automated step.
Where it tends to work
- Preparing capital call and distribution notices from data already in your fund administration system, for review before release.
- Reconciling cash, positions and investor balances across systems, and routing exceptions to the right person.
- Drafting first versions of quarterly letters, DDQ responses and investor updates from approved language.
- Extracting KPIs and covenant data from portfolio company reporting into a consistent format.
Where to be careful
Some work should keep a person firmly in charge. Valuation judgments, waterfall and allocation decisions, regulatory filings and anything that changes what an investor is owed all fall into this category. AI can still help — gathering inputs, flagging inconsistencies, preparing a draft — but the decision, and the sign-off, should stay with your team.
Make review part of the design
The firms that get lasting value from AI treat review as a feature, not an afterthought. Every automated step has an owner, a clear point where a person approves the output, and a record of what was produced and who approved it. That record is what lets a CFO, an auditor or an investor’s operational due diligence team trust the process.
It also makes improvement measurable. When you know how long review takes and how often outputs are corrected, you know which automations are working — and which should be redesigned or retired.
A sensible place to start
Pick one recurring workflow with consistent inputs and a clear owner. Measure how long it takes today and how often it needs correction. Automate the preparation, keep the review, and measure again after a quarter. The results will tell you where to go next far more reliably than any vendor demonstration.
For the method in detail — selection criteria, data boundaries, review roles, audit records and measures — see the Fund Operations AI Control Blueprint.
This article is general information and is not advice for any specific firm or situation.