A series on AI and fund administration, part eight (and last): a single investor onboarding, told two ways, to make the earlier articles concrete.
The previous seven articles in this series have each made a general argument — about efficiency, governance, data protection, skills and scale. This closing piece makes the same argument a different way: by walking through one routine task, investor onboarding into a fund, as it would have looked five years ago and as it looks today. The fund and the investor below are illustrative composites, not any real Berkley client.
An investor submits a subscription application for a Caribbean-domiciled fund: a signed subscription agreement, a certified passport copy, a recent bank statement, and a short covering letter from their wealth manager. An administrator opens each document in turn and works through it by hand — checking the subscription amount against the application form, confirming the identity document matches the applicant's stated details, verifying the bank statement supports the source-of-funds declaration, and cross-referencing all of it against the fund's own register for consistency.
Each of these checks is straightforward individually, but together they take the better part of a day for one investor, longer if any document is unclear or a detail does not immediately match. Multiply that across a fund's subscription period, with dozens of investors submitting documents within the same window, and the administration team's week fills with this first-pass checking before anyone gets to the exceptions that actually need judgement.
The same documents arrive today. An AI tool reads each one, extracts the relevant fields — subscription amount, identity details, bank account information, source-of-funds indicators — and flags its confidence in each extraction, along with anything inconsistent across the document set. A person reviews that output rather than the raw documents: confirming the fields the tool extracted correctly, correcting the small number it did not, and investigating the one or two genuine inconsistencies it flagged, such as a name spelled slightly differently between the passport and the subscription form.
That review takes an hour, not a day, for a comparable investor — and the hour is spent on the parts of the task that actually needed a person's attention, rather than on confirming, line by line, that a passport number was typed correctly.
The sign-off did not change. No investor is admitted to the register without a person confirming the onboarding is complete and correct, and that confirmation is recorded exactly as it would have been five years ago. The judgement calls did not change either — a genuinely ambiguous document still goes to a senior administrator, not to the tool, and a source-of-funds declaration that does not sit right still gets the scrutiny it always would have. What changed is how much of the straightforward checking had to happen before a person's judgement was needed at all.
That is, in miniature, the argument this whole series has tried to make: artificial intelligence changes the shape of fund administration's working day, not its purpose. The reconciliations, the reporting, the onboarding — all of it still ends with a person accountable for the result, working within jurisdictions that take investor data protection seriously, at a firm senior enough to apply judgement rather than follow a script. What changes is how much of the routine work stands between that person and the parts of the job that actually need them.
Berkley Fund Services administers a focused number of Caribbean-domiciled funds on that basis, and its team is glad to discuss, with any promoter or manager who asks, exactly how a particular task is handled today.
This article is provided for general information purposes only and does not constitute legal, regulatory, tax or financial advice. Readers should seek advice specific to their circumstances before acting on any matter referred to above.