Ask an investor relations team why it takes six weeks to bring a new limited partner into a fund and the answer will be about documents. The subscription pack runs to eighty pages. The anti money laundering file for a family office holding through three trusts and an offshore feeder needs certified copies of things nobody can locate. The tax forms arrive incomplete, or signed by someone whose authority to sign them is unclear.
All of that is true. Almost none of it explains the six weeks.
The document handling part of investor onboarding has quietly stopped being difficult. What remains difficult is that no single person in most firms owns the investor from indication of interest to funded position, and the people who own the fragments have no shared clock. Onboarding is slow for organisational reasons dressed up as technical ones, and that distinction matters because the two problems have completely different fixes.
Extraction stopped being the hard part somewhere around last year
For most of the past decade, the honest constraint on onboarding really was the reading. Someone had to open a scanned trust deed, find the beneficial owners, work out whether the signatory block matched the authorised persons list, and rekey the result into a checklist. Someone else had to compare a completed subscription agreement against the executed version of the fund documents to see which representations had been struck through. This was slow, dull and error prone, and it consumed the time of people who were qualified to do more interesting work.
That work compresses well. Current models read a badly scanned deed, a certificate of incorporation in a second language and a partnership agreement with handwritten amendments, and produce a structured extraction with the source page cited for each field. We use Claude for this in client implementations, mostly because the failure behaviour is better than the alternatives we have tested: on a genuinely ambiguous ownership chain it will flag the ambiguity and show its reasoning rather than commit to a confident answer that a compliance officer then has to unpick. In onboarding, a model that returns "this reads as a nominee arrangement, and here is the reason that reading is uncertain" is worth considerably more than one that is right slightly more often but silent about the cases where it is not.
So the extraction is not the bottleneck. In the implementations we have run, the machine-readable part of a complex subscription file can be processed and cross checked in minutes. The file then sits and waits.
Six weeks of elapsed time contains about six hours of work
It is worth putting rough numbers on the shape of a typical onboarding, and these are illustrative rather than measured. Take an institutional LP subscribing to a mid-market fund, with a modest side letter request and a two-tier ownership structure. The hands-on effort across IR, finance, compliance and counsel might total five to eight hours. The elapsed time is thirty to forty working days.
The gap is almost entirely queueing. Broken into stages, on the same illustrative basis:
Subscription pack issued and returned: five to ten days, most of it the LP's own internal approval, and largely outside the GP's control.
Completeness and AML review: under a day of work, three to seven days of elapsed time, because it sits in a shared inbox.
First round of queries back to the LP: one day of work, five to ten days elapsed, because the queries go in a single batch after the full review rather than as they are found.
Side letter and counsel review: two to four hours of actual drafting, ten to twenty days elapsed.
Countersignature, administrator setup and capital call notice: a day of work, three to five days elapsed.
Reading that list, the temptation is to attack the biggest number. That is usually the wrong instinct. The expensive property of this process is not the length of any single stage but that each stage is a full stop: work halts, moves to a different function, joins a queue and waits for someone to notice it. Automating the review step so it takes minutes instead of hours changes the total by a rounding error if the output still lands in a shared inbox on a Thursday afternoon.
Nobody owns the investor, only the fragments
The structural issue is that onboarding crosses four or five functions and belongs to none of them. IR owns the relationship and the commercial expectation. Finance owns the commitment record and the capital call mechanics. Compliance owns the AML and KYC file and carries the personal regulatory exposure if it is wrong. External counsel owns the side letter. The administrator owns the register. Each of these functions is doing its job competently. What none of them holds is the elapsed time.
This produces predictable behaviour. Every function batches, because batching is locally efficient. Every function reviews the whole file rather than the changes, because nobody trusts an upstream summary they did not produce. Nobody chases, because chasing a peer function is socially expensive and there is no shared measure that makes the cost of not chasing visible. The LP experiences this as silence punctuated by requests for documents they believe they already sent.
The firms that have genuinely compressed onboarding have almost always done one unglamorous thing first: named a single accountable owner for the elapsed clock, with the authority to escalate across functions, and made the clock visible to the partnership. That is a governance change. It costs nothing in software and it is harder to implement than any extraction pipeline, because it means a partner has to accept that a number with their name on it will be discussed.
Counsel review is a real constraint, and also a negotiable one
The strongest counter-argument to all of this is legal review, and it deserves a fair hearing rather than a dismissal. Side letters create binding obligations, most favoured nation clauses propagate them across the LP base, and a mistake in a transfer restriction or an excuse right is not recoverable by apologising. Counsel are slow because being slow is the observable form of being careful, and because they are reviewing your document alongside eleven other clients' documents. No amount of capability inside the GP changes the queue inside the law firm.
That constraint is real. It is also less fixed than firms treat it. Most side letter requests fall into a small number of recognisable categories, and most of what counsel does on a first pass is determine which category a request belongs to and whether the drafting departs from precedent. A firm that maintains a properly indexed record of its own executed side letters, with the accepted and rejected positions marked, can send counsel a review that starts from "this request matches our standard position in nine places and departs from it in two, here they are" rather than starting from a blank page. The legal judgement stays with the lawyers. The classification work, which is most of the elapsed time and none of the value, does not need to.
Redesign the ownership, then the tooling earns its keep
A redesigned model is not complicated to describe. One owner holds the elapsed clock for each investor. Queries go to the LP continuously as they are found, not in a batch at the end of review. Each function reviews exceptions rather than whole files, which requires that the upstream extraction is traceable to source so the reviewer can verify a flag in seconds instead of rereading the document. Counsel receives a pre-classified package. The clock is reported to the investment committee alongside fundraising progress, because an LP stuck in onboarding is a commitment that is not yet capital.
In that model the technology does real work, but it is doing a different job than most firms buy it for. It is not replacing the reviewer. It is removing the reasons a file has to stop moving: the incomplete information that triggers a round trip, the untrusted summary that triggers a full reread, the unclassified request that triggers a blank-page legal review. Capability applied to a process that still has five full stops in it produces a faster version of the same six weeks.
A week is a governance target, and that is good news
The encouraging part of this argument is that the remaining constraint is one the firm actually controls. Extraction quality is bought. Model capability improves whether or not you do anything. Ownership design, escalation authority and the decision to make elapsed time visible are all within the gift of the managing partner, and they can be changed in a quarter rather than a funding cycle.
An investor whose first experience of the firm is a process that runs cleanly, tells them what is outstanding before they have to ask, and closes in days has learned something real about how the firm operates. That impression forms once, at the beginning of a relationship that is meant to last three funds. It is an odd thing to leave sitting in a shared inbox.
