Ask a fund CFO why the quarter takes six weeks to close rather than three, and the answer usually arrives with a name attached. The administrator was slow. The administrator missed the cut-off. The administrator sent the pack back with queries that should never have been queries.
It is rarely untrue. It is also rarely the main event. When we walk through an actual close timeline with a finance team, day by day, the time that goes missing is mostly not time spent inside the administrator's four walls. It is time spent waiting for the GP to answer something.
That is an uncomfortable place to land, because it moves the problem out of vendor management, which is someone else's to fix, and into internal design, which is yours.
The close is a chase, and the chase happens on your side of the line
A quarterly close in a private fund is not really an accounting exercise. It is an evidence-gathering exercise with an accounting exercise bolted onto the end of it. Before anyone books a number, somebody has to establish what the number is: what the portfolio company actually reported, which version of the management accounts is current, whether the add-on completed before or after the period end, what the revised debt schedule says following an amendment nobody circulated, whether the valuation committee's mark rests on the same forecast that finance has on file.
Administrators cannot do that work. They are not in the deal team's inbox, they were not on the board call, and they do not hold the side letter that changes how a particular fee is treated for one investor. So they ask. Each question becomes an email to a deal professional who is mid-process on something else, and each of those emails has a latency measured in days rather than hours.
Multiply it out. On an illustrative mid-market fund with twenty-two portfolio companies, and these figures are modelled rather than measured, a single close might generate somewhere between sixty and ninety queries. If two thirds of them resolve inside a day and the remainder take four, the queue alone consumes around two and a half weeks of calendar time. Nobody in that chain was idle. The administrator was waiting. The deal team was busy. The controller was chasing. And the close was late.
Exceptions are found late because nothing is looking for them early
The deeper issue is when exceptions surface. In most firms they surface during reconciliation, which is to say at the precise moment two numbers are placed side by side and fail to agree. That is late by design. The source documents that would have predicted the disagreement have usually been sitting in a data room, a shared drive or somebody's inbox for weeks.
The amendment to the credit agreement was signed in the second month of the quarter. The management accounts arrived with a quietly restated prior period and no covering note. The capital call notice went out on a figure that assumed one investor's commitment had already closed. Each of those things is knowable on the day it happens. None of them is known until the close forces the comparison.
This is the part that is now genuinely changing. Reading a large, inconsistent pile of source material and noticing where it disagrees with itself, or with the ledger, is exactly the class of work that capable language models handle well. Not the arithmetic, which was never the hard part, but the comparison: does this quarter's EBITDA definition match last quarter's, does the debt balance in the valuation model reconcile to the amended facility, has a portfolio company changed its revenue recognition without saying so in words a keyword search would catch. We build this kind of work on Anthropic's Claude because it holds long and badly structured document sets in view at once, and because it can point at the specific clause or line that triggered a flag. That last property matters more than it sounds. A controller who has to defend an exception to an auditor needs the evidence, not a verdict.
An exception with no owner is just a message in a thread
There is an organisational failure sitting underneath the technical one. In most finance functions an exception is not a tracked object at all. It is a message. It lives in an email thread, or a comment in a spreadsheet, or a line on a call agenda, and its status is whatever the person chasing it happens to remember.
This is why exceptions take days rather than hours. Not because the answers are hard, but because the questions have no owner, no due date and no visible state. The controller knows there are roughly a dozen open items. The CFO knows there are some. The deal partner whose answer is blocking three of them knows nothing at all, because the request arrived as the fourteenth email of a Tuesday afternoon.
Making exceptions first-class objects, each with an owner, an attached piece of evidence, a deadline and a status anyone can see, is unglamorous and does more for close timelines than most technology purchases. It also creates the record you will want later. When an auditor asks why a mark moved, or an LP asks why a figure restated between quarters, the answer either exists or has to be reconstructed from memory and email.
Exception-first means the close starts before the quarter ends
An exception-first close inverts the sequence. Rather than assembling everything and then discovering what fails to agree, you run the comparison continuously and let the disagreements queue up as they arise.
In practice, source material enters the process as it arrives rather than being gathered in the final fortnight. Board packs, management accounts, amendments, executed documents, notices, bank statements: each is read against what the fund already believes to be true, and anything that does not fit becomes a flagged item with an owner and the underlying evidence attached to it.
The effect on the calendar is not subtle. To put an illustrative shape on it, and this is a model rather than an observation from any particular client, if a fund currently spends eighteen working days on a close of which eleven are query resolution, pulling even half of those queries forward into the quarter itself removes five or six days without anyone working faster. The work did not shrink. It moved into a period when nobody is under deadline, and when the person who knows the answer is not simultaneously being asked for four other things.
The second effect is on the quality of the exception itself. A flag that arrives with the clause, the page and the prior-period comparison attached is a two-minute decision for a controller. A query that arrives as "please confirm the debt balance" is half a day of archaeology for somebody else.
Administrator quality genuinely varies, and that is a separate conversation
None of this argues that all administrators are equivalent. They are not. Service models differ substantially, the experience of the specific team allocated to your fund matters far more than the logo on the contract, and some administrators are materially stronger on complex waterfalls, multi-currency structures or fund-of-one arrangements than others. A firm that has genuinely outgrown its administrator should change administrator.
The test is straightforward, and most finance teams can run it from records they already hold. Take the last four closes and mark every day of delay against its actual cause: waiting on the administrator to produce, waiting on the GP to answer, or waiting on a third party such as an auditor, a lender or a portfolio company. If the first bucket dominates, this is a vendor problem and should be handled as one. In our experience it usually does not dominate, and the exercise ends up being quietly uncomfortable rather than vindicating.
A migration is also expensive, disruptive and consumes exactly the finance capacity that was already scarce. Changing administrator to fix a problem that lives inside the GP buys a year of transition and returns you to the same six-week close with a different name on the invoice.
The close becomes a state you are in, not a deadline you hit
The direction of travel here is not really faster closes, although faster closes follow. It is a different relationship with your own numbers.
When exceptions surface as they arise, the question "where are we?" stops requiring a project to answer. The valuation committee sees the issues that will affect a mark while there is still time to get the underlying information. The CFO can answer an LP query in the middle of a quarter without hedging about unaudited figures that may move. The administrator receives a clean pack and a short list of genuinely judgemental items instead of a pack plus ninety questions, which changes what you are paying them for and lets them do the work they are actually good at.
It changes the internal conversation too. A close that runs continuously does not need the quarterly scramble that consumes the finance function for three weeks in every twelve, and the people doing it spend their time on judgement rather than on the chase. In a market where good fund controllers are hard to hire and harder to keep, that is not a small consideration.
The firms getting this right are not the ones with the best administrator. They are the ones that stopped treating the close as something that happens to them at the end of a quarter and started treating it as something they maintain all the way through it.
