The most expensive email your fund sends.

There is a particular silence that follows the discovery that a capital call notice went out wrong.

Not catastrophic. Usually the amount is right and something else is wrong — an investor whose KYC lapsed, a notice that was generated but never approved, a call where the notification was drafted and never switched on, so a third of the LPs got nothing at all.

Then comes the follow-up email. The one that begins "Please disregard our previous notice."

Nobody gets fired for it. It just quietly costs you something you can't measure: for a few weeks, every LP who received it treats your next communication with slightly more scepticism.

Capital calls are a sequence, and the notice is near the end

A call runs roughly like this: decide the amount, work out each investor's share, generate their notices, approve them, send, then record the money as it arrives and confirm receipt.

Most of the risk sits in a narrow window — the moment between the notices exist and the notices have gone out. Everything before that is internal and fixable. Everything after is in your LPs' inboxes.

So the agent lives exactly there, on the capital call's own page, and answers one question: is this call actually ready to send?

The four gates

1. Every investor's KYC is verified. Not present — verified. If one investor fails, the checklist stops. It does not carry on and tell you about the documents.

2. The notices have been generated. First a fund-level check: does this fund have a Call Notice template at all? If not, that single missing file blocks the whole call immediately, which is the correct outcome — no template means no notices for anyone. Then, per investor, do the generated documents actually exist? Commitments you've deliberately flagged to skip document generation are excluded, so an intentional exception isn't reported as a failure.

3. Those notices are approved. Generated and approved are different states, and the gap between them is where most near-misses live. A document can exist, look complete, and have never been reviewed by anyone.

4. An active email notification is configured on the call. This is the gate that surprises people, and it catches the failure that hurts most: everything is perfect, the notices are beautiful, and the delivery was never switched on. You find out from an LP asking why they haven't received their call — after the deadline has started running.

Only then does the agent confirm the call as ready to send and save its report.

⚙️ Under the hood: the checklist you can't configure

Here's the part I expect pushback on, and would defend anyway.

This agent has no settings form. Not "sensible defaults you can change." Nothing. You cannot turn a check off, reorder them, or make one advisory. Setting the agent up makes it available; it doesn't give you anything to tune.

That runs against everything enterprise software is supposed to do. Configurability is how vendors avoid arguments — ship the toggle, let the customer decide, and nobody has to defend a position.

We took the position instead, for a specific reason: every one of these four checks describes something that is unambiguously wrong.

There is no fund, anywhere, in any jurisdiction, that wants to send a capital call notice to an investor with unverified KYC. Or send a notice nobody approved. Or run a call whose notification is switched off. These aren't preferences that vary by house style — they're the definition of the call not being ready.

A toggle here wouldn't be flexibility. It would be an invitation, on a bad Friday, to make the red thing green.

Compare this with our distribution readiness gate, where the checks are individually switchable. That difference is deliberate: distributions vary genuinely between funds — some don't issue notices at all — so switching a check off can be a legitimate description of how that fund works. Here, it can't be. Same engineering, different rigidity, because the underlying risk is a different shape.

Stops at the first failure, and says so

Same principle as the distribution gate: the checklist stops at the first thing that fails. The remaining checks don't run and don't pretend to.

If KYC fails, you're not told about documents — because the documents genuinely weren't examined. A report that guessed at the rest would be more comfortable and less true.

Running it from the capital call's page gives you six live step indicators and a progress bar, updating as each gate clears. Watching it move is oddly reassuring in a way a final green tick isn't: you see the order, you see where it stopped, you know what to fix first.

And zero AI

No model calls anywhere in this agent. It needs no AI key.

Every gate is a yes/no comparison against records you already own. Is this KYC verified. Does this document exist. Is this notification active. There's no reading, no judgment, no summarising — so there is nothing an AI could contribute except uncertainty, immediately before a legally significant email goes to every LP in the fund.

	The mental pre-send check	The fixed pre-flight gate
Applied	When someone remembers	Every call, same order
Who can weaken it	Anyone in a hurry	Nobody — there's no toggle
Notification setup checked	Rarely, until it bites	Explicitly, as gate 4
Evidence afterwards	None	Saved report per call
Fund missing a template	Discovered at send	Named in a warning banner beforehand

The warning banner nobody asked for

One more detail worth stealing regardless of what software you run.

The agent's page permanently lists every fund in your entity with no Call Notice template uploaded.

Not as an error at send time. As a standing statement of a latent problem, visible before anyone starts preparing a call. It's the difference between discovering the missing template on the afternoon you planned to issue the call and knowing about it three weeks earlier when it costs nothing to fix.

Most operational pain in fund administration is like this: not a hard problem, just a problem discovered at the worst possible moment.

📊 The impact

Before: a pre-send check that depends on who's preparing the call, with no artifact proving it happened, and a notification setting nobody looks at until an LP complains.

After: four gates in fixed order, run from the call's own page with live progress, producing a saved report that names exactly where it stopped — plus a standing list of funds missing a call notice template, surfaced weeks before anyone needs one.

The error class removed: notices sent to unverified investors, unapproved notices sent, and calls issued with delivery switched off. The retraction email that follows any of those is the cost being avoided, and it's paid in credibility rather than currency — which is why it never shows up in an ROI model.

What to take from this

See it on your own structure

If you issue calls quarterly, try this on the last one: can you produce evidence — not recollection — that every investor's KYC was verified and every notice approved before it went out?

If not, DM me and I'll show you what the recorded version looks like on a call of your own shape.