A Global Asset Manager Catches Missing Custodian Files with Files.com Before the Trading Day Starts

A global asset manager runs strategies through mutual funds, ETFs, closed-end funds and UCITS vehicles for institutional and individual clients worldwide.

That product range has an operational consequence. The firm's funds trade against NYSE, Canadian, Irish, Luxembourg and UCITS market calendars, and every one of them runs on a daily cycle of files: NAVs, holdings, shadow accounting and trade data arriving overnight from between 60 and 100 custodians and data vendors. The firm's start of day depends on those files showing up.

A Missing File Makes No Sound

The firm ran its vendor-data operation on a file-triggered architecture. ActiveBatch, the firm's batch scheduler, watched the folder where vendor files landed and fired hundreds of jobs the moment the right file appeared. File-triggered architectures share one blind spot, and it is the one that matters most: a file that never arrives produces no event. No trigger fires, no job runs, and no error is thrown. The firm wanted the absence of a file to raise an alert on its own.

The stakes are not incidental. Start-of-day and end-of-day NAV and shadow accounting files from custodians carry regulatory deadlines, and the firm's investment decisions run on the same data.

The timing raises the bar. Custodian files land in the early morning hours, while the team is asleep, and a custodian's window for redelivering a file is limited. The firm wanted a miss caught inside that window, not at the start of the trading day.

A Second Job to Watch the First

Every available fix lived outside the layer it was watching. Detecting absence in ActiveBatch meant building a second job whose only purpose was to check whether the first one ran. The data engineering team had also built its own SLA trackers in Databricks, checking whether files had reached storage and opening tickets when they hadn't. And the firm wanted each feed to carry its owner and the vendor contact to chase, so a detected miss started with a notification rather than a lookup.

And any schedule-based check had to know when not to expect a file. Markets close on different days in each of the five geographies the firm operates across. Some close for half days. Some vendors follow calendars of their own that match no country's holidays. A naive check that expected a file every day at the same hour would be wrong dozens of days a year, paging the on-call rotation for files that were never due.

What the fix had to do was clear. Write down, per feed, what correct delivery looks like: which files, on which path, by when, in what shape. Evaluate every window against that definition. Know the difference between a miss and a holiday, across five market calendars. And put the alert in front of both the on-call engineer and the vendor while redelivery was still possible. The firm selected Files.com as the landing point for its vendor files, with Expectations, the platform's built-in delivery monitoring, carrying that requirement.

Delivery SLAs Defined Where the Files Land

Custodian and vendor files now land in Files.com, and the critical feeds each carry an Expectation: the folder and file pattern to watch, the deadline, and the shape the delivered files must take. When a window closes, Files.com evaluates what arrived. A file that never came is recorded as missing and opens an incident. A file that arrived in the wrong shape is recorded as invalid before any downstream job consumes it. The team scoped this deliberately, putting Expectations on the feeds that matter most rather than on everything.

Files.com custom Holiday Calendars carry the part no generic schedule could. The firm built calendars covering NYSE, Canadian, Irish, Luxembourg and UCITS market closures, half-day early closures, and the vendor-specific calendars that follow no national holiday schedule at all. An Expectation tied to a calendar simply does not open a window on a day no file is due, so a Dublin bank holiday no longer pages an engineer in the US.

When a window does fail, the alert reaches the on-call rotation, and a missed Expectation can email the responsible vendor directly, naming the file, when it was expected, and who to contact. The chase now starts with the vendor already notified.

The same evaluation validates shape on arrival. A vendor that sends a spreadsheet where a CSV was expected used to surface as a middle-of-the-night page from a broken load. Now it is flagged as an invalid delivery at intake.

Caught in the Vendor's Window, Not the Next Morning

With Expectations running against the production feeds, the firm turned a file that never arrives into a signal it can act on the same night.

• A missing custodian file is detected inside the vendor's own operating window, so a replacement can be requested and delivered before the start of the business day.
• False alerts on market holidays are gone. An Expectation tied to a holiday calendar expects nothing on days nothing is due, across all five market calendars the firm operates on.
• Malformed deliveries are caught and recorded as invalid at intake, before a downstream job runs against bad data.
• Delivery monitoring lives in the ingestion platform itself. Defining what correct delivery looks like is now a step in standing up a feed, not a second monitoring job to build and maintain alongside every feed that matters.

The File That Never Arrives Is Now an Event

For a firm where those files carry regulatory deadlines and the day's first investment decisions, that is the change Files.com made: delivery SLAs live in the ingestion layer itself, where a file that fails to arrive is something to act on, not a silence to discover the next morning.