Skip to content
RiskTemplates · The Daily Brief Sunday, November 15, 2026
Wire DOJ WHIP Whistleblower Rewards Put a 120-Day Clock on Internal Investigations OCT 20

Feature Business Continuity

When Fedwire or FedNow Is Down: A BCP That Assumes the Rail Returns

Fedwire and FedNow outage business continuity: settlement windows, a backlog triage rule, Operating Circular 6 Appendix B, and a BIA addendum.

By Rebecca Leung · November 15, 2026 ·
Table of Contents

TL;DR

  • On February 24, 2021, an automated data center maintenance process triggered during business hours knocked out all 14 Federal Reserve services. The Fed knew at 11:15 a.m. ET. Fedwire came back at 3:14 p.m. (Banking Dive).
  • Most payment BCPs assume the rail returns by the next window. Operating Circular 6, Appendix B plans for the opposite: participants who are not on the Protracted Outage Directory “will be unable to send or receive Messages,” and the Import Contingency Feature will not work either.
  • The gap to close is not a backup rail. It is a written order of battle for the backlog, agreed before the backlog exists, plus a BIA row that admits recovery is not yours to schedule.

The Fed knew at 11:15 a.m. and Fedwire was back at 3:14 p.m. In between, every payment ops desk in the country had to decide something its plan probably did not answer: what goes first when nothing goes.

The February 24, 2021 event took out all 14 Federal Reserve services. The root cause, per a Fed spokeswoman, was “an automated data center maintenance process that was inadvertently triggered during business hours” — normally an after-hours job. “This was human error.” Fedwire Funds, Fedwire Securities, and the National Settlement Service resumed at 3:14 p.m.; FedACH at 2:57 p.m.

Three hours and fifty-nine minutes. Not a catastrophic duration. Plenty long enough to matter.

The detail most people forget is the Fed’s instruction on the ACH backlog: “Please note that the backlog of files may take time to clear and do not resend files.” Do not resend. The single most obvious recovery instinct — push everything again — was the wrong one, and the only reason anyone knew that was a line on a status page.

What the rail actually is, in numbers

Fedwire Funds moved more than 835,000 transactions a day in December 2020, averaging $3.4 trillion daily, per Reuters reporting cited in the same piece. Settlement is immediate, final, and irrevocable. It is open 22 hours a day, 9:00 p.m. ET the preceding calendar day to 7:00 p.m. ET, Monday through Friday excluding holidays.

It is also expanding. The Board announced on October 9, 2025 that both Fedwire Funds and the National Settlement Service will run Sunday through Friday including weekday holidays, no earlier than 2028, at the same 22-hour daily window. Saturdays stay closed as a 26-hour maintenance slot.

Meanwhile FedNow, live since July 20, 2023 and 24x7x365, settled 4,997,811 customer credit transfers worth $274.7 billion in the second quarter of 2026 — about 54,921 a day at an average of $54,957, per Federal Reserve Financial Services data reported in August 2026. The October 2025 hours expansion deliberately left FedNow and FedACH untouched, because FedNow has no closing time to expand.

Two consequences for continuity planning, and they cut in opposite directions. Fedwire’s downtime window keeps shrinking, which retires the old assumption that overnight is always available for catch-up. And FedNow has no window at all, which means a FedNow outage has no scheduled reopening to plan around.

The part of this that is harder than a data center BCP

Data center BCPs are tractable because the failure is local. You fail over, you lose time, you restore. The rail is different in four ways that most templates do not model.

Recovery is on someone else’s clock. There is no failover you can execute. The FFIEC Business Continuity Management booklet is blunt that alternate arrangements are what you control: section V.D (p. 30) says the plan “should address alternate arrangements if payment systems fail,” with manual procedures “for calling in or faxing wire or automated clearing house requests to correspondent financial institutions.” The rail coming back is not in your column.

The fallback is shared, so it degrades first. When Fedwire is down, everyone calls the same correspondent. The manual path holds for the first twenty institutions and then the correspondent’s own wire desk saturates. A plan that lists a correspondent as the alternate without a volume ceiling is a plan that assumes you are first in line.

Designation is not automatic. Under OC 6 Appendix B §2.1, if the Reserve Banks declare a Protracted Outage, only participants they have designated can send Critical Payment Orders. Everyone else cannot send or receive. Appendix B §4.7 puts the burden of readiness on the designated participant: it “shall (i) have authorized persons available who are properly trained and have access to all information necessary” and “(ii) establish the capability to send and receive communications through the channels identified by the Reserve Banks.” Trained people and a live comms channel are contractual requirements, not nice-to-haves.

Compliance with the rules is on you. Appendix B §4.3: the Reserve Banks “are not responsible for identifying whether a Critical Payment Order complies with the restrictions” in §4.2 — receiver must be another designated participant, format per the Protracted Outage Procedures, Systemically Important Payment Activities only, within Submission Limits, Dollar Limits, and the Submission Window. The Reserve Bank will accept a non-compliant order anyway.

That last one is a control-design gift. It means a sloppy Critical Payment Order File gets accepted and settles, and the finding lands on you later.

Check your name against the directory first

Before writing anything else, answer one question: are you designated?

If yes, three artifacts already exist and need to be in the plan rather than in someone’s inbox — the Protracted Outage Procedures, the Protracted Outage Directory, and the Critical Payment Order File templates, which together form the Protracted Outage Toolkit. Confirm the current versions, confirm who holds them, and confirm the roster of authorized persons is accurate this quarter.

If no, the plan’s manual procedures only buy you a phone call to a correspondent that is itself trying to place orders. Write that down, because a BCP that implies otherwise is worse than one that admits the constraint.

The Fed’s own resiliency statement adds the piece everyone assumes is handled: reconciliation processes “identify and inform participants of transactions that may need resubmission if data loss is detected.” Identify and inform. The resubmission is yours. Certain large participants are contractually required to run at least one contingency test a year proving they can reconcile and resume — a useful benchmark for smaller ones that test annually on paper only.

The walkthrough: first four hours

ClockWhat happensWho decidesArtifact produced
0:00–0:15Fed status page or your access solution alerts; supervisors confirm it is the rail, not your instancePayment ops supervisorTime-stamped screenshot, incident number
0:15–0:45Freeze new outbound release except the tier-1 list; confirm inbound holdsPayment ops manager + treasuryHold decision, list of suppressed items
0:45–1:30Call the correspondent on the standing number; ask two questions — can you place orders, at what ceilingDesignated correspondent contactCall log with the ceiling answer
1:30–2:30Determine notification status; if a notification incident, start the 36-hour clock documentationCompliance + CISODetermination memo, even if the answer is “not yet”
2:30–4:00Client comms for anything missing a deadline today; no ETAs you cannot supportRelationship managementHolding statement, one per segment
4:00+Reconcile position: settled vs. queued vs. released-but-unconfirmedTreasury + accountingReconciliation worksheet, resubmission list

The 0:45 step is the one that gets skipped, and it is the only one that buys capacity. The two questions are deliberately narrow. “Can you place orders” is yes/no. “At what ceiling” gives you a number to triage against instead of a feeling.

Sample call-log line, because a narrative log will not survive review:

14:52 — Called M. Okafor, wire desk, Corr. Bank, standing number on file. Orders: yes, manual, ceiling 12/day over $250k, no ACH. Asked to confirm by email; received 14:58. Logged by J. Ruiz.

Write the backlog rule before the backlog

A triage rubric invented at hour three is a negotiation. One written in a calm quarter is a control. The rubric has to key off consequence of delay, not payment size — a $9,000 payroll batch can be more urgent than a $4 million intercompany transfer with a floating rate.

TierPayment typeRule during outageClient contact
1Payroll funding, same-day settlement obligations (securities, FX), time-critical closingsTo correspondent inside first ceiling slotSame day, proactive
2Vendor and tax payments with a dated penalty, loan funding with a rate lockQueue for next available slot; if the rail stays down past MTD, escalate to Tier 1Same day, on request
3Intercompany, internal transfers, discretionaryHold; no client contactNone
4Batch ACH and non-urgent disbursementsHold and suppress resend per Fed guidance on duplicate riskNone

Two rules under the table. First: never blanket-resend. 2021 taught that resending compounds the backlog and risks duplicates; reconcile to a settled position first, then resubmit only what is confirmed unsettled.

Second: every Tier 1 recipient gets a callback, not just a confirmation. The failure mode in a rail outage is silent queueing — a payment sits in a holding state, the client assumes it moved, and the miss surfaces a day later as a returned item.

The BIA addendum

The FFIEC booklet’s impact section (p. 11) treats RPO, RTO, and MTD as measurements to be “evaluated for alignment with third-party service providers’ contracted recovery expectations,” and says that where an RTO is not feasible, management should “verify whether the RTO is realistic, initiate an action plan and milestone(s),” and plan mitigation. Apply that to a rail you do not control and the honest output is a dependency row with a known ceiling.

Addendum fields, one row per rail, one row per payment type that uses it:

  1. Rail and access path — Fedwire via FedLine, FedNow, plus any indirect access through a correspondent or processor.
  2. Payment types riding the rail, with the MTD for each in hours, set by the consequence of a missed cycle.
  3. Designation status — on the Protracted Outage Directory, yes or no, checked date.
  4. Fallback — named correspondent, standing number, ceiling as last confirmed, and the date it was confirmed.
  5. Capacity ceiling — the manual path’s daily item limit at each fallback, in items and dollars.
  6. Suppression list — payment types that must not be blanket-resent, and the reconciliation step that precedes resubmission.
  7. Named decision owner per tier, with a backup who has actually released a Tier 1 payment in production.

Field 7 is where the key-person problem shows up. A backup name in a BIA is a guess until someone has run the control with a supervisor watching.

Proving it works

One tabletop per rail per year, run by the business continuity owner, not the payment ops manager. Inject at hour zero, then at hour two inject the correspondent’s ceiling at a third of what you assumed. The measured output is not “did we recover” — it is the time to produce the first Tier 1 list and the accuracy of the suppression list.

Evidence to retain: the scenario, the attendee list, the produced Tier 1 list with timestamps, every find logged as an issue with an owner, and the follow-up test date. A tabletop with a clean sheet and no findings is a rehearsal of the agenda, not of the event.

The Interagency Paper on Sound Practices to Strengthen Operational Resilience is worth reading for the framing even below its $250 billion asset threshold: firms are expected to “identify risks of third parties that provide it with public and critical infrastructure services” and to hold “processes to manage disruptions of these services” inside a board-approved tolerance for disruption. A rail is critical infrastructure in exactly that sense.

Three failure modes

The designated-but-not-ready participant. On the directory with a stale authorized-persons list and nobody trained on the Protracted Outage Procedures. Appendix B §4.7 makes that a breach, and the first notification window is a bad time to discover it. Test: produce the current roster and the date each person last exercised the procedure.

The one-rail shop with a second rail on paper. FedNow access held but never used for the payment types that would need it. During a Fedwire event the switch happens live, with no limit structure and no one who has done it. Test: run one deliberate FedNow cycle for a payment type that normally uses Fedwire.

The backlog disco. Everyone resends everything, duplicates land, and the weeks after the outage are spent on returns and client explanations rather than on the actual cause. Test: ask three people independently what the resend rule is; if the answers differ, there is no rule.

So what

Rail outages are short, rare, and disproportionately expensive — not because of duration but because they land on the payment types with the least slack. Every institution reading this has a runbook for its own data center and a paragraph about “payment systems” in the BCP that was written to satisfy a checklist item.

Four things this quarter, in order: confirm your directory designation and the current Protracted Outage Toolkit versions; write the call-log format and the two ceiling questions into the runbook; draft the triage rubric and ratify it at the next ops risk forum; then tabletop it with the ceiling degraded.

The output you want is a one-page addendum a new hire can execute at 2 a.m. without calling anyone. If it needs interpretation, it is not done.

The Business Continuity & Disaster Recovery (BCP/DR) Kit carries the BIA tiering by maximum tolerable downtime, the dependency mapping that flags a recovery time longer than the process RTO, and the test log that records whether each exercise actually met its targets.

Related: what the FFIEC BCM booklet requires in a BIA, the FFIEC BCM requirements mapped, and running a geopolitical stress scenario your board will accept.

◆ Need the working template?

Start with the source guide.

These answer-first guides summarize the required fields, evidence, and implementation steps behind the templates practitioners search for.

◆ Immaterial Findings · Weekly

Sharp risk & compliance insights. No fluff.

◆ FAQ

Frequently asked questions.

What does the Fed actually do during a Fedwire protracted outage?
Operating Circular 6, Appendix B governs it. The Reserve Banks declare a Protracted Outage and then process only a limited number of Critical Payment Orders from funds participants they have designated as Protracted Outage Participants, inside defined Submission Windows and Dollar Limits, for Systemically Important Payment Activities. Section 2.1 is explicit that participants who are not designated 'will be unable to send or receive Messages,' and that the Reserve Banks also cannot process Offline Messages sent through the Import Contingency Feature.
Does the Fed reconcile payments for us after an outage?
Partly. The Fedwire business resiliency statement says reconciliation processes exist to identify and inform participants of transactions that may need resubmission if data loss is detected. It also says certain large participants are required to participate in at least one contingency test per year to prove they can reconcile and resume processing after a recovery event. The reconciliation of your own general ledger, customer records, and correspondent position is yours.
Why does a weekend outage matter if Fedwire is closed on Saturdays?
Because FedNow is not. FedNow launched July 20, 2023 as a 24x7x365 service, and the Board's October 2025 expansion of Fedwire hours to Sunday-through-Friday explicitly left FedNow and FedACH alone. A Saturday event takes out the instant rail with no scheduled reopening to plan around, and the Fedwire Saturday downtime window is the only maintenance slot the industry has.
How long can a payment rail outage run before it becomes a notification event?
There is no rail-specific clock. Under the 36-hour computer-security incident rule at 12 CFR 53.3, the trigger is a notification incident — a materially disruptive event, which can include a third-party service provider's incident. A declared protracted outage of a rail you depend on is exactly the kind of determination a bank should be documenting as it happens, not reconstructing afterward.
What should the BIA say about a shared rail?
That recovery is not yours to schedule. The BIA should carry a dependency row for each rail, name the correspondent or access solution used, record the manual fallback, and state the maximum tolerable downtime for the payment type rather than for the platform. The FFIEC Business Continuity Management booklet, section V.D (p. 30), expects alternate arrangements and manual call-in or fax procedures to correspondent institutions to be documented in the plan.
Rebecca Leung

Author

Rebecca Leung

Rebecca Leung has 8+ years of risk and compliance experience across first and second line roles at commercial banks, asset managers, and fintechs. Former management consultant advising financial institutions on risk strategy. Founder of RiskTemplates.

◆ Related framework

Business Continuity & Disaster Recovery (BCP/DR) Kit

Set recovery priorities from a business impact analysis, catch vendors that recover slower than the processes they support, run and log tabletop exercises and tests, and report to the board on one page. Excel workbook, guide, tabletop kit, case studies and Word templates.

Immaterial Findings · Newsletter

The brief, in your inbox.

Enforcement of the week, a framework breakdown, and the prompts that are actually worth running. Delivered to your inbox. Free.