A few hours without Fedwire, and the bank promises with no fallback

The week

On February 24, 2021, several Federal Reserve financial services, including the Fedwire Funds Service and FedACH, were disrupted for several hours during the business day. The Federal Reserve attributed the problem to an operational error, not a cyberattack, and services were restored later that day.

The outage drew wide attention because these services sit at the center of the U.S. payment system. Fedwire carries large-value, time-critical payments between banks. FedACH processes a large share of the country’s batch payments, including payroll and government benefits. Banks and their customers saw delayed wires and ACH files until the backlog cleared.

Payment rails run by a central bank are usually treated as a utility: always on, outside the institution’s control, and therefore outside its planning. The February 2021 outage was a short reminder that a utility is still a dependency. Nothing catastrophic happened. Many banks still found, during those hours, that they could not quickly answer three questions. Which customer commitments were now at risk? What fallback existed? And how would they reconcile once the rail came back? Those questions divide the problem into mapping, contingency and recovery.

Mapping: which promises ride on which rail

A bank makes many time-bound promises that depend on Fedwire or FedACH. Real estate closings, securities settlement, correspondent flows, business payroll, consumer benefit deposits and the bank’s own liquidity movements all run on them. Different business lines manage these promises, each with its own view of the payments platform.

When a rail stops, the urgent question is a business one: which customers and obligations are affected, in what order of urgency, and who must be told. Banks that had mapped important business services to the underlying rails could triage within minutes. Others depended on experienced operations staff knowing whom to call.

The map needs deadlines attached. A wire that must fund a closing by mid-afternoon is a different problem from an ACH credit due tomorrow. Same Day ACH adds its own settlement windows, so a delay of a few hours can push a credit a client expected today into tomorrow. With cut-off times on each commitment, an outage becomes a ranked list. The map also tells the bank which customers to call first, before they call in. A business client warned early that payroll may land late can tell its own employees. One that hears about it from them cannot.

Building the map is less work than it sounds. Payment systems already record originator, type, value and requested settlement time. A monthly extract that ranks clients by the value and time-sensitivity of their Fedwire and ACH activity gives most of the picture. Relationship managers add the context, such as a client’s payroll date or a scheduled closing. The map should be refreshed often enough that it reflects this quarter’s clients, not last year’s, and someone in operations should own it.

Contingency: alternatives are narrow, so procedures decide

For most institutions there is no practical alternative rail for Fedwire payments. The largest banks can use private-sector systems for some large-value flows. The Federal Reserve maintains offline procedures for urgent payments when its services are impaired. Neither is a full substitute, and the offline channel is built for a small number of critical payments.

That makes contingency mostly a matter of procedure:

• Eligibility. Which payments are urgent enough for an offline channel, decided in advance by value, purpose and deadline.
• Authority. Who may request offline processing, with what credentials and what verification, kept current as staff change.
• Client guidance. What the bank tells clients about adjusting their own deadlines, and who has authority to say it.
• Liquidity. How treasury manages the bank’s own position when incoming payments are also delayed, including end-of-day balances.

These decisions are hard to make for the first time during an outage and easy to make in a tabletop exercise. Credentials for offline channels deserve particular care, because they are rarely used and often found expired or held by someone who has left.

Banks that reach the rails through a correspondent or a processor have contingency partly in someone else’s hands. A bank should know how its provider will prioritize its payments relative to other clients when access is constrained, and how it will be told. Those answers belong in the contract and in the provider’s tested procedures.

Recovery: the backlog is its own incident

When the services returned, banks faced a surge of queued payments, files submitted during the outage, and possible duplicates where operators had retried. Reconciliation, exception handling and customer service volumes peaked after the outage ended.

Recovery is often the weakest part of payments resilience. It needs written rules for resubmission, so a payment sent once is not sent twice. It needs temporary staffing for exception queues, coordinated customer communication, and a way to confirm that every critical payment completed before end of day. Some recovery steps depend on the operator’s own instructions, such as extended operating hours or guidance on resubmitting files. The bank’s runbook should leave room for those and name who watches for them.

A bank that handles the outage well but the recovery poorly still lets customers down. Measuring recovery, such as the time until the exception queue is back to normal and the number of duplicate payments that had to be reversed, gives the next exercise a baseline.

Short, contained events are the cheapest time to test assumptions. Banks that wrote down what the February 2021 outage taught them were better prepared for the always-on payment systems that followed.

With hindsight

The outage did not change the structure of U.S. payments, but the years since raised the stakes of payments resilience. The Federal Reserve launched its FedNow instant payment service in July 2023, alongside the private-sector RTP network that had operated since 2017. Instant rails run around the clock, removing the overnight buffer that batch processing once gave operations teams. Fedwire moved to the ISO 20022 message format in July 2025, a large migration every participant had to test.

The federal banking agencies’ incident notification rule took effect in 2022, and supervisory attention to operational resilience grew. Mapping important services to their dependencies became an expectation rather than a best practice. Central payment infrastructure remains very reliable but not infallible. A bank’s exposure lies in its own commitments, procedures and recovery capacity, which only the bank can map and rehearse.