Insights Contact

Payment Rails

ACH Payments Runbook: From Requirements to Live Support

By SP Sankar

⏱ 3-minute read · open “Read more” for the details

From the outside, ACH looks simple. A file goes to the bank and the money lands the next day. Inside the bank, even a small change touches many people: requirements, testing, compliance, security, release and support. This is how I take an ACH change from idea to live, safely.

ACH moves money between US bank accounts in batches. Think salaries and bills. This post shows how to build and run ACH changes safely, from the first idea to live support.

New here? Start with How a US Payment Works: 15 Steps Inside a Bank, then come back for the detail.

The lifecycle at a glance

Meet Nina, a product owner. Her team must launch same-day payroll over ACH. Here is the whole journey, from the first idea to live support.

  • Nine steps, from requirements to go-live.
  • Live support runs alongside, all the time.
  • In real sprints, several steps run at once.
ACH delivery: nine kinds of work, then live supportPhases overlap in sprints; hard sign-off gates come before productionPhase 1RequirementsPhase 2DesignPhase 3DevelopmentPhase 4Unit testingPhase 5PerformancePhase 6E2E testingPhase 7UATPhase 8ReleasePhase 9DeploymentLiveRun: L1/L2/L3Across every phaseCompliance · open source · access · certificates · software certification · alerts · disaster recovery
ACH delivery lifecycle
Read more: The lifecycle at a glance

Read the top row from left to right, then follow the arrow down to the second row. In real sprints, several boxes run at the same time. The gray box on the right is live support. The band at the bottom shows the controls that apply in every phase.

Agile, not waterfall

Nina’s team works in sprints, so the steps overlap.

  • One sprint can refine, build and test different stories.
  • Overlap the work, but never skip the hard checkpoints.
  • Money can’t be un-sent, so sign-offs matter.
Read more: Agile, not waterfall

These nine steps are the work that has to happen. They are not a strict one-by-one plan. In a Scrum or SAFe team, one sprint can refine the next stories, build the current ones and test the last ones, all at the same time.

LevelHow it works
StoryDefinition of Ready before work starts; Definition of Done (reviewed, unit-tested, automated tests green) before it closes
SprintSprint review with the product owner; performance and security checks automated in the pipeline
IncrementE2E (end-to-end: testing the whole journey) and UAT (testing by business users before go-live) on each releasable increment, often in parallel with the next sprint
Before productionHard gates: E2E (end-to-end: testing the whole journey) and UAT sign-off, bank certification when needed, CAB go/no-go (the final decision to release or not), penny tests

You can overlap the work, but never skip the hard checkpoints. Money can’t be un-sent. Sponsor banks and regulators will always ask for the sign-off evidence.

Planning vs delivery alignment

First, Nina marks the dates that never move.

  • Bank certification windows, Nacha rule dates, Fed holidays and change freezes.
  • Plan backwards from those dates.
  • Keep 15–20% spare capacity for surprises.
  • New work comes in only if something of the same size goes out.
Read more: Planning vs delivery alignment

In ACH, some dates never move: bank certification windows, Nacha rule dates, Federal Reserve holidays and change freezes. Put those dates on the calendar first. Plan backwards from them. Then check every sprint that you are still on track.

  • Keep one backlog. Link every requirement to its tasks, tests and release.
  • Plan with the team’s real speed, and keep 15–20% spare for support work and surprises.
  • Track what you need from the bank, vendors and other teams, with an owner and a date.
  • New work comes in only if something of the same size goes out, or everyone agrees to move the date.
MeasureWhat it showsHealthy sign
Say/do ratioCommitted stories delivered each sprint80% or more, steady
Feature predictabilityPlanned features delivered in the quarter80% or more
Scope changeStories added after sprint startLow, and always agreed
Milestone statusFixed-date milestones on trackGreen, or an agreed recovery plan

The biggest warning sign I see: the status stays green until the week before go-live. Report what is built and tested, not how busy people are.

The nine phases in one line each

Here are the nine steps, one line each.

  • Each step has a clear goal, an owner and a checklist.
Read more: The nine phases in one line each
PhaseWhat it deliversDone when
1. RequirementsSigned business needs, Nacha rule mapping, numbers for volumes and speedCompliance and business sign-off
2. DesignArchitecture, status model, cut-off and duplicate-payment protectionArchitecture and security review
3. DevelopmentReviewed code, NACHA file builder, PAYROLL and PURCHASE descriptionsClean scans and unit tests (tests of small pieces of code)
4. Unit testingProof of every rule: routing numbers, amounts, dates, file totalsCoverage gate met
5. PerformanceThe busiest day’s files built and sent well before the bank cut-offTargets met at peak
6. End-to-end testingHappy paths, returns, NOCs (notices from the receiving bank to correct account details), OFAC holds, fraud holds, reconciliationNo open critical defects
7. UATBusiness, operations and support confirm it works for themWritten sign-off
8. ReleaseChange approval, go/no-go, rollback plan, hotfix processCAB approval
9. DeploymentQuiet-window deploy, penny tests (tiny real payments to prove it works), two weeks of hypercareClean first windows

Exit criteria and sign-off: E2E and UAT

Before go-live, Nina needs signatures.

  • A step is done only when its checklist is met.
  • Share a test summary: results, open defects and risks.
  • No signature, no go-live.
Read more: Exit criteria and sign-off: E2E and UAT

A step is done only when its checklist is met and the right people have signed. No signature, no go-live.

Exit criterionE2E testingUAT
Coverage100% of requirements have a passed test100% of agreed business scenarios run
Pass rateAt least 98%; every failure has a plan100% of critical customer journeys
Open defects0 critical, 0 high0 critical, 0 high
Money movementCredits, debits, returns, NOCs and reversals 100% passedException queues proven by operations
ReconciliationFiles = settlement advices = ledgerReports confirmed by treasury
ReadinessBank certification tests passedRunbooks, L1 scripts, on-call rotas and customer emails ready
StageMust sign
E2E testingQA lead, engineering lead, ACH operations lead
UATProduct owner, business owner, ACH operations lead, compliance
Release (CAB)Release manager, change manager, product owner

How sign-off works: we share a test summary with results, open defects and risks. We sort the defects by severity, hold a short review, and store the signatures with the release. If something isn’t met, the product owner can ask for a written waiver with the risk, workaround, owner and fix date. If the waiver touches money movement or compliance, compliance must approve it too.

Internal and external audit

Auditors will ask how Nina’s team proved each step.

  • Keep evidence as you go.
  • Watch for manual fixes in the live database with no second person checking.
  • Then an audit means handing over evidence, not rebuilding it.
Read more: Internal and external audit

Keep evidence as you go. Then an audit means handing it over, not rebuilding it.

AuditHow oftenFocus
Internal auditRisk-based, usually yearlyControls designed well and working
Nacha Rules compliance auditEvery year, by December 31Compliance with the Nacha Operating Rules
External audit (IT controls)YearlyAccess, change management, reconciliation (checking that every record matches)
Regulatory examsExam cycleIT, BSA/AML, OFAC (the US office that keeps the sanctions list), consumer protection
Sponsor bank reviewPer agreementThird-party duties, limits, Nacha compliance

Here is an example auditors love to find. Someone in operations fixes a stuck batch by changing the payment status straight in the live database. There is no second-person check, no record of who did it, and the records may no longer match the bank or the books. The fix: remove direct write access, allow status changes only through the application with a reason and a second approver, and log every database change against a ticket.

The full PDF has complete internal and external audit checklists, and the step-by-step response for this example.

Controls that run across every phase

Some controls run through every step.

  • Compliance: Nacha Rules, including the 2026 fraud monitoring rules.
  • Open source: license checks and security scans.
  • Access, certificates, keys and software sign-off.
Read more: Controls that run across every phase
  • Compliance: Nacha Rules, the 2026 fraud monitoring rules, Regulation E (the US rule that protects consumers’ electronic payments), BSA/AML and OFAC, with evidence kept for audit.
  • Open source: license checks, security scans and a list of every software part in each release (a software bill of materials, or SBOM).
  • Passwords and access: long passwords, a block on leaked passwords, MFA everywhere, no forced 90-day changes (as NIST now advises), and an access review every three months.
  • Certificates and keys: the digital keys that secure the link to the bank (TLS, PGP and SSH keys) are tracked and renewed automatically, and the bank connection is re-tested after any change.
  • Software sign-off: release sign-off, vendor security reports and license renewals tracked in one list.

Penny testing: the last proof before customers

Right after go-live, Nina’s team sends tiny real payments to prove it all works end to end.

  • Never use customer money.
  • Label every test entry with the release number.
  • If a penny test fails, the rollout stops.
Read more: Penny testing: the last proof before customers

Test environments can’t prove that the real bank, the ACH operator and real accounts still work with your release. A penny test can. Right after go-live, send tiny real payments between company-owned accounts. Open the change to customers only when they pass.

Penny testAmountWhat it proves
PPD credit, Same Day window$0.01File built, sent, accepted and settled; credit posts
CCD debit, Same Day window$0.01Debit path, settlement advice, ledger posting
Payroll credit with “PAYROLL”, next day$0.01Company Entry Description and future-dated path
Incoming entry to your own account$0.01Inbound processing and posting
Reversal of a test entry$0.01Reversal logic and audit trail
  • Never use customer money. Fund tests from a separate account with a small daily limit.
  • Label every test entry with the release number, so finance can match it and reverse it.
  • If a penny test fails, the rollout stops. The feature stays off until it is fixed or rolled back.
  • Keep the evidence: file IDs, trace numbers and settlement advices.

This is not the same as micro-entries. A penny test checks your release with your own accounts. Micro-entries (credits under $1 marked “ACCTVERIFY”) check a customer’s account before you debit it.

Disaster recovery: files must still go out

If the main site fails, ACH files must still reach the bank before the cut-off.

  • No payment lost, and none sent twice.
  • Test failover and failback regularly.
Read more: Disaster recovery: files must still go out
WhatExample target or rule
Payment hub and databaseBack within 1 hour; at most 5 minutes of data lost
Bank connectionBack within 30 minutes, from a DR connection the bank has already certified
No duplicatesBefore sending from DR (disaster recovery: how systems keep running if a site fails), check which files the bank already acknowledged; send only the missing ones
Cut-off fallbackIf systems can’t recover in time, upload the last good file through the bank’s backup channel, under maker-checker
TestingMonthly restore tests, quarterly failovers, and a yearly full DR test with a real test file to the bank

The full PDF includes recovery targets for every component, a seven-step failover and failback runbook, and the DR test plan.

Running it live: L1, L2 and L3

Once live, support runs around the clock.

  • Monitoring alerts the L2 person on duty in Slack or Teams.
  • No answer in 10 minutes? The backup and the L3 lead are called.
  • Never roll back a file that has gone to the bank. Fix it with a reversal or a new entry.
Read more: Running it live: L1, L2 and L3
LevelWhoDoes
L1Service deskAnswers customers, checks status by trace number, follows scripts
L2ACH operations and app supportWorks rejects, returns, NOCs and holds; reruns files under the runbook
L3EngineeringFinds root causes, fixes code and data, ships hotfixes

Alerts climb by themselves. Monitoring posts to a Slack or Teams channel and calls the L2 person on duty. If nobody answers in 10 minutes, the backup and the L3 lead are paged. After 30 minutes, the incident manager joins. If a cut-off is at risk, leaders and the bank are pulled in. Customers get plain emails about holds, returns and delays, never with full account numbers.

Golden rule: never roll back a file that has already gone to the bank. Fix it with a reversal or a new entry, within the Nacha time limits.

Download the full runbook (PDF, 38 pages)

Every phase in detail, with planning vs delivery alignment, audit checklists, disaster recovery, exit criteria and sign-off templates, a full penny test plan, checklists, a RACI, severity levels, alert templates, hotfix steps, return codes and more.

Download the PDF (1 MB)

The files behind this runbook

Every phase of this runbook touches the same handful of files. Here is where each one shows up. For what is inside each file, with samples, see ACH Payments, File by File.

File or momentWhere it matters in this runbook
Client file arrives (NACHA, pain.001, pain.008, CSV)Files in, monitoring and L2 alerts
Gateway checks and pain.002Exit criteria: rejects and statuses tested
Outbound NACHA file to FedACH or EPNPenny testing and disaster recovery: files must still go out
Returns (addenda 99) and NOCs (addenda 98)Compliance controls and L2 daily work
Reversal batchGolden rule: never roll back, reverse within the Nacha time limits
Statements (BAI2, camt.053) and reconciliationAudit evidence and daily sign-off

More payment runbooks

My take

I’ve lived through what happens when an ACH batch overruns. A performance issue once stretched our batch past an hour. Outbound files sat stuck on the server instead of reaching the operator, payment statuses froze mid-cycle, and a set of payments dropped into the exception queue.

The recovery was painful. We updated statuses by hand in the database, replayed the stuck files and corrected timestamps so the records matched what had really happened. Every manual step was another chance to make a mistake, under pressure and close to a cut-off.

That night taught me that in ACH, time is a control. Since then I push for four safeguards on every ACH platform:

  1. A batch-duration alert that fires well before the cut-off, not after it is missed.
  2. File delivery checks: confirm every file has left the server and been acknowledged by the operator, rather than assuming it did.
  3. A tested replay tool instead of manual database fixes: re-send files and update statuses through an audited, repeatable process, with timestamps set by the system, not by hand.
  4. Reconcile before you repair: match exceptions against what was actually sent before anyone retries them, so no payment goes out twice.

Safe ACH delivery isn’t about one big test at the end. It’s small gates at every step, controls that never switch off, and a support model that knows exactly who acts before the next cut-off.

Working on a payments project?

I help banks and fintechs deliver real-time payments, ISO 20022, and AI-driven AML and compliance change, from strategy to go-live.

Connect on LinkedInSee how I can help