Australia’s Batch Payments Delay: Why AI Still Wins

Australia processes an eye-watering $17.4 trillion each year through a batch-based payments system that most customers never think about—welfare, pensions, salaries, and bill payments. That system, BECS, was meant to have a clear retirement path with a target switch-off date of June 2030. This week, that target date was removed, pushing the transition further out.

Most people will read that and assume it’s a step backward for modern banking. I don’t. A delayed BECS shutdown is frustrating, sure—but it’s also a reality check: real-time payments infrastructure isn’t “better” unless it’s more resilient than the legacy rails it replaces. And that lesson matters a lot for banks investing in AI in finance.

Because here’s the thing about AI-driven banking: it thrives on real-time signals. But it breaks—fast—when the underlying payment rails are unstable, inconsistent across institutions, or poorly instrumented for monitoring and controls.

BECS isn’t “old”; it’s operationally dependable

BECS is more than 30 years old, and yes, it’s batch-based—payments often settle overnight rather than instantly. But its reputation is built on one trait financial infrastructure must never lose: reliability.

The goal of moving bulk payments to modern rails like the New Payments Platform (NPP) is sensible. Real-time settlement improves liquidity management, reduces customer confusion (“Where’s my money?”), and enables smarter products.

But the timeline shift makes a blunt point: a stable legacy system can be safer than a modern system with higher outage rates—especially when the modern system becomes the default path for critical payments like salaries and welfare.

From an AI in FinTech perspective, reliability isn’t boring plumbing. It’s the foundation for:

• Real-time fraud detection that can intervene before funds leave the account
• Instant risk scoring and step-up authentication
• Anomaly detection at the system level (outages, bank-to-bank routing issues, suspicious spikes)

If your payment rail can’t guarantee consistent uptime and predictable behaviour, AI models become harder to trust—and much harder to govern.

The real issue: migration isn’t just a technical program

A key reason the BECS end-date was removed is that the industry still lacks a shared agreement on what “good” looks like in the future account-to-account payments system.

This isn’t a detail. It’s the main story.

When different organisations oversee different rails, and when adoption of the modern rail isn’t uniform (especially among smaller institutions), migration becomes messy:

• Product teams can’t design around consistent capabilities.
• Ops teams can’t standardise incident response.
• Risk teams struggle to define controls that hold across all participants.

AI systems don’t like ambiguity. They need clean inputs, consistent process flows, and clear policy boundaries. A fragmented transition slows not only payments modernisation, but also the rollout of AI-driven financial services that depend on real-time rails.

Real-time payments are AI-ready—batch payments aren’t

Real-time payments and AI are a natural pair. Batch payments and AI can coexist, but they force compromises.

What AI can do with real-time rails

When payments are real-time, banks can use AI to make decisions at the moment it matters:

• Pre-transaction fraud scoring: block or step-up verify before the payment executes
• Behavioural biometrics and anomaly detection: match the payment to user patterns in milliseconds
• Dynamic limits and contextual controls: higher scrutiny for new payees, unusual geographies, or atypical times
• Payment message enrichment: AI-assisted categorisation that improves reconciliation and small business cashflow views

This is where AI in banking becomes visible to customers: fewer scams, faster transfers, cleaner bookkeeping, fewer “pending” mysteries.

What batch systems force you to do instead

Batch systems tend to shift intelligence from “prevent” to “detect and recover.” That’s not where you want to be, especially with scams.

With batch processing, you’re more likely to see:

• Fraud identified after a file is submitted
• Recalls and reversals that depend on timing and counterpart action
• Higher operational workload for exceptions

AI still helps—particularly for post-event detection and investigation—but the value ceiling is lower. Real-time is where AI is most effective.

The hidden risk: “modern” rails can concentrate outages

One concern raised in the broader discussion is that loading more volume and criticality onto newer rails—if they experience more frequent outages—creates systemic fragility.

That risk is very real, and it’s not hypothetical. Payments systems behave like utilities: once everyone depends on them, even short outages cascade into:

• missed payroll windows
• delayed welfare disbursements
• retail settlement issues
• call centre spikes and complaint storms

In late December, this matters even more. Australian businesses are closing books, running holiday payroll schedules, processing higher retail volumes, and dealing with end-of-year reconciliations. Payments reliability becomes a board-level issue quickly.

Here’s my stance: the industry is right to slow down if resilience isn’t proven. Speed is good; stability is non-negotiable.

What “resilience” should mean in a real-time payments future

If Australia wants an AI-ready payments foundation, the future rails must be built to fail safely.

Resilience should include:

1. Clear contingency paths (including fallbacks for critical payments)
2. Better participant-level monitoring (so one institution’s issue doesn’t become everyone’s issue)
3. Standardised incident playbooks across banks and payment operators
4. Stronger end-to-end observability—timestamps, message tracing, and consistent error codes

Those may sound operational, not “AI”. But they’re exactly what enables responsible automation at scale.

How banks can use AI now—without waiting for BECS to retire

A delayed BECS switch-off doesn’t mean banks should pause AI initiatives. It means they should focus on the parts that create value regardless of which rail carries the payment.

1) Build a cross-rail fraud brain

Banks often run detection rules and models separately for different payment types. That’s a mistake.

A better approach is a shared fraud intelligence layer that scores events across:

• batch payments
• real-time payments
• card payments
• digital wallet transfers

The model doesn’t need identical rails; it needs consistent features and feedback loops. This is one of the fastest paths to measurable reduction in losses.

2) Invest in better payment data (it’s the fuel)

AI in finance fails more often due to data problems than modelling problems.

Focus on:

• Normalising payer/payee identifiers across systems
• Improving payment message quality (clean references, categories, structured descriptions)
• Capturing decision outcomes (approved, blocked, recalled, disputed)

If you can’t trace a payment decision end-to-end, you can’t train models properly—and you can’t explain them to auditors.

3) Use AI for operations: outages, exceptions, and reconciliation

This is the underused win.

Even if customers don’t see it, AI can reduce operational drag through:

• Anomaly detection for payment file volumes, failure rates, and latency
• Auto-triage of exceptions (e.g., invalid account details, duplicate files)
• Reconciliation support for businesses and government agencies receiving bulk payments

In practice, this can reduce manual workload and shorten time-to-resolution during incidents.

4) Treat model governance as part of payments governance

Payments modernisation and AI governance should be tied together—same risk committee, same operational metrics.

A practical governance checklist:

• Can we explain why the model blocked a payment?
• Do we have appeal and escalation pathways?
• What happens when the rail is degraded—do models switch behaviour?
• Are we monitoring drift by customer segment and institution type?

If you’re building AI for real-time decisioning, governance isn’t paperwork. It’s what keeps you out of the headlines.

What to expect next: a longer runway, and better outcomes if used well

The industry is working toward a shared roadmap for the future of account-to-account payments, with clearer deliverables and milestones expected during 2026.

That timing matters. A lot of banks are setting 2026 budgets right now, and many are deciding where to place big bets: payments, fraud, digital identity, or generative AI programs.

My advice is to link them.

A modern, resilient payments rail is what makes AI-driven banking feel instant, safe, and personalised. But AI also makes modern rails safer by improving detection, monitoring, and operational response. They’re not separate transformation streams—they’re the same program viewed from different angles.

Batch payments will stick around longer than planned. That’s not a defeat. It’s a reminder that finance runs on trust, and trust is earned through uptime, predictability, and controls.

If you’re planning AI in finance projects for 2026, the question to ask your team is simple: Are we building models that assume the future is perfect—or systems that perform well even when the rails are stressed?