101001101110100100001010010110010110110101101101001 bits & atoms
·
AI AdoptionBankingFinancial AnalysisStrategy

The Statutory Signal

AI adoption, audited. What Australian banks actually spend—not what they say—reveals who's building competitive advantage.

9,000+ Engineers

CBA now employs more engineers than Google, Canva, and Atlassian in Australia combined

Engineering headcount comparison showing CBA with 9,000+ engineers

  • CBA: 16% engineering density. One in six employees writes code. That’s a software company with a banking license.
  • NAB builds. Peers rent. +6.7% engineering FTE growth. Meanwhile, Westpac/ANZ hold headcount flat while external spend inflates.
  • Proprietary data demands proprietary talent. You can’t fine-tune models on customer data if your engineers work for Accenture.

Cash → Code

Capitalisation rate reveals who’s building IP vs. renting it

Scatter plot showing tech expenses vs build intensity across Australian banks

  • Westpac: $3.14bn in, 25% capitalised. Three-quarters evaporates as run cost. No durable asset formed.
  • CBA: 49% capitalisation. Every second dollar becomes owned IP on the balance sheet.
  • Macquarie: 15% build intensity. Deliberate exit from the asset game. (See next section.)

Westpac rents its intelligence. CBA compounds it.

The Counter-Bet

Macquarie exits the build phase

Dual-axis chart showing Macquarie's OpEx rising while CapEx additions collapse 54%

Note: Figures per Macquarie's software capitalisation policy.
  • Peers build assets. Macquarie buys outcomes. Deliberate shift from fixed IP to variable SaaS.
  • -54% CapEx additions (FY25). Software capitalisation collapsed from $391m to $180m. The build phase ended.
  • The wager: SaaS ROE > depreciating proprietary code. Asset-light in a balance-sheet industry.

The Outsourcing Tax

Freeze headcount. Inflate vendors.

Chart comparing internal FTE growth vs vendor inflation across banks

  • Westpac/ANZ froze hiring to cut costs. Demand didn’t disappear—it migrated to vendors. Result: +17-21% services inflation.
  • NAB insourced instead (+6.7% FTE). Productivity captured internally. External inflation capped at 3.5%.
  • Model costs fall. Integration costs rise. Without internal density, you pay someone else’s margin to deploy the tools.

The Capability Void

Everyone has fraud AI. The gap is everything else.

Matrix showing AI deployment across banks by category—fraud, marketing, coding, lending, infrastructure

  • Fraud AI is universal. Every bank has it (SaferPay, BioCatch). Regulatory mandate, not competitive advantage.
  • CBA is the only full-stack deployer. Verified AI in all five categories: Fraud, Marketing, Coding, Lending, Infrastructure.
  • The Void: ANZ and Macquarie have disclosed nothing beyond fraud. Westpac has a single lending tool. NAB is strong but has gaps.

2 Hours vs. 14 Days

What full-stack AI deployment enables

Log-scale comparison of lending approval times—CBA at 2 hours vs industry legacy at 14 days

  • AI compresses decision cycles. CBA’s lending AI (BizExpress) delivers 2-hour approvals—168× faster than manual processing.
  • Speed is the new pricing power. In broker channels, same-day response wins the deal. Multi-day wait loses it.
  • The gap is structural. Banks without production lending AI cannot match this velocity.

Ledger, Not Lobby

Auditing the receipt, not the press release

Side-by-side comparison of what banks say vs what they spend

  • Surveys measure intent. Filings measure action. Every “AI Readiness” score you’ve seen was based on what management said. This analysis is based on what they spent.
  • Source: audited statutory notes. Amortisation schedules. Capitalised labour. Vendor disclosures. Not interviews. Not surveys. Receipts.

This dataset doesn’t estimate AI adoption. It audits the receipt.

Sources: FY25 Annual Reports and statutory notes for CBA, NAB, Westpac, ANZ, Macquarie, Bendigo. Engineering headcount from LinkedIn Talent Insights and company disclosures.