One Big Idea
Banking AI's largest peacetime experiment has gone backwards. That's not editorial framing. That's what the data says.
McKinsey reports that 94% of companies see no significant value from their AI investments. The MIT Media Lab puts the number at 95% for generative AI specifically. And in banking, US productivity has fallen 0.3% per year since 2010 — at the same time global banking technology spending grew 9% annually. We are running the most expensive technology rollout in the history of finance, and the productivity curve is pointing the wrong way.
The big idea this week is uncomfortable but useful: productivity is not profit. AI that helps a credit specialist finish reports 10% faster does not become revenue automatically. The employee just has more free time. Real return appears only when the workflow gets restructured around the new capacity.
This newsletter is about getting comfortable with that uncomfortable truth — and what to do about it.
The Insight
Most banks deploy AI onto existing workflows, call the gap between investment and impact "the AI productivity paradox," and wait it out. There is a J-curve at play, and the University of Tokyo's 2026 study on US bank AI adoption is honest about it: high-performing banks invest heavily in GPU, data scientists, and AI infrastructure now, taking earnings hits today to position for dominance later.
That framing is half-true. The J-curve is real. But it also gives banks permission to keep running unfocused AI pilots indefinitely without ever measuring anything, on the theory that "the value will show up eventually."
McKinsey's data quietly destroys that excuse. Banks that simply add AI to existing workflows capture 5% productivity gains. Banks that redesign their operating model around AI capture 15 to 20%. Three to four times the return for the same investment — and the difference is operating model redesign, not budget size.
The J-curve is real. The patience excuse is not. When the curve does not turn, leaders should redesign the workflow, not extend the runway.
Framework of the Week · The Boardroom Equation
This week's framework is the one I use whenever a banking CFO asks me to defend an AI program. Built on three pillars, one equation:
AI Deployment Cost − ( Pillar 1 + Pillar 2 + Pillar 3 ) = Net Value. If the result is not strongly positive, the architecture is flawed — not the budget.
Pillar 1 — Avoided Hiring. Planned headcount multiplied by fully-loaded cost per role, for every role AI removed from the hiring plan. This is the simplest pillar to measure and the one that ties most cleanly to an auditable line on the P&L.
Pillar 2 — Vendor Consolidation. BCG's 2025 finance-leader study (n=280) found that top-performing teams dedicate a separate AI budget AND require vendor reductions to be proven before continued funding — exactly the discipline applied to any major capital deployment.
Pillar 3 — Protected Risk-Adjusted Cash Flow. HSBC's AI-driven AML deployment with Quantexa delivered 228% return on investment over three years, with 60% reduction in case volumes and 2 to 4 times more suspicious activity detected (Forrester TEI 2025).
If your AI program cannot produce a strongly positive number through this equation, the architecture is flawed — not the budget. You do not solve an architecture problem with more money.
Use Case · Vietnamese Banking Benchmarks
Three Vietnamese banks already have proof points worth studying.
Techcombank has stated publicly that it intends to become "the first fully AI-operated bank in Vietnam." Sixteen percent of staff already specialize in data science or AI. The AI analytics platform contributes 100 billion VND in operating income directly. The eKYC platform processes 400,000 transactions per day with 99% success against deepfake attempts — driving 15% customer growth and 20% revenue growth (Asian Banker · 2025).
Vietcombank's VCB Digibot handled 88.5% of customer requests in its first six months of deployment — roughly 2 million successful interactions, replacing planned hiring in customer service while accelerating response times.
BIDV's private banking AUM grew 37% to 300 trillion VND in 2024 — driven explicitly by an AI-supported digital wealth advisory platform. HNWI clients grew 36%. Market share at 8%.
These cases map cleanly to the three zones: Risk Intelligence (Vietcombank), Knowledge Operations (Techcombank's underwriting AI), Decision Layer (BIDV's wealth advisory). Vietnamese banks do not need to wait for global benchmarks. They have their own.
Risk Note
Two risks worth flagging this week.
First, algorithmic coupling. If many banks adopt the same AI models for AML, fraud, and credit decisions, a single model failure could trigger correlated shocks across the financial system. The University of Tokyo 2026 study found measurable evidence of this risk in US banking. Vietnam's regulator (SBV) will likely require model diversity disclosures in coming circulars.
Second, AI cost savings erode to competition. McKinsey 2025: pure AI cost-reduction gains pass to customers, not bank shareholders. The competitive advantage is in HOW you deploy, not WHETHER. Treat AI as defensive necessity, not offensive moat.
Vietnam's AI Law 134/2025 takes effect March 2026. Plan compliance now.
Latest Video
This week I published the first video of The AI Banking Reality Check — an 8-minute executive briefing walking through three brutal truths about banking AI ROI and the Boardroom Equation framework.
Watch: youtu.be/KeUAIUH0P7k
The video drops links to every source cited above, plus chapter timestamps so you can jump directly to the section your board needs. Next week's video applies the Boardroom Equation retrospectively to a Vietnamese commercial bank's existing AI portfolio — showing how the framework reclassifies roughly sixty percent of typical AI spend as architecturally flawed without restructuring the whole budget.
Hit reply and tell me which of the three pillars is hardest to measure at your bank today. I read every response. Forward this to a banking executive who needs to see the data.
Was this forwarded to you? Subscribe to The AI Architect Letter — free, every Saturday.
Minh Tran · AI Business Architect · LinkedIn · Workshops & advisory: aibusinessarchitect.ai