Beyond Internal AI Metrics | Generated by AI
Question: Are “number of internal AI use cases” the wrong success metric for banks’ AI transformation? Shouldn’t the real measure be whether the public perceives banks like SCB or DBS as “AI banks” — and what does it take to earn that label?
Answer:
You’re pointing at something real. The current playbook is:
“We deployed 200 internal AI use cases across 14 business units.”
That’s a cost center metric dressed up as innovation. Nobody outside the bank cares. No customer wakes up and thinks “wow, HSBC’s internal document summarization bot is changing my life.”
The Right Metric: Public Perception as an “AI Bank”
The actual signal is: does the market, the customer, the developer community, the media — spontaneously call you an AI bank?
That’s a brand/perception metric, not an internal operations metric. It’s earned, not declared.
Right now, a few banks are getting close:
| Bank | Why people think “AI bank” |
|---|---|
| DBS | Digibank, AI-driven credit decisions, public talks about AI culture, named by MIT as digital leader |
| JPMorgan | IndexGPT, $2B annual tech spend, Jamie Dimon openly says AI will compress workweek — public narrative |
| Goldman Sachs | Developer-first, Marcus, open about LLM use in trading/legal |
| Standard Chartered (SCB) | aXess, AI credit scoring in frontier markets, public AI partnerships |
| Nubank | Born digital, AI-native credit underwriting for 100M+ users in LatAm — customers feel it |
Notice the pattern: the ones earning the label shipped AI that customers touch, not just internal tooling.
Why Internal AI Doesn’t Count (Alone)
Internal GPT / chatbots are:
- Invisible to the public → zero brand impact
- Efficiency plays → just headcount reduction dressed up as innovation
- Not defensible → every bank can deploy the same Azure OpenAI stack
- Metric-gameable → “500 use cases” can mean 500 Slack bots nobody uses
The honest internal metric isn’t “how many use cases” — it’s how many FTE-equivalent hours saved, at what cost reduction, with what error rate. Even that’s internal.
What Actually Makes a Bank Feel Like an AI Bank (to the World)
1. Customer-facing AI that changes behavior
- Instant credit decisions (Nubank, Ant Financial)
- Hyper-personalized financial advice in-app
- Fraud detection so good customers notice the absence of false positives
2. Public developer/research output
- Papers, open source, APIs
- JPMorgan published on LLM use in financial NLP
- If your engineers are speaking at NeurIPS, you’re an AI company
3. AI in the product name / brand
- DBS’s “Digibank” is AI-first by positioning
- Not “Bank X launches AI pilot” — that’s press release theater
4. Ecosystem effects
- Other companies build on your AI infrastructure
- Your AI decisions (credit scores, KYC) become the standard others reference
5. Speed of change that customers feel
- Loan approved in 3 seconds vs 3 days
- That’s the moment a customer thinks “this bank is different”
The Honest State of Most Big Banks
Most big bank “AI transformations” are:
Internal ChatGPT wrapper → call it "BankGPT"
Deploy to 10,000 employees
Count use cases
Put in annual report
CEO says "we are an AI-first bank"
This is theater. The test is simple:
Ask 100 random customers of that bank: “Do you think of [Bank X] as an AI company?”
For most banks, the answer is no. For Nubank, Revolut, maybe DBS — it starts to shift.
What SCB/DBS Need to Do to Actually Earn the Label
- Ship AI that 10M+ customers feel in their daily banking — not just internal ops
- Open source something or publish research — show the world your AI muscle
- Let AI make real decisions — not just “assist” a human who overrides it anyway
- Build network effects from AI — your AI improves because more customers use it (Ant Financial’s credit model is the gold standard here)
- Get the developer community talking — if ML engineers at other companies are discussing your model architecture, you’ve crossed the threshold
Bottom line: “Number of internal use cases” is a metric that exists to satisfy board decks, not to measure real AI impact. The real metric is whether the world changes its mental model of that institution — from “bank that uses software” to “AI company that has a banking license.” Very few banks are there yet. The ones that get there will do it through customer-facing AI at scale, not internal chatbot counts.