Running AI as a Portfolio of Monitored Bets

A leadership team approves an AI strategy, signs off a multi-year business case, and treats everything after as delivery. The assumptions underneath it keep moving: what the technology can do, what it costs, and what customers will want. Most operating rhythms have no way to notice when one of those assumptions has expired.

The business case is approved once, but its assumptions decay continuously

The gap between AI activity and AI value is now well measured. McKinsey's State of AI in 2025 reports that 88% of organisations use AI regularly in at least one function, yet only around 39% report a measurable effect on earnings, and only about 7% have fully scaled AI across the enterprise. [1]

Activity is not the constraint. The constraint is that a business case is a snapshot of assumptions taken on the day it was approved, and the organisation rarely returns to check whether those assumptions still hold. An AI strategy is built on assumptions with a half-life, and the job is not to set the strategy faster but to re-test it on a clock. [2] Organisational intelligence compounds when bets are monitored against their assumptions, not when tools are rolled out faster.

The design question for a board is therefore not whether to commit to an AI programme. It is whether the operating model can notice when one of its bets has stopped being true.

DBS treats AI as a portfolio of governed bets, not one wager

DBS, the Singapore-headquartered bank, runs AI as a portfolio rather than a programme. By early 2026 it reported more than 430 use cases in production supported by over 2,000 models, up from roughly 370 use cases and 1,500 models in 2024. [3][4]

That growth is the point. A portfolio of 430 live use cases is not a strategy that was set once and delivered. It is a book of bets that keeps moving, where some scale, some plateau, and the mix is never settled.

The value is spread across the book, not staked on a single deployment. DBS reports around S$370m of value from AI in 2023 [5] and S$750m in 2024, and set a target of more than S$1bn for 2025, a goal it framed in 2022. [4][6] It has since reported reaching about S$1bn of AI value for 2025 on its own benchmarked, not independently audited, basis. [5]

The structure matters more than any single number: value compounds across many governed use cases, so no one bet has to carry the case.

A leadership team copying only the headline number misses the mechanism. The mechanism is the portfolio shape and the guardrails that hold it together.

The centre owns the guardrails, the business owns the bets

DBS splits responsibility along a clear line: the centre sets the standards every use case must meet, and the business units own the individual bets. Every use case is assessed against a single common standard, the PURE principles, that AI should be Purposeful, Unsurprising, Respectful, and Explainable. [7]

Underneath that standard sits a risk-based governance regime, where oversight is tiered by how material a use case is. DBS describes an AI protocol and model registry, risk-based materiality tiering, senior accountability, and human-supervised oversight, with a senior, cross-functional governance committee setting direction across the bank. [7][8]

The centre also owns shared central platforms, one for data governance and one serving as the AI protocol and model registry, while around 700 data professionals organised as a Data Chapter are embedded inside the business units. [4][8] The standards live in the centre; the people who build the bets sit at the edge.

A federated portfolio puts guardrails at the centre and bets at the edge, with evidence from live systems deciding what scales and what is reversed. The reverse path is illustrated below by CBA, not DBS.

This shape is what lets a large book of bets stay governed without freezing. It is also what makes monitoring possible, because every bet is registered, tiered, and held to the same standard.

Monitoring catches a decaying assumption before it becomes a workforce decision

The reason monitoring matters is concrete: an unmonitored bet can scale straight into an irreversible decision about people. Commonwealth Bank of Australia shows what that looks like.

In 2025 CBA cut 45 roles in its Customer Service Direct business on the assessment that an AI voice-bot had reduced call volumes, citing roughly 2,000 fewer calls a week. [9] The Finance Sector Union said call volumes were in fact rising, with staff offered overtime and team leaders pulled onto the phones. [10]

On 21 August 2025 CBA reversed the redundancies and called it an error, stating that its initial assessment "that the 45 roles in our Customer Service Direct business were not required did not adequately consider all relevant business considerations and this error meant the roles were not redundant." [10][9] Affected staff were offered the choice to keep their role, redeploy, or take voluntary redundancy.

The lesson is not that the voice-bot failed. It is that a single bet's assumption, fewer calls, was treated as settled fact and acted on before anyone tested it against what was actually happening on the floor. Monitoring is the control that catches a decaying assumption while reversing it is still cheap.

What a board should decide before the next pilot scales

The portfolio shape does not require DBS's scale. A registry of bets, a single standard each must meet, and a clock to re-test the assumptions works with 10 use cases as well as 430, and the discipline matters more the smaller the team. The decision a board owns is whether scaling a pilot triggers a check on the assumptions that justified it, or whether approval is treated as the end of the question.

Key Takeaways

• Run AI as a registered portfolio of bets, not a committed programme. Require every use case to be logged, tiered by materiality, and assessed against one common standard before it scales, the way DBS assesses every use case against its PURE principles. [7]
• Put a clock on the business case. Name the two or three assumptions a major AI bet depends on, set a date to re-test them against live evidence, and decide in advance who can reverse the bet if an assumption has expired. CBA's reversal shows the cost of finding out late. [10]
• Read the full method. The half-life principle, the re-test cadence, and how to decide what to scale or stop are set out in The Half-Life of AI Strategy. [2]

Sources

[1] McKinsey & Company. "The State of AI in 2025: Agents, innovation, and transformation." McKinsey, 5 November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

[2] Hughes, Christopher. "The Half-Life of AI Strategy." cgh.dev, 2026. https://cgh.dev/thinking/half-life-of-ai-strategy/

[3] Tan, Aaron. "DBS rewires operating models for AI reasoning era." Computer Weekly, 9 March 2026. https://www.computerweekly.com/news/366639844/DBS-rewires-operating-models-for-AI-reasoning-era

[4] Singapore Economic Development Board. "How DBS, Southeast Asia's largest bank, is capturing the full value of AI and Machine Learning in Singapore." Singapore EDB, 29 September 2024. https://www.edb.gov.sg/en/business-insights/insights/how-dbs-southeast-asias-largest-bank-is-capturing-the-full-value-of-ai-and-machine-learning-in-singapore.html

[5] Forrester. "DBS Bank's Billion-Dollar AI Dream: Realized." Forrester, March 2026. https://www.forrester.com/blogs/dbs-banks-billion-dollar-ai-dream-realized/

[6] CNBC. "CEO of Southeast Asia's top bank DBS says AI adoption is already paying off." CNBC, 14 November 2025. https://www.cnbc.com/2025/11/14/ceo-southeast-asias-top-bank-dbs-says-ai-adoption-already-paying-off.html

[7] DBS. "Responsible and Ethical AI in Banking and Finance." DBS Bank, accessed June 2026. https://www.dbs.com/artificial-intelligence-machine-learning/artificial-intelligence/ethical-and-responsible-ai-in-banking.html

[8] DBS. "Responsible AI in banking: Gaining a competitive edge." DBS Bank, accessed June 2026. https://www.dbs.com/artificial-intelligence-machine-learning/artificial-intelligence/responsible-ai-in-banking-gaining-a-competitive-edge.html

[9] Bloomberg. "Commonwealth Bank of Australia Reverses Move to Replace 45 Jobs With AI." Bloomberg, 21 August 2025. https://www.bloomberg.com/news/articles/2025-08-21/commonwealth-bank-reverses-job-cuts-decision-over-ai-chatbots

[10] Sadler, Denham. "CBA reverses AI-driven job cuts, admits 'error'." Information Age (ACS), 21 August 2025. https://ia.acs.org.au/article/2025/cba-reverses-ai-driven-job-cuts--admits--error-.html