The article was originally published by The Business Times
This budgeting season, chief financial officers are facing complex artificial intelligence investment decisions.
How capability is depreciating
AI depreciates human and organisational capability through two distinct mechanisms.
First, technology forces some skills into obsolescence. Accountants stopped doing manual arithmetic; spreadsheet mastery, long prized in finance, is now being handed off to a chatbot. People retain the skill but the business stops valuing it.
Disuse, in contrast, is a function of management, policy and company culture.
Cognitive offloading has always been a rational way for humans to preserve mental resources. But now, AI inadvertently facilitates the delegation of judgment, a quality that exists only as a by-product of doing the work.
When arithmetic (a terminal skill) was lost, nothing downstream suffered. But judgment is generative – and precisely what is required to govern the AI doing that same work.
The company’s own capital decisions may therefore be degrading its essential organisational capability.
In our work at The Global Institute For Tomorrow, we have seen this trend accelerate within leadership cohorts over the past year.
Writing skills, conceptual thinking, and sadly, even the ability to communicate clearly are in decline.
A cohort in 2024 toiled over a flip chart mapping a new business model in boxes and arrows. In 2026, the default is to ask a large language model (LLM) to visualise it for them.
Where once a non-native English-speaking professional struggled through articulating his or her ideas, thereby mastering business concepts and linguistic nuance, text and visuals are now generated in seconds.
Through polished outputs, LLMs give the appearance of competency so the human professional demurs.
Addressing this strange new dilemma through human-resources learning and development – budgeted as an operating expense – is unreliable and structurally vulnerable to the vagaries of the business cycle.
Traditional development spending cannot be capitalised because the firm lacks sufficient control over the benefits: Staff can leave after undergoing training. Indeed, AI fluency makes them more attractive in a competitive talent market.
Here lies the problematic asymmetry: AI hardware and software are capitalised, depreciated and reliably funded, while the human capability required to make them productive is relegated to discretionary expenditure – making it the first line item to cut when margins tighten.
S&P’s 2025 Voice of the Enterprise survey found 42 per cent of companies abandoned most AI initiatives that year. The mispricing and implementation failures of early enterprise resource planning roll-outs serve as a recent precedent: Similar mistakes in underfunding and process redesign should be avoided.
Companies cannot change accounting requirements but they can change how internal decisions are made and who owns the outcomes.
Governing AI as capital allocation
Like R&D, organisational development requires an extended payback time, produces uncertain returns and confers benefits that accrue across multiple years.
No serious company manages R&D as discretionary overhead; it is expensed in the accounts for compliance but governed as capital allocation.
When thinking machines work alongside humans, capability development deserves the same treatment. It must be managed through a capital allocation framework, and should not sit as an overhead owned by a function with no accountability for the asset it enables.
Organisational reinvention should produce durable firm-owned artefacts: verification standards specifying which classes of LLM output require human adjudication and deliberate mechanisms, such as for preserving supervisory capacity. Unlike training outcomes, these are codified, auditable and persist even if individuals leave the firm.
Governed as a capital allocation, this complementary investment in organisational reinvention should incur similar obligations: being tied to a named asset, producing defined artefacts, measured against a baseline, subject to a specific hurdle rate, and owned by an executive accountable for the technology’s return instead of sitting with the people function.
Depreciation is continuous, in professional skills as in hardware. In an AI era, it may be that human skills depreciate even faster.
The critical 2027 planning question is therefore not how much to spend on AI, but rather what degree of organisational investment is required, who owns it, and how to measure whether it worked.
Renée Chu is the Head of Programmes and Content at GIFT HK, where she oversees the Programme Management team and content creation. Her interests include business management, organisational culture, and female advocacy.
Thomas is an intern at the Global Institute For Tomorrow (GIFT), where he supports the team on programme delivery and research across leadership development, sustainability, political economy, and broader international development issues in Southeast Asia. He is currently in his final year, pursuing a Bachelor of Social Science (Honours) in International Relations at Taylor’s University, Malaysia.
