AI only matters when the economics move. Measure the capacity released, how much of it becomes real business value, and the full programme cost before treating productivity as ROI.
21,600 hours released. But what did they become?
Consider a company with 180 knowledge and office users in scope using AI in suitable recurring work.
If each person genuinely saves 2.5 hours a week across 48 working weeks, that is:
180 × 2.5 hours × 48 = 21,600 hours of gross annual capacity.
That is material. But it is not automatically a cash saving.
If 50% of the released time is actually redirected into useful work, realised capacity is 10,800 hours.
Even then, payroll has not necessarily fallen. Cash saving occurs only where overtime, outsourcing, hiring or other spend is genuinely reduced or avoided.
If the capacity helps the company handle more orders, the benefit is not the additional revenue. It is the additional profit after the costs of delivering those sales.
If the released time is simply absorbed by meetings, checking, idle time or lower-value work, the financial value may be close to zero.
What must management prove before scaling?
Ask for five things.
| Question | What management should show |
|---|---|
| What moves? | The cost, revenue, margin, cash, capacity or decision metric expected to improve. |
| What is the baseline? | Current time, error rate, throughput, cost or other measure before AI. |
| How will capacity be realised? | Overtime removed, hiring avoided, more profitable output delivered, cycle time reduced or another measurable use. |
| What does the programme really cost? | Software, integration, data work, training, human review, support, governance and change effort. |
| Who owns the result? | A business owner, finance check, review date and clear criteria to scale, redesign or stop. |
NIST’s AI Risk Management Framework calls for relevant metrics, benchmarks and ongoing performance assessment rather than relying on the fact that a system functions.[3] ASEAN guidance likewise emphasises data governance, evaluation and appropriate human oversight when AI is deployed.[4]
Test your own economics
The website’s Materiality Check — AI & Automation model lets management enter its own assumptions:
people affected × hours saved × working weeks × loaded hourly value × realisation rate, less programme cost.
The important input is not the most optimistic productivity gain. It is the realisation rate: how much released capacity actually becomes measurable economic value.
Keep the outputs separate.
Capacity value is not cash saving.
Revenue opportunity is not profit.
Unused capacity is not realised value.
A learning pilot can still be worth funding before the P&L moves. But it should have a capped question, a measurable baseline and a decision date.
The next phase deserves funding when management can explain how the technology changes the economics — and how that change will be measured after rollout.
Source notes
PwC Malaysia — 29th Global CEO Survey, Malaysia findings (January 2026). The survey included 37 Malaysian CEOs. PwC reported that 23% saw additional revenue from AI, 17% saw cost reductions and 26% reported an increased cost base over the prior 12 months.
Deloitte Southeast Asia — Asia Pacific CFO Pulse Survey, Southeast Asia findings (July 2026). Among 74 Southeast Asian CFO respondents, 46% reported pockets of AI use across their organisations and 12% reported extensive use.
NIST — Artificial Intelligence Risk Management Framework (AI RMF 1.0) and Measure guidance. Supports selecting appropriate metrics, benchmarks and ongoing performance assessment.
ASEAN — Expanded ASEAN Guide on AI Governance and Ethics: Generative AI. Supports data governance, evaluation, safeguards and human oversight in deployment.