Only 23% of Malaysian CEOs Say AI Grew Revenue — Is Even That Too High?
Ask a CEO if AI grew revenue and 23% say yes. Attach an audit trail, and the number falls to single digits.
Note
Only 23% of Malaysian CEOs Say AI Grew Revenue — Is Even That Too High?
Ask a Malaysian CEO if AI grew their revenue this year, and 23% will say yes. Ask the same question with an audit trail attached instead of an opinion, and the number keeps shrinking — all the way down to single digits.
Four surveys asked a version of the same question this year, each with a slightly stricter bar than the last:
| Source | Who was asked | The actual bar | Result |
|---|---|---|---|
| PwC Malaysia | 37 CEOs, self-reported | “Did revenue grow, in your own view?” | 23% yes |
| PwC Global | 4,454 CEOs, self-reported | “Did revenue grow and costs fall?” | 12% yes, 56% got nothing |
| McKinsey | 1,719 executives, 97 countries | Audited EBIT impact ≥5% | 6% yes |
| MIT (NANDA) | Audited enterprise pilots | Measurable P&L impact | 5% yes |
The interesting part isn't that the number keeps falling. It's why it falls in exactly this pattern, every time someone asks more honestly.
The basics get skipped — usually on purpose, not by accident. PwC's own global chairman said as much at Davos: the companies getting nothing from AI mostly hadn't cleaned up their data or processes first. In Malaysia, only 30% of AI-using businesses can say who owns their AI initiatives. Skipping that isn't ignorance. Governance is slow and unglamorous; launching isn't. Someone chose speed.
AI gets bolted on, not built in — because redesigning a workflow means admitting the old one was wrong. McKinsey found real earners are 2.8x more likely to have rebuilt the process around AI (55% vs 20%). Everyone else drops AI into the process unchanged. That saves a person time — 80% report exactly that — but a P&L doesn't move until someone with authority over the old process agrees it was the problem.
The money follows what a board can see, not what an auditor can find. MIT's audit of live deployments found budgets chase visible, front-office use cases, while the real returns sit in the boring back office: $2–10 million a year on document review and outsourced support, $1 million on financial-risk monitoring. A sales chatbot shows up in a demo this quarter. A compliance tool shows up in next year's cost line, quietly, and by then it's hard to say whose win it was.
Nobody checks the output — because checking it means owning what happens if it's wrong. McKinsey's best performers use human sign-off on AI decisions at almost 3x the rate of everyone else (65% vs 23%). Over half of CFOs say they lack real authority over AI governance. Sign-off isn't a technical feature. It's a person agreeing to be accountable for a decision they didn't fully make.
One failure, four faces
Line these up and they stop looking like four separate problems. They're the same one, repeating: the number only moves when somebody in the organisation gives up a little control — the process owner who admits the old way was worse, the executive who funds the unglamorous project over the impressive one, the manager who puts their name on an AI decision instead of hiding behind it.
That's a harder ask than "buy more AI." It's also the only thing that shows up, independently, across four different studies that had no reason to agree with each other.
So before the next AI number gets quoted in your office — 23%, or any other — the sharper question probably isn't what AI changed. It's who in your organisation would have to lose a bit of control for it to actually work, and whether that conversation has happened yet.
Sources: PwC Malaysia CEO Survey; PwC Global CEO Survey via The Register; PwC Chairman at Davos, via Fortune; McKinsey State of AI 2026, via TechTimes and CX Today; MIT NANDA "GenAI Divide," via Legal.io; Deloitte Q2 2026 CFO Signals; Malaysia AI governance gap, ITbrief Asia.