A CFO was sitting across from a customs officer, trying to explain why his system had charged the wrong tax rate on a set of transactions. His defence was simple: “The system was set up wrongly.”
— RMC audit encounter · Malaysia
The officer was furious. The CFO had nothing else to say.
That exchange is not unusual. And it is exactly the danger with AI in finance. If the system is wrong, the CFO still owns the output. A wrong model, wrong mapping, or wrong tax assumption does not become harmless because it now sits inside new software. It just becomes repeatable.
Section 01The e-Invoicing Misread
Everyone wants to talk about AI in finance right now. Vendors promise faster close, smarter forecasting, better dashboards, and cleaner reporting — especially as e-Invoicing pushes companies to revisit their systems and data flows. The opportunity is real.
But many CFOs miss the actual implication. They see AI as a tool for speed and e-Invoicing as a trigger for system change, but they forget that e-Invoicing was built for tax visibility and enforcement, not management convenience. Once the system is feeding structured data outward, bad logic becomes easier to spot, not harder to hide.
“Bad logic doesn’t disappear inside new software. It just becomes repeatable.”
The Chart of Accounts Problem
Many CFOs now see the chart of accounts as a strategic asset — something to improve segment reporting, profitability analysis, and AI-readiness. That instinct is right. But the chart of accounts is not only an accounting structure. It is also the place where tax starts.
When tax authorities review a business, they do not begin by reading every invoice one by one. They begin with the chart of accounts, management accounts, and how the business groups revenue and costs. That is where they look for patterns, margins, and mismatches.
If the COA has been redesigned for management reporting but not updated for current tax logic, SST scope, or transfer pricing consequences, the business has created a more elegant view of its own risk.
Tax is not like payroll software. Payroll changes often look like a rate update or a threshold change. Tax is a system made up of income tax, SST, transfer pricing, company law, accounting treatment, governance, and documentation obligations — and those pieces keep shifting. Many finance teams treat these as minor annual updates when they are structural changes with system consequences.
Transfer Pricing Is a Defence Tool
Transfer pricing is often treated as a yearly compliance document. It is much more than that. Transfer pricing is a defence tool. It explains who does what, who owns what, who takes which risks, and why profit is allocated the way it is.
But even saying “TP is a story with a rubric” is too soft. In an exam, words on paper may be enough. In transfer pricing, they are not. The story has to survive evidence. The authority is not only reading the TP report — it is checking whether the narrative matches contracts, organisation structure, payroll, margins, chart-of-account categories, management accounts, and e-Invoice trails.
If the evidence does not support the story, the story collapses.
Done properly, transfer pricing is a shield backed by evidence. Done badly, it becomes a self-inflicted wound — a polished document that locks the company into a weak explanation and hands the authority a script to attack.
Where AI Becomes Dangerous
AI can draft a TP report. But it cannot see the supporting documents unless someone gives them context and meaning. It does not naturally understand the ledger entries, internal contradictions, or operational nuance of one specific group. It only works with the prompts, answers, and source materials it receives. If the human feeding the system does not know the real insider story, the output will still be weak.
Building a TP system quickly surfaces the problem. Even after feeding it TP books, OECD material, and local guidelines, the machine still needs answers to a long list: who makes decisions, who controls risk, who owns assets, how work really moves across the group, what the commercial reality actually is. Each company is different. The quality of the TP output depends on the quality of the human story supplied to it.
The teams that get the most out of AI on transfer pricing are not the ones that prompt it best. They are the ones whose human insider knowledge is sharp enough to feed it well.
When Structure and Substance Part Ways
A useful example is the increasingly popular “licensed manufacturer” setup. One company handles procurement, another becomes the licensed contract manufacturer, and on paper the structure looks tidy for sales tax purposes. In reality, the same warehouse is still used, labour sits in questionable form, materials are booked away from the manufacturer, and the entity with manufacturing income ends up showing very little material cost.
The numbers no longer look like a real manufacturer. The label says one thing; the cost structure says another. For LHDN and RMC, that is low-hanging fruit. The financials no longer tell a coherent story, and the mismatch is visible.
Add e-Invoicing, analytics, and AI-enhanced reporting, and the inconsistency becomes easier to detect across entities and over time.
The Sequence That Actually Works
AI is not the enemy. It is not the saviour either. It is a loudspeaker. If the tax spine is strong, AI helps finance move faster and see more clearly. If the tax spine is weak, AI simply broadcasts that weakness in a more organised way.
For any CFO thinking seriously about AI, the sequence matters. Start by updating the chart of accounts with tax, SST, and TP consequences in mind. Make sure the operating structure matches the legal and tax story. Treat transfer pricing as an evidence-backed defence tool, not an annual writing exercise. Then layer AI on top of a finance function that can survive scrutiny.
Because in this environment, “the system was set up wrongly” is not a defence. And “AI generated it” will not be one either.
