Your tax office has better data
than your accountant.

A video came across my feed explaining how a tax bill went from 250,000 to 2,500 through transfer pricing. The comments were full of people asking for the playbook. Nobody asked the question I would ask first: when was that arithmetic last true?

Before I go further — I am a technology CEO, not a tax adviser. What follows is not advice on anyone’s structure. It is an observation about what AI does to the economics of enforcement, which is a business question, and it happens to be the one nobody in those comment threads is asking.

The audit used to be a lottery

Tax planning has always been priced on a probability. Not on legality — on detection.

Detection was expensive, because auditing meant a human being reading documents, comparing figures, requesting files and waiting for them. Expensive looking forces sampling: an authority examines a fraction of taxpayers, and everyone else prices their risk against that fraction. That is the arithmetic under every aggressive position ever taken. Not “is this defensible” but “what are the odds anyone opens the file.”

The whole industry that grew around this was, in effect, selling a bet on scarcity of attention.

AI collapsed the cost of looking

When looking becomes nearly free, you stop sampling. You check everyone.

China’s Golden Tax System Phase IV cross-references invoices, revenue declarations and bank records in real time, using big data and AI to surface inconsistencies across sources. On transfer pricing specifically, the authority pairs that analysis with cross-border information exchange — country-by-country reporting and the Common Reporting Standard — to flag high-risk related-party transactions automatically.

Malaysia is assembling the same foundation, and almost nobody discusses it in these terms. MyInvois is a continuous transaction control model: commercial transactions validated through the tax authority’s platform in near real time, 55 structured data fields per invoice, and from January this year individual e-invoices required above RM10,000 with consolidated invoices no longer accepted.

Strip away the compliance language and describe what that actually is: a complete, structured, machine-readable ledger of national commerce, sitting in one place. Whether or not heavy analytics are running on it today is almost beside the point. Datasets that exist get analysed eventually. That is the one reliable law in this field.

The audit is ceasing to be a lottery and becoming a search.

The reversal nobody announced

Here is the part I find genuinely striking, and it is a business observation rather than a legal one.

For a century the analytical advantage sat on the taxpayer’s side. Large firms fielded specialists, modelling and expensive advice against tax offices with limited headcount. Money bought better analysis, and better analysis bought defensible positions.

That asymmetry has quietly inverted. Your accountant holds your data. The tax authority is coming to hold everyone’s — yours, your suppliers’, your customers’, your competitors’ — in one structured format, continuously updated.

AI applied to a complete dataset beats AI applied to a partial one, every time. Your adviser can tell you whether your margin looks reasonable. The authority can see whether it looks reasonable against every comparable transaction in the country. For the first time, the better data is on their side of the table.

The beat nobody prices

Now the part that gets left out of every one of these videos: this runs backwards.

Tax authorities examine past years, and the records are already being stored. So the real question is not whether an arrangement survives today’s scrutiny. It is whether it survives scrutiny that has not been built yet, applied to data that already exists.

A structure created in 2024 gets assessed with 2029 capability. That is a long commitment made against a short-cycle assumption — the same mistake I have written about in an entirely different context. Anyone selling a strategy priced on the old detection probability is selling at a stale price, and the person holding it when the price corrects is not the seller.

My father’s advice on tax

My father put it to me in a way I have never been able to argue with. Paying tax is a responsibility. It is also proof of capability.

Sit with the arithmetic. Tax is a function of profit. You only pay a large tax bill if you built something that produced a large profit. The bill is not a punishment for success — it is a receipt for it.

And notice what the tax-scheme industry is quietly telling you: that your business is not strong enough to simply pay. Every hour spent engineering the denominator is an hour not spent on the numerator. I have watched owners work harder to avoid RM100,000 of tax than to earn another million — and the second is the easier problem, because nobody audits you for growing.

He gave me that as a moral position, years before any of this existed. AI has quietly turned it into the practical one.

To be fair

Three things this argument must not be allowed to become.

Legitimate planning is not evasion. Structures with real substance — people, functions, assets and risk genuinely located where the profit is booked — remain entirely proper, and they survive scrutiny precisely because they can be explained. Nothing here says pay more than you owe. It says the gap between what you owe and what you declare is becoming visible.

Enforcement capability is not enforcement judgement. A system that flags everything also flags a great deal that is innocent. More visibility will mean more queries into ordinary businesses that did nothing wrong, and the burden of answering falls hardest on the small operator with no tax department. That is a real cost, and it deserves to be named.

And AI cuts both ways. This is the half I find genuinely hopeful. Documentation quality that once cost a fortune from a large firm is becoming reachable for a small business. Historically the big company had professional defensibility and the SME had a shoebox — and the shoebox only worked because nobody looked. Now that everyone gets looked at, AI is the reason an SME can meet the standard at all.

The bottom line

Assume you are visible. Then spend accordingly.

Money that used to go into cleverness is better spent on two things: documentation that explains your position, and growth that makes the position worth defending. Get your records machine-legible, keep your intercompany arrangements explainable in plain language, and put the saved energy into revenue.

And if someone is selling you a structure, ask the question I ask about every adviser: what do they carry if it goes wrong? When the assessment lands with penalties, the seller is not on the invoice. You are.