“Show me the 5%.” Four words from a reviewer, and every transfer pricing team knows the feeling that follows. The policy says 5%. The question is whether anyone can show, quickly and in one coherent trail, how the books got there.
The paradox, stated plainly
Here is the transfer pricing paradox: a policy can be technically sound and the booked result can still be wrong.
The policy chooses the method, the tested party and the target return. It is usually reviewed by experienced people, sometimes by outside advisers, and it often holds up well on paper. But a policy is a design document. The actual result is produced later, by data feeds, spreadsheets, invoicing runs, journal entries and people working to a close calendar. Those are different things, owned by different teams, and they change on their own schedule.
In my 18+ years in transfer pricing, at EY and then about a decade in-house, most recently at Citigroup, the hard conversations were rarely about whether the method was defensible. They were about tying the numbers back together: showing that the policy in the report matched the data in the system, the charge on the invoice and the entries on both sides of the transaction.
That is the gap Transfer Pricing Lab exists to close. We bridge the gap between what your policy says and what your books show.
Why the gap opens
This isn’t only my experience. In Aibidia’s State of Transfer Pricing 2026 survey of 97 in-house transfer pricing professionals, mostly at large European groups, only 38% described their Operational Transfer Pricing as well defined and managed. 60% said they run it mainly on spreadsheets, email and manual processes, and half said last year’s year-end adjustments were moderately or very significant.
The gap rarely comes from one big mistake. It usually builds from ordinary business change that nobody connected back to transfer pricing. Five patterns come up again and again.
1. The business changes and the paperwork doesn’t
A reorganisation moves people and functions between entities. A new entity starts providing services. The intercompany agreement still describes last year’s arrangement, and nobody is assigned to notice.
2. The data moves underneath the calculation
An ERP migration remaps cost centres or accounts. The transfer pricing calculation still pulls from the old mapping, or from a report someone built years ago, and the cost base quietly changes.
3. The calculation lives in a spreadsheet
The markup changed in the policy but never reached the spreadsheet. A manual override was added one quarter and never removed. There are three versions of the file, and the one that was used is not the one that was saved.
4. The two sides book it differently
One entity records the charge in December and the other in January. One uses the month-end exchange rate and the other uses the average. The intercompany balance doesn’t eliminate, and the difference is cleared with an entry nobody can explain later.
5. The evidence is scattered
The approval is in an email thread. The data extract is on someone’s desktop. The person who knew how it all fit together has moved roles. Each piece may exist, but no one can produce the trail without a reconstruction project.
None of these means the policy was wrong. Each means the policy was not carried out the way it was written, and that needs a different kind of fix: ownership, data, controls and evidence, rather than a new benchmarking study.
Why this matters more than it used to
Tax authorities have been moving, steadily and publicly, toward testing what actually happened rather than what was written down.
- Conduct over contract. The OECD Transfer Pricing Guidelines (2022) say that where the economically relevant characteristics of a transaction are inconsistent with the written contract, the actual transaction should generally be delineated according to the conduct of the parties (paragraphs 1.42 to 1.45). The US regulations take the same approach: written terms are respected if they are consistent with economic substance, and in judging substance the greatest weight goes to the parties’ actual conduct (Treas. Reg. §1.482-1(d)(3)(ii)(B)(1)).
- Documentation that exists at the time. The OECD notes that contemporaneous documentation helps restrain taxpayers from developing justifications after the fact (paragraph 5.7), and recommends that the local file be finalised no later than the tax return due date (paragraph 5.30). In the US, documentation counts for penalty protection only if the principal documents are in existence when the return is filed and are provided within 30 days of a request (Treas. Reg. §1.6662-6(d)(2)(iii)(A)).
- Formula-based outcomes. The OECD’s Amount B simplified and streamlined approach sets returns for qualifying baseline distribution activities from a pricing matrix, and IRS Notice 2025-04 lets US taxpayers rely on it for taxable years beginning on or after 1 January 2025, pending regulations. When the target return is set by formula, the open question shifts from “what should the return be?” to “did the books actually deliver it?”
The stakes are real, though they depend on the facts. In the US, a net section 482 adjustment above the lesser of $5 million or 10% of gross receipts can bring a 20% penalty, and above the lesser of $20 million or 20% of gross receipts, 40% (26 U.S.C. §6662(e) and (h)). Good documentation is the main route to excluding adjustments from that test. A gap between policy and books does not mean a penalty will follow. It does mean the evidence you would rely on is weaker than it should be.
The six links between policy and evidence
The most useful way I know to find the gap is to follow one material intercompany charge through six links, and ask one question at each.
| Link | The question |
|---|---|
| Policy | What was intended, for which transaction, entities and period? |
| Agreement | What did the parties agree to do, and does it describe how they actually behave? |
| Source Data | Which records, accounts and mappings fed the result, and are they complete? |
| Calculation | How did approved inputs become the charge: which logic, version and adjustments? |
| Accounting | Where did each participant record it, in what period, accounts and currency? |
| Evidence | Can the team retrieve a coherent trail, with reviews and approvals, without reconstructing it? |
Northstar’s policy says its shared-service company earns cost plus 5% on management services. The year-end ledger shows 3.8%. Following the links backward: the policy and agreement agree; the calculation applied 5% correctly; but an ERP migration in the second quarter moved two cost centres out of the service cost pool. The markup was applied to a smaller base. The break sits in Source Data, and the evidence shows nobody owned the mapping change.
All six links have to work together for the transfer pricing to perform as it should. Finding the weakest one is usually more valuable than polishing the strongest.
From annual scramble to year-round control
Closing the gap is less about one big project and more about running transfer pricing as a business process all year. TPL frames the work in three steps.
- Diagnose: find the gaps faster. Trace one material flow across every entity it touches, and test the data, ledger, controls and evidence against the policy.
- Improve: build smarter processes. Assign owners to each link, fix the mappings, document the calculation, set exception thresholds and decide what approval looks like.
- Monitor: stay ahead of the risks. Check results against the policy monthly or quarterly, so an exception shows up in the period it happens and the year-end true-up is easier to explain.
A true-up may still be needed. That’s fine. The aim is that it confirms what you already knew, with the evidence already in place.
Where AI agents can help, and where they can’t
This is where I believe agentic AI can make a real difference, with experienced people still guiding the decisions. Within approved access and testing, an AI agent can help retrieve the current policy and agreement, check that a calculation’s inputs are complete, recompute a result and compare it with the defined rule, flag a bilateral mismatch, and assemble an evidence index.
What it cannot do is make an undefined process reliable. If nobody has written down the rule, the approved data source, the exception threshold and the approver, an agent will simply move the ambiguity faster. And people stay accountable for everything that carries judgment: interpreting a rule, resolving an exception, changing a price, approving a journal entry and responding to a tax authority.
So the order matters. Define the links first, then decide where assistance fits. Intelligent agents. Expert oversight.
This is a tax and finance problem, together
Operational Transfer Pricing doesn’t belong to one team. Tax sets the policy. Controllership books it. Shared services invoices it. IT owns the systems the data comes from. When each team only sees its own piece, the gap opens in the handoffs between them. In the same Aibidia survey, 68% of respondents put coordination between tax, finance and IT among their top three obstacles to Operational Transfer Pricing, second only to getting timely, good-quality data.
The teams that close it treat transfer pricing as a shared process with named owners at each link, and they bring tax and finance into the same room before close week rather than during it. That is the transfer pricing practice I want to help build, and if you work in tax or finance, I suspect it’s one you’d recognise.
Where to start
Pick one material, recurring intercompany flow. Try to locate all six links using the records your team would normally rely on, and mark each one Found, Partly, Not found or Don’t know. Note who owns the next step. That alone will tell you a lot. (Please don’t send confidential records through any public form, ours included.)
If you want a structured way to go further, the free 15-question Operational TP & AI Readiness Checklist covers ownership, data and calculations, accounting and monitoring, evidence and change management, and AI readiness. It is an educational self-assessment, not a tax opinion or certification.
And if the exercise surfaces a problem worth solving, the $999 Operational TP & AI Opportunity Scan helps prioritise where to look first, and the Operational TP Diagnostic & AI Readiness (from $10,000) tests one defined flow across all participating entities. Start with whichever step matches where you are.
If any of this sounds like your team, book a free 30-minute call. We’ll look at one intercompany flow, where it tends to break, and whether AI can help. You’ll leave with one practical next step, whether or not we work together. Book a free 30-minute call.
The policy is where transfer pricing starts. The books are where it has to hold up, every period, all year. Could your team show them the 5%?