A transfer pricing study is the document that proves the prices charged between related companies match what independent parties would have agreed. It is built in a fixed order: delineate the transaction as it actually happened, characterize what each entity does and risks, pick the tested party and the method, build and tie out the cost base, benchmark it, then crosswalk the file to the documentation rules that buy penalty protection. In the files we build, the economics are rarely the problem; most of the work goes into the supporting evidence. The steps below follow the order we use with clients.
The engagement I keep coming back to is a software group with its IP-owning company in a jurisdiction with no corporate income tax and a Delaware development company doing the build. The development company invoiced its costs plus a markup, the cost base grew into the millions by the second year, and every intercompany payment had to be tied back to an invoice after the fact.
Part 1: Planning the study, from conduct to tested party
Delineate the transaction as it actually happened
Before any method or benchmark, write down what flows between the entities (who provides what to whom, on what cadence, under which agreement, settled how), then compare it to what the ledgers and the settlement records show. Under Treas. Reg. 1.482-1(d)(3)(ii)(B), written terms count only if they match the economic substance, and the greatest weight goes to the actual conduct of the parties. In practice, a written agreement that the payments contradict offers little protection, and the absence of a written agreement does not by itself invalidate the pricing.
This client's first year had no executed intercompany agreement. We reconstructed the terms from conduct: the year-end invoice, the ledger behind it, and the settlements that paid it. The pricing was accepted, but without a signed agreement the position depended entirely on those records.
The planning output is a one-page pricing policy dated before the year opens: cost base definition, exclusions, markup, invoicing cadence, settlement terms, year-end true-up. Our guide to multi-entity structures for tech startups covers where these transactions usually arise.
Characterize the parties, then choose the tested party
The functional analysis asks three questions of each entity: what functions it performs, what assets it uses, and what risks it bears and controls. For a software group the asset question is the IP question, so I map each DEMPE function (development, enhancement, maintenance, protection, exploitation) to the entity that performs it and the entity that funds and controls it. The entity that owns the IP and bears market risk is the principal; the entity that builds to specification and is paid whether or not the product sells is a routine service provider.
That characterization decides the tested party. Under Treas. Reg. 1.482-5(b)(2) it is the party whose results can be verified with the most reliable data and the fewest adjustments, usually the one with no valuable intangibles. For a development company whose only revenue is the intercompany charge, a revenue-based indicator is circular, so the profit level indicator is operating profit over total costs, which in practice means a cost-plus markup.
Build the cost base before you open a database
A cost-plus study is won or lost on the cost base, and it is built from the ledger up: operating expenses by account; every exclusion with its reason (shareholder costs, owner draws, foreign exchange gains that are not a cost of the service); allocation keys for anything shared; then the chargeable base. The invoice has to equal that base times one plus the markup. Item 6 in the table below requires exactly this description, so the workbook is a principal document, not supporting material. Decide while the year is still open which records you will keep (payroll by role, vendor invoices, proof of delivery, proof of payment). Collecting them monthly takes far less time than reconstructing them after the year has closed.
Part 2: What the study has to include, and the evidence beneath it
The US penalty regulations do not leave the contents to taste. Treas. Reg. 1.6662-6(d)(2)(iii)(B) names exactly ten principal documents; the right-hand column shows how we prepare each one in practice.
| Treas. Reg. 1.6662-6(d)(2)(iii)(B) requires | What it is in practice |
|---|---|
| 1. Overview of the business and the economic and legal factors that affect pricing | What the group sells, how it earns, which entity owns the IP |
| 2. Organizational structure, including an organization chart | Entity register with ownership and effective dates; one chart per year the structure changed |
| 3. Documentation the Section 482 regulations explicitly require | The intercompany agreement and pricing policy; for services, the cost allocation method |
| 4. The method selected and why its conditions were met | Method selection memo: best method rule applied, tested party and profit level indicator named |
| 5. Alternative methods considered and why they were rejected | One paragraph per rejected method with the specific reason |
| 6. The controlled transactions, their terms, the internal data used; for cost-based methods, how costs are determined and allocated | The cost base workbook: opex, each exclusion with its reason, allocation keys, markup, invoice, settlement |
| 7. The comparables used, how comparability was evaluated, and any adjustments | Screening criteria, source log with filing dates, accepted set, rejection log |
| 8. The economic analysis and projections relied upon | Range computation, where the tested party lands, sensitivity tests |
| 9. Relevant data obtained after year-end but before filing | True-ups and timing differences between the invoice and the return |
| 10. A general index of principal and background documents and the recordkeeping system | The exhibit index, with a stated retention location |
Read top to bottom, the ten items are also a build order. Items 1 through 8 must exist when the return is filed and be produced within 30 days of a request; items 9 and 10 are the only ones you may finish afterwards.
The evidence layer the report sits on
The report has to include the exhibits it refers to. These are the supporting documents, in the order we prepare them:
- Cost base workbook. Opex by line, included or excluded with a reason, reconciled to the ledger, the invoice and the return.
- Payroll and vendor analysis. Every payee with role and location, every material vendor invoice catalogued, so each subcontractor cost is supported by an invoice.
- Settlement evidence. Every intercompany payment matched to the invoice it settled, with the paying and receiving account mapped to the entity that controls it. A bank statement does most of that job; for the client from the intro, which settled through accounts a bank statement did not cover, we built the schedule ourselves, and it took longer to prepare than any other exhibit.
- Proof of delivery. An invoice does not satisfy the benefit test. On this file we used the development company's source control history: commits by month, pull requests merged by quarter, a deliverables log tying each quarter to the periods invoiced. That proves work was performed more than it proves the recipient accepted it, so we now also ask for sprint sign-offs from the recipient, which show that it accepted the work.
- A signed management representation letter. Management's statement that the facts are complete and the conduct matched the documentation.
Canada and the UAE ask for the same file on different clocks
The ten-item list is American, but the other jurisdictions we file in want the same substance. In Canada, subsection 247(4) of the Income Tax Act requires records of the transaction, the parties, the functions and risks, and the methods considered by the filing-due date, produced within 3 months of a request, with a 10% penalty on large adjustments under subsection 247(3). In the UAE, Ministerial Decision No. 97 of 2023 adds a master file and local file above revenue thresholds, but the arm's length requirement applies at any size. In practice, we prepare one core file and add the local documentation each jurisdiction requires.
Part 3: Our experience, from a file that ran three years
What starting late costs
This client's first-year study was assembled after the return had been filed. Under the US rules that one fact matters more than any number in the report: the documentation safe harbor requires the principal documents to exist at filing, so year one cannot claim it, and penalty protection for that year rests on reasonable cause and the business records themselves. From year two, the pricing policy was adopted by resolution in January and the study was finished before the return went out.
The reconstruction is the cost. Rebuilding a closed year meant cataloguing a year of vendor invoices, matching payroll to the accounts it was paid from for contributors spread across many countries, and ending with dozens of exhibits behind a single controlled transaction. The stakes are set by IRC Section 6662(e): a 20% penalty where the price claimed is 200% or more (or 50% or less) of the arm's length amount, or where the net adjustment exceeds the lesser of $5 million and 10% of gross receipts, with a 40% tier above that. The 200% test has no size threshold, so it applies to a two-entity startup from its first intercompany invoice.
Benchmarking without a commercial database
Groups at this size rarely subscribe to a comparables database, and should assume the examiner has one. We say so in the study, then make the manual search as rigorous as a database one: screen listed IT-services companies from their own annual filings by industry code, services share of revenue, size and independence; convert operating margins to markups on cost; compute the interquartile range as the regulations define it; keep a rejection log naming every company dropped and why. Two things make the result defensible. Disclosure: a range described as directional support from public filings is credible, and the same range described as a database search is not. And the self-test the IRS documentation FAQ asks for: show how the range changes when each comparable is removed in turn, and disclose it if the tested party falls outside the range.
Quick answers about transfer pricing studies
What is a transfer pricing study? The analysis and documentation showing that prices between related entities, for goods, services, IP or financing, are what unrelated parties would have agreed. In the US it is organized around the ten principal documents in Treas. Reg. 1.6662-6.
When does a startup need a transfer pricing study? The day one entity pays a related entity in another tax jurisdiction for anything: development services, a license fee, a management charge, an intercompany loan. The US price-based penalty test has no revenue floor.
Does a transfer pricing study protect against penalties? In the US it is the condition for the reasonable-cause defense to the Section 6662(e) penalty, provided it existed when the return was filed and is produced within 30 days of a request. It does not prevent an adjustment, but it can prevent the 20% or 40% penalty that would otherwise apply.
My take, and what I tell clients
Treat transfer pricing as a monthly accounting discipline, not an annual report, and that comes from watching the alternative on the file from the intro: a year that closed with no agreement in force, no monthly invoices, and a cost base nobody had looked at since the ledger was posted. Rebuilding it took months, and the documentation safe harbor for that first year was no longer available.
The procedure we follow now is the consequence: the pricing policy is signed before January 1; intercompany invoices go out monthly against the actual cost base; settlements are reconciled to the ledger as part of monthly bookkeeping; a year-end true-up invoice brings the monthly invoices to total costs times one plus the target markup; and the study, the index and the representation letter are dated before the return is filed. The files we run pair a Canadian company with a BVI entity, a US company with an SVG entity, or a UAE operating company with an offshore principal, and the failure mode is the same across all of them: the economics were fine, the evidence was late. On the file from the intro, the year that took the most work to support was the first one, because its documentation was prepared after the return was filed.
One caveat. None of this makes a study examiner-proof. An examiner can still disagree with a comparable set or a cost allocation; the regulations promise only that a reasonable, documented position avoids penalties, not adjustments. The outcome depends on your facts, so this article describes the process and does not replace a study.
If you have two entities and an invoice between them, we will scope the file against your actual ledgers and settlement records. See what our tax compliance work covers for cross-border groups, or book a call and bring your intercompany agreement if you have one.