Transfer pricing documentation has a reputation as a writing exercise: once a year, someone assembles a Master File and a stack of Local Files, updates the benchmarking, and files the package away against the day a tax authority asks for it. The reputation is out of date. The documents still follow the OECD Action 13 architecture of Master File, Local File, and CbCR, but the work behind them has become a data pipeline problem, and the pressure on that pipeline is rising from two directions at once.
The first direction is regulatory convergence on the same underlying data. Pillar Two obliges groups to align transfer pricing data, CbCR-type data, and GloBE information return inputs; a 2026 compliance checklist puts it plainly: align Pillar Two and CbCR data early, and check transitional safe harbors before assuming they apply. Amount B adds a scoping overlay for baseline distributors that has to be monitored annually as OECD guidance moves. A Local File that contradicts the CbCR, or a CbCR that contradicts the GloBE return, is no longer an embarrassment discovered years later; it is a cross-check a tax authority can run on receipt.
The second direction is filing mechanics. The documentation-on-request model is eroding jurisdiction by jurisdiction. Romania’s ANAF Order No. 828/2026 revised TP documentation rules for transactions from 2026 onward and requires large taxpayers to submit documentation electronically via SPV on an annual basis, rather than holding it available for an auditor who may never come. Annual electronic submission converts TP documentation from a contingency into a filing, with a filing’s deadline discipline.
What the automation actually does
The vendor market has settled on a consistent shape, and the consistency is informative. KPMG describes its engine as automating data collection, benchmarking analyses, review, and report generation in one place. EY’s solution populates OECD-compliant Local File templates through dynamic placeholders and generates the submission package from them. Deloitte and others describe equivalent generation workflows for global TP reports. Read together, the claims define the current automation boundary: first-draft generation, data extraction, benchmarking acceleration, and controlled assembly of documentation modules. None of them claim autonomous sign-off, and the omission is the honest part.
The pipeline behind those claims runs in recognisable steps:
- Source the data. Pull intercompany transaction data, trial balances, segment information, contracts, and entity metadata from the ERP and adjacent systems. Interfaces to ERP systems and databases are now standard product features rather than integration projects, which matters because manual data collection was historically where most of the hours went.
- Normalise and reconcile. Standardise master data, map legal entities and transactions, and reconcile TP inputs to statutory and management accounts. This is the step that decides everything downstream: a Local File built on transaction data that does not reconcile to the accounts is a defense file that impeaches itself.
- Benchmark. AI-assisted comparable searches, FAR analysis support, and margin computation compress work that consumed specialist weeks. Human review stays on method choice and comparability adjustments, where the analysis is argument rather than arithmetic.
- Draft. Populate jurisdiction-specific Local File and Master File templates from the structured data, using placeholder libraries and prior-year files as extraction sources. The draft inherits its numbers from the reconciled dataset, so a figure that appears in three documents comes from one place.
- Overlay Pillar Two and Amount B. Test whether the same dataset supports CbCR, the GloBE return, transitional safe harbor evaluation, and Amount B scoping for qualifying distributors. The overlay is where single-purpose documentation tools show their limits, because the value is in one dataset serving several regimes coherently.
- Package. Preserve workpapers, contracts, reconciliations, benchmarking outputs, and version history in an indexed defense file.
Consistency is the new deliverable
The deeper shift in 2026 is what the output is for. When documentation lived in a drawer, internal inconsistency was a latent risk. Now that authorities receive CbCR, increasingly receive the documentation itself, and will receive GloBE returns built from adjacent data, consistency across the package is the first thing an automated review on the authority’s side will test.
That reframes the automation business case. Faster drafting is worth having, but the differentiator is whether the system connects ERP data, benchmarking, Action 13 documentation, Pillar Two inputs, and Amount B scoping into one controlled process, so that a change in the underlying data propagates everywhere it should. A group that drafts its Local Files from one extract, its CbCR from another, and its GloBE workings from a third has automated the production of contradictions.
The transaction inventory deserves particular attention as the keystone dataset. Every documented intercompany flow should tie to the ledgers, to the relevant agreements, and to the policy that prices it. Where the inventory is reconciled and current, the rest of the pipeline is assembly. Where it is rebuilt annually from memory and emails, no drafting tool will save the file that sits on top of it.
Controls, versioning, and the defense file
Because the output of this pipeline is what the group hands a tax authority in a dispute, the engineering priorities look like audit priorities. Version control on documentation modules, so the file defended in 2029 is provably the file prepared in 2026. Controlled evidence: workpapers, contracts, and reconciliations indexed against the assertions they support. A recorded review trail showing which specialist reviewed which module, what changed between draft and final, and who approved release. Vendors increasingly build this in, and it is the feature set to weight most heavily in selection, ahead of drafting fluency.
Generated text needs its own control. A model drafting functional analysis narrative from prior files and contracts will occasionally produce fluent assertions the evidence does not support, and in a TP dispute an unsupported assertion is ammunition for the other side. The same rule that governs numbers should govern prose: statements of fact in the file trace to a source in the defense file, and the review step checks the trace on material assertions, not only the grammar.
The operational reality is that most of the programme’s cost sits before the drafting: entity master data, transaction mapping, reconciliation to statutory accounts, and the annual maintenance of benchmarking and scoping logic as OECD guidance moves. Groups that fund the drafting tool and starve the data layer get the characteristic 2026 failure: beautifully formatted Local Files, produced quickly, resting on numbers nobody can tie back to the accounts. The groups that get it right treat documentation as the visible output of a governed dataset, and the annual filing deadline, where it now exists, as the pipeline’s regression test.
FAQ
Can AI-generated Local Files be filed without a specialist review?
No, and the vendors themselves do not claim otherwise. The credible automation boundary in 2026 is first-draft generation, data assembly, and benchmarking acceleration, with method selection, comparability judgments, and sign-off staying with people. The practical gain is that specialists review a populated, evidence-linked draft instead of assembling one, which changes where their hours go rather than whether they are needed. A file that reaches a tax authority without documented human review is a governance failure regardless of its quality.
How should we prepare for jurisdictions moving from documentation-on-request to annual electronic filing?
Treat it as a deadline architecture change. Romania's 2026 rules requiring large taxpayers to submit TP documentation electronically via SPV annually are an early example: documentation that could previously be finalised when an audit letter arrived now has to be production-ready every year. That pulls data collection, reconciliation, and drafting into the annual compliance calendar alongside statutory filings, and it punishes processes that depend on a scramble. Groups with entities in such jurisdictions should schedule TP documentation like a filing, not like a contingency.
Where does Amount B fit into an automated documentation process?
As a scoping and policy overlay, not a separate workstream. For qualifying baseline distribution activities, the process should test entities against the scoping criteria annually, apply the simplified approach where it holds, and preserve the scoping analysis and its outputs in the defense file. Because OECD guidance continues to evolve, the scoping logic needs an owner and an annual review, and the version of the rules applied in each year should be recorded with the documentation it shaped.