Pillar Two is a data problem wearing a tax costume

The 15% rate is the easy half. One clean, defensible number per entity is the hard half, and it is a finance-architecture job. I have designed that architecture at global scale, lived its translation layer, and I still run it end to end every month.

Public finances have rarely been under more strain: ageing populations, the accumulated shocks of a pandemic, wars and an energy squeeze, and commitments that have outrun what tax bases comfortably fund. That is the backdrop against which more than 135 jurisdictions agreed a floor under corporate tax competition: a 15% minimum effective rate, tested jurisdiction by jurisdiction, implemented through domestic law, and collected as a top-up where the floor is not met. It is not a verdict on tax planning, which remains part of finance's ordinary duty to a business and its owners; what changes is the floor beneath the planning, and the standard of evidence behind it.

The global minimum tax gets discussed as a compliance deadline. Underneath the rate calculation it is asking for something more demanding: entity-level data that a management team and a tax authority can both trust. I have built finance architecture with that underlying requirement before, and the tax rules, however carefully applied, cannot compensate for weak inputs.

A note on scope: I write from the finance seat, not the tax seat, and where something is my judgement rather than the rulebook, I say so. Tax owns interpretation of the rules; finance owns the data, controls and reporting spine that make the calculation defensible.

The argument in 60 seconds:

• Pillar Two is a data-architecture problem with a tax specification: entity-level inputs feed a jurisdiction-level 15% test.
• The transitional relief was extended in January 2026, but it runs once out, always out: the first year your data fails in a jurisdiction, that relief is gone there for the rest of the transition.
• The first filing season, June 2026, proved the point in public: emergency coordination on filing, and national extensions, because the data was not ready.
• Design the entity data model first and the tax calculation becomes a controlled process; bolt it on late and you build a shadow spreadsheet layer instead.
• The same foundations now decide what your reporting, your next transaction and your AI investment are worth.

The calculation is not the hard part

Pillar Two sets a 15% minimum effective tax rate for groups with consolidated revenue of EUR 750m or more in at least two of the four preceding fiscal years, worked out jurisdiction by jurisdiction, using a defined measure of income and covered taxes for each entity in the group's consolidation perimeter, including entities left out of it only on materiality grounds. Fall below the minimum anywhere and a top-up tax makes good the shortfall. And the net is already wider than the "multinational" label suggests: under the EU's directive and the UK's rules, a large purely domestic group can be in scope without a single foreign entity. The rules are complex but defined: a capable tax team can apply them consistently once the right entity-level inputs exist.

The catch sits one layer down, and it is worth being precise about it: entity-level inputs feed a jurisdiction-level effective-tax-rate and top-up calculation. The rules start from each entity's accounting income on the standard used in the group's consolidated financial statements, before consolidation eliminations and subject to defined adjustments, whatever basis the local statutory accounts were prepared under. So most entities' data has to pass through a translation layer before the 15% test can even be run. And where a country runs its own domestic minimum top-up tax, that computation may be required on the local accounting standard instead, which means the translation layer is not one translation per entity but potentially two.

What a group typically discovers in its first filing cycle is that the inputs are the entire project. The problem is rarely one system: the ERP keeps the entity ledgers, the consolidation system produces the group result, the tax-provision tool holds a third view, and the gaps between them are patched in spreadsheets. Getting a clean, defensible, entity-level number, on a consistent basis, out of that landscape is not a tax exercise. It is a data architecture exercise wearing a tax deadline.

Where entity numbers quietly go wrong

The translation layer is not an abstraction to me; for nine months it was my month-end. As European Financial Controller at Shionogi, the Japanese pharmaceutical group, I ran the month-end close, budgeting and group reporting for the EMEA entities into the Japanese parent, and their statutory consolidation: local books on local terms, group reporting on the group's basis, IFRS and local GAAP reconciled entity by entity, on SAP Business ByDesign. Multi-currency, multi-entity, and a parent nine time zones away that needed each European entity's number to mean exactly what the group basis said it meant.

That seat teaches you the thing Pillar Two now examines at scale: the translation layer is where entity numbers quietly go wrong. Not through error, usually, but through accumulation, a mapping judgement here, a timing difference there, each defensible alone, none documented together. A group that has never had to prove its entity-level translation discipline is about to find out how much of it lives in one person's head.

Designing for the challenge before it arrives

Towards the end of my time at GSK, my manager and I co-led, together with one of the Big Four firms, the redesign of the group's global and intercompany trading model as part of the wider SAP ERP implementation. The scope ran across the entire supply chain, drug discovery operations and the commercial functions, and the design principle was specific: build each entity's trading-profit position in the system itself, so it stands up to a transfer-pricing challenge in every country where a group operates.

A decade before the acronym existed, that programme posed the same underlying data question the 15% test now asks everywhere at once: can this entity's number, produced by the system rather than reconstructed at year-end, survive examination by a local authority. Show the entity's number, and show your working.

The design question underneath both

The part of that programme that mattered most was not the trading model itself. It was the reporting split behind it: one system, two audiences. Management reporting served the business users who ran the entities day to day. Legal-entity and statutory reporting served each country's regulatory and tax obligations. Both had to trace back to the same source data, so the two views never quietly diverged. Built correctly, that architecture is designed to outlast the programme that funded it: the numbers keep reconciling long after the consultants have left the building.

Pillar Two now joins them as a third standing audience on the same spine. It is not a project specification that tax hands over once; it is a permanent consumer of the close, refreshed every reporting cycle, with safe-harbour positions revisited year by year. A programme that treats it as a one-off filing build has misread what it is.

My Finance Director seat for Europe at GSK gave me the other half of the same view, from the outside in: profitability by legal entity, the statutory reporting obligation attached to each country, and restructuring decisions that had to keep that architecture intact as the business changed shape underneath it.

And I knew at first hand what the scrutiny it was designed for actually feels like. In an earlier GSK role I was joint finance lead, with my manager, on one of the largest UK transfer-pricing defences of its time, engaging HMRC directly alongside the group's Big Four advisers. It was a separate matter entirely from the trading-model redesign, but it is exactly the scrutiny that design principle anticipated: a defence holds when the entity-level positions were defensible from the start, not because they were argued well after the fact.

Transfer pricing had first crossed my desk years earlier still, at Chiron: an international business whose profit sat squarely in the product as it moved between entities and markets, the classic ground of the discipline as I first met it. The discipline prices what each entity actually does: the functions it performs, the assets it uses, the risks it assumes. Pillar Two keeps a version of that instinct in its substance carve-out, which shelters a routine return on real payroll and tangible assets, entity by entity, jurisdiction by jurisdiction; payroll data that usually sits in HR systems no one has ever reconciled to the finance ledger. And that is the shift worth naming: a discipline that concentrated where physical product moved across borders is now, through Pillar Two, in effect every large group's problem, whatever its profit is made of.

The systems thread, and why AI raises the stakes

There is a longer thread underneath this for me, and it is the part I would flag to any board weighing the programme. My first corporate seat, at IMS Health in 1999, a $1.4bn US-listed pharmaceutical market-information group, was spent on special projects for group finance: implementing Hyperion Enterprise for consolidation and Hyperion Pillar for budgeting and forecasting, with a revenue-recognition workstream underneath, capabilities that carry through to today's Oracle Cloud EPM. My first implementation, in other words, was a product literally called Pillar; twenty-seven years later the word has a second meaning on my desk. That was the era before the consolidation layer and the ERP layer had been made to talk to each other, and every seat since has run the same lesson at a different scale, from the GSK trading model inside a global SAP implementation to cloud ERP at Shionogi to the systems my own company runs today. The lesson has never changed: the value a business gets from any reporting programme is decided by the data architecture underneath it, not by the software above it.

That lesson now has a sharper edge, because AI consumes the same entity-level foundations. Clean, well-architected data compounds the value of every system above it; poor foundations quietly cancel the business case that funded the investment, exactly as a generation of ERP programmes discovered. That deserves its own piece, and I will write it separately. For a Pillar Two programme the implication is immediate: the entity data model you are about to build is not a compliance artefact. It is the foundation the next decade of reporting, and your AI ambitions, will stand on.

The same test, at every scale

The discipline did not retire when I left the corporate world. For the last eleven years I have stayed directly accountable for multi-currency, multi-jurisdiction reporting and control in my own consumer-products company across the US, UK and Europe, on a Far East supply chain: statutory accounts, US sales taxes that vary state by state, VAT in the UK and Europe, customs duty at import, and the platform's digital services tax arriving passed through as a fee line. Every obligation must be produced from the same set of books and survive examination by its own authority, every month, with my own capital behind the answer.

The test is the same in kind as the one Pillar Two now sets: one set of source data, several authorities, each owed a number that reconciles to the others. When I say entity-level discipline has to be produced by the system rather than reconstructed at deadline, that is not a recollection from 2011. It is what I did at month-end this month.

Readiness has quietly become exit value

The day an in-scope group completes on your business, your entities are inside its Pillar Two perimeter, and buyers now test entity-level data readiness in diligence. Private equity sees this from both sides. A fund vehicle generally qualifies for investment-entity accounting, so it is not required to consolidate its portfolio line by line, and the rules' deemed-consolidation test respects that: each platform is tested as its own group, on its own consolidated accounts. But the structure breaks exactly where the accounting does. A fund that does not qualify as an investment entity, a corporate holding company sitting above two platforms, a joint venture or a partially-owned parent all change the perimeter, and the merger rules aggregate prior-year revenues when groups combine, so a bolt-on can take a platform over the line in the year it completes.

In a deal seat, that turns into a checklist I would run before signing: the acquisition perimeter mapped against the threshold; the mid-year acquisition arriving on a different chart of accounts with no comparable prior-year data; the carve-out with no standalone statutory history; the ERP still running under a transitional service agreement; who holds the seller's data, and for how long; and what the platform's cash-tax forecast and lender reporting absorb after completion. For a sponsor, entity-level readiness is tested on the way in and priced on the way out.

If your group is below the threshold

Below EUR 750m there is ordinarily no GloBE filing and no top-up under the OECD scope test, and it would be easy to file the whole subject under someone else's problem. Three routes bring it to your door anyway.

Growth first: the two-of-four-years test means scope arrives shortly after the revenue does, and the data architecture takes longer to build than the revenue takes to grow. Then the standard cascades: once tax authorities receive entity-level evidence from large groups as routine, it becomes the benchmark in transfer-pricing enquiries generally, and auditors and lenders follow. And the machinery is already built: by May 2026 the OECD's central record listed 37 jurisdictions with a qualified domestic minimum tax or income inclusion rule applying from 2024, with more added since. Domestic law moves where treaties cannot, and in my judgement the direction is one way: EUR 750m is where the line sits today, not where it retires.

The decade ahead

This is no longer a rule waiting to happen, and the first filing season made the point better than any argument could. The first GloBE information returns, due within eighteen months of the first in-scope year-end, fell due on 30 June 2026 for calendar-year groups. Getting there took emergency plumbing: on 18 May 2026 the OECD published a common understanding on central filing because portals and exchange arrangements were not ready, HMRC set out a transitional approach accepting a qualifying central filing in place of a separate UK return, and France, Portugal and Belgium extended their local deadlines into the autumn. None of that was a failure of tax-technical work. It was a failure of readiness at the plumbing layer, in public.

The January 2026 package changed the timetable, not the exposure. The transitional safe harbour that lets many groups lean on country-by-country data was extended to fiscal years beginning on or before 31 December 2027, at a 17% transition rate for years beginning in 2026 and 2027, with a permanent simplified test to follow it. But the transitional relief runs on a once out, always out basis, jurisdiction by jurisdiction: fail to qualify in a jurisdiction in any year you are inside the rules, and you cannot return to it there for the rest of the transition, while the successor test is deliberately more forgiving and permits conditional re-entry. That distinction is the sentence I would put in front of a board. An extension buys time to build. It does not buy back a jurisdiction you have already lost, and whether you lose it is decided by data you either have or do not have.

The same package settled the American question. Groups headquartered in the United States can elect a deemed top-up of zero under the income inclusion and undertaxed profits rules for fiscal years from 2026, with the US minimum-tax regime running alongside; domestic top-up taxes elsewhere still apply to their subsidiaries, and earlier years remain governed by the pre-2026 rules and the safe harbours then available. Two systems, one floor. And while the top-up rules are carefully ordered so that one shortfall is collected once, there is still no dedicated binding multilateral mechanism for GloBE disputes: at the edges, parallel regimes and domestic top-ups computed on different bases leave the burden of unwinding an overlap with the group. The group that can show its number, entity by entity, reclaims faster and disputes less.

What this means for a Pillar Two programme now

Run as a tax project with a data appendix, a Pillar Two programme inverts the order that works: a specification is written first, and finance or IT is asked to source the numbers into it late, under deadline pressure. The data model, one entity, one number, tracing to the management view, the statutory view and the group reporting basis, has to be designed first, by whoever owns the group's reporting architecture, with the tax rules treated as a specification the system needs to satisfy. Three things then happen in a different order than most programmes assume:

1. The entity inventory and its data sources get mapped before the tax specification is finalised, not after.

2. One data model gets designed to serve every audience, rather than two systems bolted together once the gap is discovered.

3. The tax team pressure-tests real entity output months before the filing deadline, not the week before it.

The hardest single input, for most groups, is deferred tax at entity level, pulled from provision tools and legacy ledgers that were never asked for it. In practice I would build the whole thing in six layers, each with a named owner and an audit trail: the perimeter (every constituent entity, permanent establishment, joint venture and exclusion, kept current through deals); the accounting bridge (local ledgers to the group standard, with the adjustments documented); the tax data (current and deferred covered taxes, credits, elections, classifications); the calculation layer (jurisdictional blending, the substance carve-out, the safe-harbour tests); the control layer (reconciliations and evidence an authority can walk); and the outputs (the information return, local returns, the cash-tax forecast, board reporting and transaction readiness).

On an interim mandate the first ninety days have a defined shape: the entity inventory and data-source map in the first month; the accounting bridge and the tax-data gaps sized and owned in the second; real entity output in front of the tax team by the third, months before anything is due. It is worth naming what kind of work this is: a defined, cross-functional build with a hard external deadline, delivered while the permanent team keeps closing, forecasting and running the business, and an architecture decision that outlasts the programme. That is the shape of mandate groups bring in senior interim finance leadership to deliver.

Pillar Two relies on an existing foundation. It requires transfer-pricing-grade entity data, at filing speed, with nowhere left to hide a gap. And it needs more than transfer pricing ever asked of a ledger: covered and deferred taxes, entity classifications, ownership allocations, elections, payroll and tangible assets, all traceable.

In brief

Who does Pillar Two apply to? Groups with consolidated revenue of EUR 750m or more in at least two of the four preceding fiscal years, under the OECD's GloBE Model Rules, with top-up taxes applying from 2024 in early-adopting jurisdictions including the UK; under the EU directive and the UK's rules, large purely domestic groups are in scope too. The effective tax rate is tested jurisdiction by jurisdiction at 15%.

Does Pillar Two apply below EUR 750m? Not directly: below the threshold there is ordinarily no GloBE filing and no top-up. It still arrives through deals, growth and expectations. Acquisition by an in-scope group brings your entities inside its perimeter from completion, the two-of-four-years test means scope follows revenue quickly, and entity-level evidence is becoming the benchmark tax authorities, auditors and lenders apply generally. Groups expecting to cross the line, or to sell to one that has, build the architecture before it is mandatory.

Is Pillar Two really a systems problem, not a tax problem? The tax law is real and the mechanics are defined. What is hard is producing a clean, defensible number for every entity in the group, on a consistent basis, from a landscape of ERP, consolidation, tax-provision and spreadsheet layers that were built for one consolidated result. That is systems design, with a tax specification attached.

What is the "once out, always out" rule? The transitional country-by-country safe harbour, extended in January 2026 to fiscal years beginning on or before 31 December 2027, can only be claimed for a jurisdiction if it was claimed there in every earlier year the group was in scope. Fail to qualify once, and that jurisdiction needs full GloBE computations for the rest of the transition. Its permanent successor is more forgiving; relying on either is a data decision before it is a tax one.

Who should own the data architecture behind a Pillar Two programme? Whoever owns the group's reporting architecture, which in most groups means the CFO's office, not the tax function alone. Tax specifies the rules the system has to satisfy; finance designs the system that satisfies them for every audience that needs the number.

What goes wrong when groups get the order backwards? A shadow layer of spreadsheets appears alongside the ERP, rebuilt every filing cycle to patch data the system was never designed to hold at entity level. It is expensive, fragile, and recreates the manual reconciliation burden a properly designed system should have removed.

The first useful conversation about a Pillar Two data build is usually about one entity, one number, and the system that has to support it. I take those conversations directly: get in touch.

Discuss this piece on LinkedIn: linkedin.com/in/jatinderpurewal · More Insights

References

• OECD, Tax Challenges Arising from the Digitalisation of the Economy: Global Anti-Base Erosion (GloBE) Model Rules (Pillar Two), December 2021 - the 15% minimum effective rate, jurisdictional blending, the EUR 750m consolidated-revenue threshold and the two-of-four-years test.
• OECD, Pillar Two administrative guidance and commentary - the GloBE income and covered-taxes definitions on the consolidated accounting basis, and the substance-based income exclusion.
• HM Revenue & Customs - the UK's Multinational Top-up Tax and Domestic Top-up Tax, applying to accounting periods beginning on or after 31 December 2023, including purely domestic groups meeting the threshold.
• OECD/G20 Inclusive Framework, GloBE Information Return and administrative guidance - the standardised return, due within 18 months of the first in-scope fiscal year-end (30 June 2026 for calendar-2024 groups), and the common understanding on central filing published 18 May 2026.
• HM Revenue & Customs, transitional approach to GloBE Information Return filing and exchange - acceptance of a qualifying central filing with an on-time overseas return notification in place of a separate UK information return.
• OECD/G20 Inclusive Framework, administrative guidance package of 5 January 2026 (the "side-by-side" package) - the elective deemed top-up of zero under the income inclusion and undertaxed profits rules for groups headquartered in a qualified side-by-side jurisdiction, for fiscal years beginning on or after 1 January 2026; the extension of the transitional country-by-country safe harbour to fiscal years beginning on or before 31 December 2027 (and no fiscal year ending after 30 June 2029) at a 17% transition rate for 2026 and 2027; and the permanent Simplified ETR Safe Harbour for fiscal years beginning on or after 31 December 2026.
• OECD, Central Record of legislation with transitional qualified status - 37 jurisdictions with a qualified domestic minimum top-up tax or income inclusion rule applying from 2024, as at 1 May 2026.

The views here are my own. They do not represent the position of any current, former or future employer or client, and nothing here draws on confidential information.

© Jatinder Purewal 2026. All rights reserved.