The Authorised OECD Approach (AOA) has been a cornerstone of the international tax system for over a decade, providing the framework for attributing profits to Permanent Establishments (PEs) by treating them as if they were independent enterprises.
The AOA was formally introduced in 2010 through the OECD’s Report on the Attribution of Profits to Permanent Establishments, and its principles are closely linked to Article 7 of the OECD Model Tax Convention, which establishes fair and limited taxing rights for the host country when a foreign business creates a taxable presence through a PE.
Under the AOA, the OECD requires a detailed functional and factual analysis to establish where functions, assets, and risks should be allocated. Central to this analysis are Significant People Functions (SPFs), which serve as the key indicator of:
- Who exercises economic ownership and control over assets.
- Who assumes and manages the enterprise’s significant risks.
In the financial sector, this concept developed as Key Entrepreneurial Risk-Taking Functions (KERTs), such as making lending decisions or managing loan risks. Because financial assets and risks are closely aligned, KERTs provide a clear tool to allocate profits.
Outside the financial sector, the OECD broadened the concept to SPFs, since not all assets carry significant risks, and not all risks are tied to assets. Therefore, in practice, SPFs act as the OECD’s “location tool,” as identifying SPFs in a business allows one to determine which jurisdiction should be allocated the profits from controlling those assets and risks.
This people-centric logic has served as the backbone of transfer pricing and PE attribution for decades. But in today’s environment of accelerating automation and algorithmic decision-making, a crucial question arises: What happens to SPFs when decision-making (and the direct management of associated risks) is carried out by algorithms rather than people, or when human involvement disappears altogether?
The Fading Relevance of SPFs in a World of AI
Businesses across industries —from software and finance to logistics and e-commerce—are increasingly relying on AI-driven algorithms and autonomous AI agents to perform activities once handled by key human decision-makers. For example:
- A global logistics company may use machine learning to optimise routing and inventory allocation without human approval at each step.
- A financial institution may rely on AI models for credit scoring and fraud detection, decisions that once required human oversight.
- A multinational tech platform may deploy algorithms to set dynamic pricing in multiple markets simultaneously.
- At the most extreme, a company could in theory be composed entirely of AI agents, from the C-suite down to every department and function. This is the vision behind the recently announced “Macrohard” project from Elon Musk’s AI startup xAI. The name—a witty play on “Microsoft” reflects a serious ambition: to build a purely AI-driven software enterprise capable of replicating the operations of a traditional software giant, but with no human workforce. While Macrohard is currently an early-stage concept rather than a functioning business, its announcement illustrates how quickly the boundaries of automation are moving.
In each of these scenarios, the role and location of human SPFs become increasingly ambiguous. If key economic decisions are generated by algorithms, can we still argue that people are performing the functions and managing risks that justify allocating profits to a jurisdiction? This raises a fundamental challenge for the SPF test under the AOA.
From Human Functions to Algorithmic Control
To address this gap, policymakers and tax authorities may need to redefine what constitutes control in an automated economy. Instead of focusing only on individuals, attention may turn toward functional control of algorithms.
Key questions include:
- Who designs, owns, trains, or supervises the AI?
The entity responsible for developing or acquiring the algorithm may be seen as controlling the value-creation function. - Where is the decision logic embedded or executed?
If decision-making is tied to server locations, this could reintroduce a location-based test, but it risks distorting outcomes in cloud-based environments. - Where are oversight functions located?
Even in highly automated systems, some “human-in-the-loop” governance remains, such as compliance checks or exception handling. The location of these oversight roles may become the new proxy for SPFs.
This shift could lead to the emergence of “digital SPFs”, which are proxies for people functions that reflect control of automated systems rather than direct decision-making.
Potential Breakaway from the PE Concept
Beyond redefining SPFs, there is a possibility that reliance on PE as a trigger for taxation may decline altogether. The digital economy has already strained the PE concept, since highly digitalised businesses can generate significant value in a market without having any physical presence or employees there.
In this context, international debate has accelerated around alternative approaches, such as:
- Formulary apportionment (Pillar One, Amount A): Allocating profits based on a formula that accounts for sales, users, and markets, rather than people or physical presence.
- Destination-based taxation: Shifting taxing rights to where customers or final consumption occurs.
- User-based or market-based nexus concepts: Recognising that value in the digital economy often arises from user participation, data generation, and market engagement, regardless of physical presence.
These approaches reflect a broader shift away from the AOA’s reliance on SPFs and toward frameworks that capture market-side value creation.
This trend is already visible in real-world tax policy: several European countries, including France, the UK, Spain, and Italy, have implemented digital services taxes (DSTs) targeting user-generated revenues from advertising and platforms (e.g., France’s 3 % DST introduced in 2019).
Even within the U.S., Maryland attempted to tax digital advertising revenue through a user-location-based mechanism, marking a clear move away from traditional PE logic.
Implications for Multinational Enterprises (MNEs)
For MNEs, the move toward digital SPFs or alternative nexus rules (as those applied for DSTs) raises several practical challenges:
- Increased compliance complexity: Companies may need to document not only where their people are located but also where their algorithms are designed, trained, and supervised.
- Greater scrutiny of intangible assets: Ownership and control of AI systems, data, and platforms could become central in allocating taxable profits.
- Potential double taxation risks: Different jurisdictions may adopt different interpretations of digital SPFs, server location, or user-based nexus, leading to disputes.
- Alignment with BEPS initiatives: The OECD’s BEPS 2.0 project, particularly Pillar One, suggests an ongoing shift toward destination-based principles, but its coexistence with AOA rules may complicate compliance in the short to medium term.
A Turning Point for the AOA
The Authorised OECD Approach was built on the idea that people drive value creation. In an automated future, that assumption no longer fully holds. As AI-driven decision-making expands, the SPF test risks becoming outdated, forcing policymakers to rethink both the definition of control and the role of PE in allocating taxing rights.
Possible solutions may involve identifying digital proxies for SPFs, shifting focus to control of algorithms, or even moving beyond SPFs and PEs altogether in favour of market-based allocation rules.
What is clear is that the rise of AI agents and algorithmic decision-making represents not just a technological disruption but a fundamental challenge to the international tax framework. For both tax authorities and businesses, the coming years will require new approaches to ensure that profits are fairly and consistently attributed in the age of automation.
