How Tax Law Must Adapt as AI Spreads Across Enterprise Operations
Companies are embedding artificial intelligence directly into business workflows-from finance to legal to human resources-and generating profits that tax authorities struggle to allocate. The problem: these gains often come from work performed by people in multiple countries using shared data and standardized processes, creating unclear transfer pricing obligations.
Unlike earlier waves of software adoption, enterprise AI doesn't require technical staff to deploy. A finance team can build an AI workflow that standardizes analysis across the organization. A legal department can create agents that draft initial documents. Operations groups can embed AI into email, spreadsheets, and knowledge bases where employees already work. The efficiency gains compound across geographies and teams, but the value generated sits in a gray zone for tax purposes.
Where Value Actually Gets Created
Companies realize AI value less through standalone tools and more by connecting models to enterprise data, wrapping outputs in approval workflows, and standardizing how teams use them. A tax professional might use AI to draft a first-pass analysis that a colleague then refines. A risk team might use it to run scenario analyses that would be infeasible to calculate manually. A marketing group might generate tailored messages at scale.
The real productivity gains come from integrating these capabilities into daily operations so time savings accrue continuously rather than as one-off pilots. Teams standardize prompts, templates, and review steps. AI becomes part of "how work gets done."
This creates a transfer pricing headache: when a workflow developed by one team generates benefits used globally, which entity should be compensated for that work, and how much?
The ERP Comparison Breaks Down
Companies have deployed enterprise resource planning systems for decades by investing heavily in integration, configuration, and controlled rollouts. Enterprise AI follows a similar pattern-accessed through third-party licensed models and managed platforms, with value realized through integration with company data, configuration of workflows, and decisions about who can change what.
The analogy stops there. ERP customization typically stays under central IT control with formal change management. AI capabilities often get created directly by operations, finance, or legal staff through prompts and workflow design. A finance manager might build an agent that proves so useful it spreads across countries overnight-without the implementation discipline that accompanies ERP rollouts.
This speed creates transfer pricing problems. An enhancement in one location can generate benefits globally, but nobody tracked the development costs or documented who owns the capability.
When AI Work Becomes an Asset
U.S. tax regulations treat an intangible asset as existing when it provides value beyond routine services. The question for AI: does the activity produce an identifiable, reusable capability with ongoing usefulness that the group can deploy repeatedly?
If yes, the work looks like intellectual property development. If no, it's a routine service.
AI work typically rises to IP development when teams build something reusable and scalable that captures proprietary know-how-a standard agent design deployable across countries, durable workflow logic that improves output quality, or curated datasets that enhance performance.
By contrast, work stays in the service category when it focuses on operating and administering the capability without creating something transferable-day-to-day use within an established playbook, human review and exception handling, access administration, monitoring, and routine updates.
Attributing Value Across Jurisdictions
Transfer pricing scrutiny will focus on which entity controls the key decisions: build versus buy, which use cases matter, how to define the data, how models get tested, and who has release authority. Tax regulations call this DEMPE-development, enhancement, maintenance, protection, and exploitation.
Companies should separate value created by AI across the contributing functions, assets, and risks, documenting who owns and controls each one. This also helps isolate returns tied to third parties (licensed foundation models, cloud services) from returns tied to group-created capabilities.
When it's extremely difficult to determine where AI value was generated, a profit split model may work best. When a central entity maintains strict control over AI development and charges affiliates based on token usage, cost-based service charges may apply. The outcome depends on facts: cost-based service charges for routine work, shared-service allocation for central enablement, or residual profit split for non-routine DEMPE functions.
Documentation Is the Foundation
Companies need governance materials, decision logs, funding approvals, model documentation, and change-control records that show who made the economically significant decisions. Without this paper trail, tax authorities will struggle to accept any transfer pricing position, and companies won't be able to defend their allocation of AI-generated profits.
Enterprises should establish clear governance over how AI is created, deployed, and reused so productivity gains are traceable to specific capabilities. Define who owns key responsibilities: decision rights over design and change, accountability for risk, and stewardship of ongoing enhancements.
Document how AI-enabled value is generated and scaled. Show who performs and controls the day-to-day AI-enabled services and where non-routine developments occur. This clarifies whether work should be characterized as services or intangible-related returns, and it strengthens the basis for any intercompany charging model.
Tax law hasn't caught up to how quickly AI spreads through organizations. Operations professionals building these systems need to treat them as governed capabilities, not just tools, and maintain the documentation to back up whatever transfer pricing position the company takes.
Your membership also unlocks:
