A recurring dispute in Indian transfer pricing litigation involves the TPO selecting comparables with high margins and the taxpayer arguing that those comparables are functionally dissimilar. A recent Bangalore ITAT order in the Quotient Technology case follows this pattern, but the specific reasoning used by the Tribunal is worth examining closely. The Tribunal excluded Evoke Technologies and Great Software Laboratory from the comparable set because they were engaged in advanced work involving AI, machine learning, cloud computing, IoT and analytics. Quotient's Indian entity, by contrast, was found to be a captive performing routine development and support work for its US parent, with core R&D, intangible ownership and strategic decision-making retained offshore.
The functional comparability requirement under Rule 10B(2) has always asked whether a comparable performs similar functions, bears similar risks and deploys similar assets. What is notable in this order is the specific marker the Tribunal used to establish dissimilarity: AI/ML capability itself. A company that builds its own machine learning products or offers machine-learning-driven consulting was treated as not comparable to a routine captive, on the reasoning that AI work signals ownership of valuable intangibles, entrepreneurial risk, and margins that a cost-plus routine development model would not justify.
The reasoning that benefits captives in this case may create a different problem for captives that have themselves moved into AI-native work. Several Indian Global Capability Centres have shifted over the past two years from routine coding and support tickets toward fine-tuning models, building internal agentic workflows, running prompt-engineering and data-annotation pipelines, and in some cases owning product-level AI tooling for their global parent. If a TPO or a future Tribunal applies the Quotient reasoning symmetrically, an Indian captive that has taken on this kind of work while still operating under an intercompany agreement describing it as a 'routine software development service provider' could face a characterization problem. The functional profile recorded in its intercompany agreement, its Form 3CEB/Form 48 disclosures, and its TP documentation may no longer reflect the work the entity actually performs.
The margin stakes run in the same direction as the Quotient dispute, only reversed. A TPO applying an entity-level or AI-adjusted comparable set to a GCC that has outgrown its routine captive label could argue for a materially higher arm's length return than the captive's existing cost-plus markup assumes. The NTT Data case illustrates the scale of this kind of gap, with segmental margins computed at 10.17% against entity-level margins of 3.90%. GCC parent groups that have not updated their functional analysis to reflect AI-related work now being performed in India are exposed to the same mismatch between documented function and actual activity that TPOs look for.
Practitioners advising GCCs need a clearer sense of where the line sits between an Indian captive using AI tools to perform routine work faster, which should remain comparable to standard BPO/SWD benchmarks, and an Indian captive performing AI functions that create value similar to the comparables excluded in Quotient, which could warrant a different characterization and markup. No authority, whether the OECD, the CBDT, or any Tribunal, has drawn that line yet. The comparables jurisprudence is nonetheless building the vocabulary that is likely to be used to draw it eventually, and that vocabulary is emerging first from the taxpayer-favourable side of the argument. TP advisors should review their GCC clients' functional profiles now and test whether the entity's actual day-to-day work would survive the same AI-based scrutiny that currently protects it.
No comments:
Post a Comment