Where has discretion relocated under Faceless Assessment and the Risk Management System?

Customs clearances, GST assessments, and income tax adjudication once depended on officer judgment; they are now governed by algorithms and risk engines. The Risk Management System releases the routine consignment without officer intervention. Faceless Assessment under CBIC and Section 144B routes flagged matters to an anonymous officer in a different zone. Discretion has not disappeared; it has relocated. The officer now sees only the exceptions, which raises the baseline of suspicion at each case. What does this architectural shift mean for how a company constructs the documentary record, and where does the relationship that matters now actually sit?

The single-window question examined the portal layer. This question examines what sits behind it: the shift from officers to algorithms. Officer-driven regulatory processing is giving way to system-driven processing in specific domains, and this shift is the most significant structural change in India's administrative architecture in the past decade. But it is partial, uneven, and widely misunderstood.

ICEGATE now processes customs clearances through the Risk Management System (RMS), which assigns risk scores to Bills of Entry based on importer history, HSN classification patterns, valuation benchmarks, and other algorithmic parameters. A consignment that clears the RMS threshold is released without officer intervention. This is the algorithm layer, and for routine shipments from established importers, it works. But when the algorithm flags a consignment for examination, the matter moves from the digital queue to an officer's desk, and from that point forward, it operates on the institutional logic of the pre-digital system: assessment, adjudication, escalation, appeal. The officer processing the flagged consignment exercises the same discretion that existed before RMS. The difference is that the officer now sees only the exceptions, which means they approach each case with a higher baseline of suspicion than when they were processing the full volume.

The relationship that mattered in the pre-digital system was the one that could get a consignment released. That relationship has largely been displaced by the algorithm for routine clearances. The relationship that matters now is different: it is the one that helps the company understand why the algorithm flagged a particular pattern, how the risk parameters are calibrated, and what corrective steps reduce the risk score for future consignments. The first relationship was transactional. The second is analytical. Most companies have not made this transition.

CAAR illustrates how discretionary architecture persists within the modernised layer. Between January 2021 and August 2024, 414 advance rulings were issued; only 46.2% within the statutory 90-day timeframe. The Nokia SFP transceiver case saw CAAR Delhi classify a product contrary to CESTAT's prior rulings, with the Delhi High Court eventually setting aside the CAAR order. Rulings bind only at the port of issue, not nationwide; the same product can face different classification treatment across ports.

The Faceless Assessment architecture extends the algorithmic logic beyond risk-scoring into the adjudication function itself. Under CBIC's Faceless Assessment Scheme, rolled out nationally from 2020, a Bill of Entry flagged for examination is no longer processed by the officer at the port of import; it is routed to an officer in a different zone, anonymously assigned through the system, who examines the consignment entirely on the documentation furnished through the platform. The importer does not know which officer is processing the matter; the officer does not know which importer is behind the Bill of Entry; the physical separation is designed to remove the relationship layer that previously governed the examination. CBDT's Faceless Assessment Scheme, operating under Section 144B of the Income Tax Act, applies a similar architecture to income tax: the assessing officer is no longer in the taxpayer's jurisdiction but assigned algorithmically from a national pool, and all communication runs through the National Faceless Assessment Centre. Both schemes have yielded specific institutional patterns that companies engaging the system have had to recalibrate around. The assessing officer, having no personal context for the taxpayer, now relies almost entirely on the documentation placed on record; representations that would previously have been clarified through a meeting are adjudicated on the papers alone, and submissions that do not anticipate the officer's specific queries yield adverse orders that then travel through the standard appellate architecture. The GA function's role has shifted from managing the officer relationship at the port or at the jurisdictional range to constructing the documentary record that will travel through an anonymous adjudication. A company that has not recalibrated its assessment strategy around this architectural shift is engaging a system whose institutional geometry it has not fully absorbed.

The extent of the shift varies dramatically across domains. In customs, RMS handles the bulk of clearances. In DGFT licensing, Advance Authorisation applications are tracked on the portal with system-generated deficiency notes, but the approval itself still involves officer review, and the officer's reading of "whether the applicant's export obligation can be met" is a judgment call, not an algorithm. In environmental clearances, PARIVESH has digitised the application process, but the Expert Appraisal Committee that evaluates the project is a room full of people exercising professional judgment. In pharmaceutical regulation, CDSCO's approval process remains substantially officer-driven. The companies that treat "digital" as meaning "no relationships needed" are as mistaken as those that assume nothing has changed.

The system has been digitised. The institutional culture that designed the system has not. And in the space between the two, both algorithmic fluency and institutional relationships operate. The GA head who has built relationships but has not studied how the risk engine classifies shipments is operating with half the map. The GA head who studies only the algorithm and assumes discretion has been eliminated is operating with the other half.