Strategic Vision

Procurement Guidance for Government AI

A handbook for decision makers on procuring AI infrastructure under GFR 2017, eCourts Phase III, and GeM guidelines.

jhanaPublic Sector Team

January 10, 2026

10 min read

TwLiLink

सत्यं वद । धर्मं चर ।

सत्यान्न प्रमदितव्यम् । धर्मान्न प्रमदितव्यम् ।

— Taittirīyopaniṣad, Śikṣāvallī, 1.11.1

1. Global DO & DON'T Checklist

The pace of AI releases creates what we call "technical stagflation": tools improve rapidly, but institutional procurement cycles lag by years. By the time a traditional RFP concludes, the technology landscape has shifted. The guidance below synthesizes best practices from GFR 2017, CVC Procurement Guidelines, Supreme Court e-Committee norms, and GeM standards—designed to help decision-makers avoid procuring tools that are obsolete on arrival.

Key Takeaway

The core principle of modern government procurement is derisking. Derisking means:

using off-the-shelf products rather than bespoke custom services
working with clear contracts, ideally with commercial terms and scalable fixed-prices per token or page used
picking technical winners based on AI benchmarks and tests

Governments should avoid working with not-for-profits with unclear provenance of funds, opaque incentives, and data capture or data sharing objectives. Instead, mature product platforms are a virtuous asset:

AI models are already trained and error-corrected on 100s of 1000s of prior user interactions and feedback—AI uniquely benefits from data economies of scale
They are validated and loved by actual end users, not irrelevant decisionmakers
Their cybersecurity is publicly evident and validated

But many products are not public or open, not even for paid use. Even their demos are safekept as secrets. Users cannot even register onto or try these products until their manager has already signed a contract. This is typically a crucial red flag, and results in contracts without any real ROI.

We also recommend not reinventing the wheel. AI is too fast-moving for legacy companies or governments to innovate in-house without losing pace. Fixed-price, off-the-shelf products are already validated and loved by users.

Item	Do	Don't	Remark
Require off-the-shelf, fixed-price products	+ Predictable, validated products built by world-class teams.	– Developing tech in-house with no derisking or data economies of scale.	Legacy service companies and governments are unlikely to produce cutting-edge technology in-house.
Require ontological architecture	+ Formal verification with line-level citations and audit trails.	– Black-box AI with unverifiable outputs.	Every claim must be traceable. Every output auditable (aligns to defensibility and auditability expectations under public procurement).
Require comprehensive legal/admin datasets	+ Proprietary databases with live, versioned, complete legal/admin data.	– Generic models and GPT wrappers with knowledge cutoffs.	Government work requires daily indexing of new judgments, circulars, amendments, etc.—not Google/GPT.
Require operational fluency	+ Forward-deployed engineers with domain expertise; on-site process mapping.	– Remote-only support; ticket-based resolution; lawyer- or sales-only teams.	Technology adoption requires hands-on implementation by technically fluent operators.
Require public products with real users	+ Products that are loved bottom-up survive; forced procurements go to waste.	– Decision-makers captured by fake demos of unavailable products.	Public pilots and real end-users reduce procurement risk and improve defensibility.
Avoid non-profit dependencies	+ Sustainable commercial model with aligned incentives and transparent data ownership.	– Grant-funded projects with uncertain continuity and opaque data-sharing.	Long-term institutional partnerships require commercial viability and clear incentives.
Require platformization AND interoperability	+ Pool data into a singular system to reduce exposure; ensure that system is interoperable.	– Point solutions that only do one thing; tools that don't talk to other tools.	Reduces exposure to multiple vendors and cybersecurity points of failure.
Require API-based, self-owned data sovereignty	+ Self-owned infrastructure with role-based permissions, admin, analytics, and bulk ops.	– Web-only or app-only consumer products with vendor data access.	Government data is sensitive and large volume; must be stored on government servers.

Source: jhana.ai/courtroom (GovTech product note)

2. Rules for Courts per eCourts, SCI & GFR

Summary: Procurement paths that actually work for AI pilots

Fastest low-friction path: the eCourts Phase III DPR explicitly allows directly procuring empanelled cloud services via GeM, including managed services. Specifically, for radically novel technologies like jhana Courtroom, courts can also contemplate single-source procurement on the basis of uniqueness of the services by issuing a Proprietary Article Certificate (PAC). This is when "...extraordinary circumstances justify exceptions to competitive tendering and the use of single-source procurement."¹
Neutral procurement route: For most courts, the best path is to issue an EOI to map the market, then award a paid pilot with clear success metrics; on successful pilot, scale via GeM tendering (QCBS-style technical-first) or a single-source / PAC route where uniqueness is demonstrable.¹
GeM is the default posture in court tech: Supreme Court procurement rules and e-Courts Phase III DPR both point to a GeM-first procurement stack.²
Quality is everything, L1 is meaningless without AI benchmark and feature sets: modern frameworks expressly support technical-first evaluation (QCBS) and documented exceptions for specialized tech.³

¹ e-Courts Phase III Vision Document, at 109–12, link.
² Supreme Court of India, Procurement Procedure, 2023, link; e-Courts Phase III DPR, at 96, link.
³ Ministry of Finance, Manual for Procurement of Consultancy & Other Services (2017) para. 5.1.4.

The judiciary is not a typical government department. It operates under its own constitutional mandate and has developed procurement norms through the eCourts Phase III Vision and the Supreme Court's Procurement Procedure, 2023. These frameworks explicitly shift away from purely L1 (lowest bidder) tenders toward Quality-cum-Cost-Based Selection (QCBS) and Single-Source (PAC) procurement for specialized technology. The rationale is simple: the cheapest AI is rarely the most reliable, and reliability is non-negotiable for the administration of justice.

Key takeaway.If you want auditable, court-grade AI to enter the institution quickly (without procurement deadlock), the most defensible sequence is: EOI → paid pilot → scale via GeM tendering (QCBS-style) or PAC (where uniqueness is evidenced).

A. The General Legal Framework

Constitutional Mandate: All government procurement must satisfy Article 14 of the Constitution—actions must be non-arbitrary, transparent, and fair. This does not mean lowest-price-wins; it means the process must be defensible, the criteria must be disclosed, and the selection must be justifiable.

Rule 173 & 166 — General Financial Rules (GFR), 2017

Rule 173 is the fundamental guidance for procurement of goods and services by Central Government ministries and departments.

Rule 166 (Single Tender Enquiry): Critical for procuring unique or proprietary technology. It permits procurement from a single source in defined circumstances and is commonly operationalized via a Proprietary Article Certificate (PAC)—a formal declaration that no substitute exists.

Key takeaway.In practice, PAC becomes strongest after a pilot because the institution can point to (i) observed performance, (ii) integration constraints, (iii) data-security posture, and (iv) measurable outcomes that were not matched by alternatives.

B. The Shift from L1 to QCBS

The "L1" system (lowest price wins) is often unsuitable for complex AI software where quality varies drastically. Quality-cum-Cost-Based Selection (QCBS) is the standard for consultancy and technology services.

GUIDANCE

Central Government Framework: The Manual for Procurement of Consultancy & Other Services, 2017 (Para 5.1.4) states that for consultants, "Quality- and Cost-Based Selection (QCBS) is the most commonly used method."

Weightage: The manual recommends a 70–80% weightage for Technical Score and 20–30% for Financial Score. This ensures that the highest quality solution is selected, provided the price is reasonable.

Key takeaway.For AI systems used in adjudicatory administration, QCBS is the most natural "defensibility" format: it allows the Court to publish technical criteria upfront (auditability, security, accuracy, support, dataset completeness) besides auditing features for various stakeholders (registrars, filing clerks, listing clerks, judges, research teams, publications, comms, etc.) and then treat price as a secondary variable rather than the decision-driver.

C. Modern Procurement Channels: GeM, EOI/Pilot, and PAC

Government e-Marketplace (GeM)

GeM is the national public procurement portal. It supports multiple procurement methods:

Direct Purchase: Up to ₹25,000, buyers can purchase directly from any seller (not necessarily L1) based on quality and suitability.
L1 Bidding & Reverse Auction: For higher-value standardized goods.
Cloud & SaaS: The eCourts Phase III DPR (Page 49, 96) explicitly contemplates procuring empanelled cloud services via GeM, including managed services.

Key takeaway.For a startup-grade but court-ready product (like jhana), GeM is not just compliance—it is speed. Where the Court can do a GeM-aligned pilot purchase quickly, it avoids long bespoke RFP cycles and instead buys outcomes.

Supreme Court Mandate

The "Procurement Procedure, 2023" of the Supreme Court of India establishes an unambiguous GeM-first policy.

Rule 8(a)(i): Up to ₹25,000 — Direct online through GeM from any supplier (not necessarily L-1).
Rule 8(b)(i): ₹25,001 to ₹5,00,000 — Through GeM Tendering.
Rule 8(c)(i): Above ₹5,00,001 — Through GeM Tendering / CPPP.

Key takeaway.When an institution says "we can only do L1," the Supreme Court's own procedure is the clean rebuttal: GeM-first with defined value bands, and direct purchase not necessarily L1 at the smallest band.

Pilot Projects & Expressions of Interest (EOI)

For genuinely novel technology, a phased approach is recommended. The eCourts Phase III Vision Document (Page 112) advocates:

EOI: Survey the market and assess available solutions.
Paid Pilot: Limited-scope, funded pilot using delegated financial powers.
Scale-Up: Use pilot results to justify a larger procurement—via QCBS or via single-source PAC where uniqueness is demonstrable.

Key takeaway.If your goal is to procure a startup like jhana.ai without diluting quality into a lowest-bid contest, the EOI/pilot route is the most procurement-safe bridge: it creates a documented record of (i) requirements, (ii) measured performance, and (iii) operational fit—so the scale-up tender (QCBS-on-GeM) becomes easy to draft, or the PAC justification becomes easy to defend.

D. High Court Autonomy & Decentralization

The eCourts Project Phase II Policy Document (2014) laid the groundwork for decentralized procurement. Chapter 2, Para 1 states: "The Project must be decentralized with greater responsibility being placed on the High Court to ensure that infrastructure, hardware and day-to-day issues are taken care of at their end." Each High Court is designated as the "Implementing Agency," granting it autonomy to manage its own tenders and procurements—including adopting QCBS or GeM routes independent of central delays.

Key takeaway.For court technology, procurement can be structured as an institution-led modernization project, not a generic government IT purchase: the High Court has the administrative standing (as Implementing Agency) to run an EOI/pilot process and then choose the most defensible scaling route.

"If extraordinary circumstances justify exceptions to competitive tendering and the use of single-source procurement..."
— e-Courts Phase III Vision Document (2022), pgs. 109–110

This is a direct policy endorsement from the highest judicial technology body for using the Proprietary Article Certificate (PAC) route under GFR Rule 166. When a technology's unique capabilities can be demonstrated—often following a successful pilot project—single-source procurement is explicitly sanctioned. This is the pathway for procuring genuinely novel AI infrastructure without the delays and compromises of lowest-bidder tenders.

APPENDIX A: STATE-SPECIFIC FRAMEWORK

The Framework in Karnataka: KTPP Act & Rules

State-level procurement adds another layer. In Karnataka, the primary legislation is the Karnataka Transparency in Public Procurements (KTPP) Act, 1999 and the KTPP Rules, 2000. Crucially, these rules explicitly authorize quality-based evaluation—providing the statutory basis for QCBS even within the state framework.

Rules 27, 28 & 25 — KTPP Rules, 2000

Rule 27 (Pre-qualification): Screening based on experience, capability, and financial status.
Rule 28 (Two Cover Tenders): Technical-first evaluation (QCBS mechanism).
Rule 25 (Lowest Evaluated Price): Permits scoring-based evaluation where specified—core to QCBS.

Other states have similar frameworks. The principle is consistent: quality-based procurement is not a workaround—it is explicitly authorized by statute when technical complexity demands it.

Key Takeaway (operational recommendation)

Key Takeaway

Recommended procurement posture for jhana.ai: (1) issue an EOI scoped to court-grade AI (audit trails, citations, data security, on-prem / VPC options, and measurable turnaround time), (2) award a paid pilot with published acceptance criteria, and (3) scale through GeM tendering (QCBS-style technical-first evaluation) or a PAC/single-source route where the pilot demonstrates uniqueness.

Rule Hooks (for citation / file notes)

GFR (2017): Rule 173 (general procurement), Rule 166 (single tender enquiry; PAC pathway).
SCI Procurement Procedure (2023): Rule 8(a)(i), 8(b)(i), 8(c)(i) value-band methods; GeM-first posture.
Manual for Procurement of Consultancy & Other Services (2017): Para 5.1.4 QCBS; typical 70–80/20–30 technical/financial weightage.
eCourts Vision / DPR: pilot/EOI framing; cloud/SaaS procurement through empanelled GeM channels.
KTPP Rules (2000) (Karnataka): Rules 25/27/28 enabling pre-qualification and technical-first evaluation.

"The rising tide shall lift all boats."

jhana.ai · Made in India

Key Takeaway

jhana meets all criteria in this guidance. Our technology is deployed with 5+ High Courts and 3+ Central Ministries. We are GeM-registered, offer API-first self-owned infrastructure, and provide forward-deployed engineering support.Contact us for a procurement briefing tailored to your institution.