Reserved for MSEs, Invisible to AI: What the 25% Procurement Mandate Reveals About Discovery

A 25 percent floor, by law

The Public Procurement Policy for Micro and Small Enterprises Order, notified by the Ministry of MSME and effective from 1 April 2012, set a minimum annual procurement target from MSEs for Central Ministries, Departments, and Central Public Sector Undertakings. The original order set that floor at 20 percent of total annual procurement. An amendment effective 9 November 2018 raised it to 25 percent, with 4 percent earmarked for MSEs owned by Scheduled Caste and Scheduled Tribe entrepreneurs, and 3 percent for MSEs owned by women.

The policy carries real mechanics behind the headline number, not just an aspiration. MSEs quoting within a price band of 15 percent above the lowest bidder (the "L1+15%" rule) are given the chance to match that price and still supply at least 25 percent of the tendered value. MSEs are exempt from earnest money deposit. Tender documentation is provided free of cost. These are structural advantages written into procurement law specifically to correct for the disadvantages small firms face against larger, better-resourced bidders.

It is, on its face, among the most direct visibility guarantees a government offers small business: not a subsidy, not a grant, but a mandated share of a market that runs into the tens of lakh crore. The Government e-Marketplace, the digital platform through which most of this procurement now flows, reports cumulative gross merchandise value of more than ₹16.41 lakh crore since inception, as of 30 November 2025.

The mandate is being met, on aggregate

By the numbers the government publishes, the policy is working better than its own target. The Department of Commerce's year-end review for 2025 reported that MSEs, with more than 11 lakh registered on GeM, contributed 44.8 percent of cumulative order value for FY 2025-26 as of 30 November 2025, against the mandated 25 percent, and had collectively received orders worth more than ₹7.35 lakh crore. A parliamentary committee report from March 2026 put the broader CPSE figure even higher: overall MSE procurement risen to 47.4 percent of total procurement, "significantly exceeding" the mandate.

At that level of aggregation, the record reads as a policy success. One level down, at the level of which firms inside that aggregate are actually winning orders, the pattern is less even, and it resembles a pattern documented separately in how AI answer engines choose which businesses to name: aggregate coverage appears strong while a specific, identifiable segment stays thin, not because it is barred from the market, but because it is harder for the system to find, verify, and correctly categorize.

Registered is not the same as visible

The clearest evidence of that gap comes from the government's own data on women-owned MSEs. GeM's SWAYATT initiative, marking its sixth year in February 2025, reported that women entrepreneurs comprise 8 percent of the total seller base on the platform, with a cumulative 1,77,786 Udyam-verified women micro and small enterprises registered. Those sellers had fulfilled a cumulative order value of ₹46,615 crore. But GeM's own CEO stated in the same release that women's overall share of procurement stood at 3.78 percent, well below their 8 percent share of the seller base, and set an explicit goal of raising that percentage.

That gap can be stated in a single comparison: a segment can be 8 percent of registered sellers and still capture under 4 percent of what is procured. Registration is a real step and a necessary one, but the SWAYATT numbers indicate it is not the step that determines outcome. Something between registration and order fulfillment, discoverability inside the catalog, correct categorization, the buyer's ability to find and trust the listing, appears to be where the share compresses.

The March 2026 parliamentary committee report on the Ministry of MSME's Demands for Grants shows an even starker version of the same pattern for a different sub-target. SC/ST-owned MSE procurement stood at just 1.85 percent against a mandated 4 percent, "less than half the mandated" floor, representing a cumulative shortfall the committee estimated at approximately ₹18,000 to ₹20,000 crore over six years. Seventy-six CPSEs were penalized for sub-target non-compliance in FY 2023-24 alone. Women-owned MSE procurement, by contrast, stood at 3.46 percent in the same report, above its 3 percent target though down from the 3.78 percent SWAYATT had reported a year earlier under a slightly different measurement window, both readings sitting well under the segment's actual share of the registered seller base.

What the government's own auditors found inside the system

The mechanism behind the gap is not a mystery; it has been audited directly. The Comptroller and Auditor General's Union Government (Commercial) Audit Report No. 18 of 2020 examined GeM, incorporated as a Special Purpose Vehicle in May 2017 and made mandatory for central government procurement under Rule 149 of the General Financial Rules, 2017. Its findings on the platform's data layer are specific: it found "deficiencies in the input controls for buyer and seller registration," and concluded that "the process of registration and verification of users needed further strengthening along with regular cleaning and updation of legacy data and wrong data."

The same audit paragraph noted that "despite being mandatory for all central government offices, high number of dormant or inactive users indicated that universal acceptance may not have been achieved." In other words, the government's own auditor found, at the level of the platform's underlying records, exactly the condition that produces an SME-visibility gap: registrations that are not clean, not verified, and in a meaningful share of cases, not active. A catalog cannot surface a seller reliably if the seller's own record in that catalog carries unresolved data-quality problems.

This is not a claim that GeM is dysfunctional; the aggregate MSE procurement numbers argue the opposite, that the mandate broadly works. It is a claim that the mandate and the underlying data infrastructure are two different things, and that a firm can clear the first (registration, Udyam verification, MSE certification) while still sitting on a weak footing in the second (a clean, current, correctly categorized, verifiable listing that a buyer or a search function can actually surface). The mandate reserves the seat. The data layer decides who is legible enough to be seated.

Why this is the same problem as the open market, not a different one

Public procurement and open-market AI visibility can appear to be separate systems: one runs on a legally mandated purchase preference and a government catalog, the other on how a large language model chooses which businesses to name in an answer. The mechanism underneath both appears to be the same. A buyer, human or synthetic, needs to find a small firm inside a large body of information, verify that the firm is who it claims to be and can supply what it claims to supply, and trust that record enough to act on it. GeM's buyer is a procurement officer searching a category inside a government catalog. An AI answer engine's "buyer" is a model synthesizing an answer from what it can parse and corroborate across the open web. Both systems reward the same underlying property: a business record that is structured, current, consistent, and verifiable, and both systems quietly filter out the business record that is not, even when that business is formally eligible, formally registered, and formally entitled to be found.

The Vendor Development Programme run by the Ministry's Development and Facilitation Offices illustrates this point, though not by design. Between 2019 and 2023, 9,474 MSMEs went through a structured programme specifically to help with "GeM registration and detailed understanding of Public Procurement." The government has, in other words, concluded that registration alone is not sufficient, and has run a hand-holding programme at national scale to close the distance between eligible and actually able to compete. That is an institutional acknowledgment that the gap this analysis describes is real, even where it is framed as a training gap rather than as a data-legibility problem.

The parallel research on AI answer engines makes the same point from the other direction. The peer-reviewed Generative Engine Optimization study (Aggarwal et al., KDD 2024) found that measurable properties of a source, cited statistics, quotations, corroborating authority, changed whether that source was cited inside a generated answer. The lever in both systems is not marketing spend or brand recognition; it is whether the record is legible and corroborated enough for the system doing the choosing to trust it.

What "seated at the table" actually requires

None of this indicates that the 25 percent mandate is ineffective. The aggregate numbers, 44.8 percent GMV share on GeM and 47.4 percent across CPSEs overall, are real and exceed the legal floor by a wide margin. The finding is narrower and, for an individual firm, more useful than "the mandate does not work": the mandate creates the opportunity, but it does not, by itself, determine which specific firms inside the eligible universe of more than 7.6 crore Udyam-registered enterprises actually capture that opportunity. The evidence for this is the gap between a segment's share of registration and its share of procurement, visible in the women-owned MSE numbers, and more pronounced still in the SC/ST-owned MSE numbers.

The practical reading for an individual MSE owner is close to the one this research strand has made about AI visibility in the open market: legal eligibility and platform registration are the entry gate, not the finish line. A Udyam certificate and a GeM seller account establish that a firm is allowed to compete for the reserved share. Whether a specific firm is actually found inside that reserved share, by a procurement officer searching GeM's catalog or increasingly by any system reading structured business data, depends on a layer of legibility that registration alone does not guarantee: current and correctly categorized listings, verifiable credentials, and a record clean enough that neither a government buyer nor an automated system has to guess.