Supervising what you cannot inspect

by Maninder Singh Juneja and Renuka Sane.

In traditional financial regulation, supervisors are able to inspect the thing being regulated. For example, a scorecard by a bank was usually a short list of factors that one could comprehend. These factors could be traced to key documents within the bank, banks had a stated rationale for why they were being used, and there was some common sense in making weighted averages. The scorecard itself was static - the same weighted factors were applied over a period of time. Inspecting this process of constructing the scorecard and using it for management decisions was how trust was built. The AI world is different. Models are rented rather than built, change continuously, and behave probabilistically. They cannot be fully inspected even by the institution deploying them let alone by the regulator.

How should we then think of regulation?

One approach is to intensify the traditional approach where regulators demand more explainability, more documentation, and more validation. This will drive up the costs of compliance. But more importantly, this is ill suited to the new world where the technology changes rapidly, where the bank does not control the AI it uses, where there is no clear artefact that the bank can give the supervisor such as a model or a document (Board of Governors et al 2026). When regulators push traditional approaches, banks will respond by choosing AI models which are easy to document rather than the ones best for them, or defer AI deployments altogether. All these are unhappy consequences. What we need are policy makers who understand the live systems of the new world.

In this article, we analyse these emerging problems from first principles. We start from scratch, understand the landscape of market failure in the world of AI in banking, and think about how regulators can grapple with this world.

Our key idea is that AI systems resist replicability. Regulatory strategies that demand replicability will flounder or choke technology deployment. We suggest the regulatory standard applied should be AI deployments that are "supervisable" - the outcomes should be observable, they should be attributable to causes, and reversible by humans.

How AI is actually deployed

Before we get to the puzzles faced by regulators, we need to describe what AI in banking is. This categorisation is not unique to banking. Banking is simply where their consequences are regulated. Five modes span the range.

AI as Tool: AI augments a step, a search, a calculation, a first draft that the human controls. The human is cognitively engaged.

AI as Collaborator: The human and AI co-produce iteratively, and the human participates at every stage.

AI as Recommender: AI generates scores or options, and the human makes the decision. This is the classic human-in-the-loop, but credible only if the human can meaningfully interrogate the recommendation and not degenerate into cognitive surrender.

AI as Preparer: AI does the work and the human signs off. The approval here is closer to a check by a supervisor rather than a re-derivation.

AI as Autonomous executor (agentic AI): AI executes autonomously inside guardrails and the human monitors on exception. Emerging forms of agentic AI for banking run from single-task agents (a payment released, a service query resolved) to multi-step workflows and customer-facing agents that transact.

A central issue here is the true (de facto) role of the human. A reviewer who approves a thousand recommendations a day is not overseeing a model, the model is overseeing her. A feature of any deployment is the measured divergence between the model recommendation (and estimated uncertainty) vs. the human decision. Managers of banks will need to worry about relapses of human behaviour inside the organisation, a bit like how hospital managers worry about bad behaviour by doctors within their organisation.

What AI does to market failure in banking

We now shift gears to look at the standard knowledge on market failure in banking. Regulation may be justified when (and only when) there exist market failures which cannot self-correct fast enough, and there is adequate state capability in banking regulation to be able to correctly identify them and intervene. AI's distinctive feature is that it can cure several classic failures. Better default prediction reduces credit rationing, better fraud detection cuts deadweight loss, richer risk assessment lets banks serve customers they previously could not price, AI advisors help customers avoid some malpractices by the bank. But there are also some new problems that are anticipated.

Information asymmetry: This happens in banking at two levels: borrower-to-lender (adverse selection, hence credit rationing) and firm-to-consumer (product complexity, hence mis-selling). AI narrows the first through alternative data and may widen the second. On one hand, the customer armed with AI can see through many things proposed by the bank which are not in her best interest. But the consumer cannot observe why they were shown a product, offered a price, or steered toward a specific insurance plan. The sales process itself becomes more opaque. Personalised pricing approaches first-degree price discrimination, extracting consumer surplus. A single flawed model can mis-sell to millions simultaneously, converting isolated conduct failures of the pre-AI world into a big correlated event. And redress weakens when a denial comes from a model the firm itself cannot explain.

Systemic externalities from shared infrastructure: Each bank chooses its models, data sources and vendors to optimise its own performance. When multiple entities choose the same ones, the sector's exposures become correlated, which is a cost no individual bank prices in. We list the vulnerabilities below:

Correlated model risk: banks on similar models and the same foundation providers respond identically to an event. The regulator, at the system level, has to manage what happens when institutions move together, because no single entity has experience of such behaviour or of the impact synchronisation adds.

Third party concentration: one vendor's failure propagates everywhere at once. India has already run this experiment, when a ransomware attack on one shared technology provider knocked roughly three hundred cooperative and regional rural banks off the payments network.

Correlated cyber breach: shared stacks mean one exploited vulnerability is every institution's vulnerability. AI lowers the attacker's costs (automated vulnerability discovery, deepfake social engineering) and adds new attack surfaces (data poisoning, model inversion, prompt injection against agents that can move money).

Runs on banks at level 3. The bank runs of old were a queue on the pavement. Then we got to Silicon Valley Bank where over a weekend, customers took away money from the bank. Now we can be at level 3: autonomous agents managing customer cash can turn a shared signal into a self-reinforcing run at machine speed.

The various market failures listed above behave differently across the five modes of AI use. For example, when a human constructs the offer, opaque pricing can get contained. However, when an agent personalises autonomously at scale, this may become severe. Systemic correlation is moderate when AI advises and severe when fleets of similar agents act simultaneously. Any regulation that grades by model type alone, or by use case alone, misses half the object. The next question is the mode of regulation itself.

Regulatory strategy

Regulation can work in two ways. Process-based regulation is ex ante: it prescribes how the firm must operate, defines required controls, mandates oversight, validation standards, limits on autonomy. Outcome-based regulation is ex post: it prescribes ends, fair treatment, solvency, and judges results, leaving the choice of methods to the firm.

Outcome-based regulation is the efficient default. It is technology-neutral, so it does not ossify as methods change; it lets firms find the least cost route to compliance; and it does not require the regulator to understand the firm's production function better than the firm does. But it has important preconditions: the outcome must be observable and measurable; it must be attributable to the firm and, ideally, to the cause; and the harm must be reversible or compensable. Process regulation is the right departure from the default when those preconditions fail and where outcomes are unobservable, harm is catastrophic or irreversible, or damage manifests only systemically or with a lag.

Traditional doctrine treats these preconditions as given: examine the activity, choose the mode. This does not work for AI. Whether an AI deployment's outcomes are observable, attributable, and reversible is an engineering choice, which needs to be settled at design time. We suggest that regulation should mandate observability. This makes it possible to have an "outcome-based supervision" model. There are three ways to ensure observability.

1. Telemetry implies that the institution keeps a track of every decision such that the system records which version of itself it was using, what information it was given, and any time a person stepped in to overrule it. This makes outcomes attributable. That way if a certain group starts getting more (or less) approvals than before, the organisation can evaluate what caused the shift - was it the model, or the group itself. Such a capability is being mandated elsewhere in the world for similar use-cases (European Parliament and Council of the European Union, 2024). The regulator should also consider if it wants to set a minimum common telemetry standard.

2. Boundaries and rollback include putting caps on what the system is allowed to do, rolling out new updates to just a small number of cases first (instead of everyone at once), and having a tested plan for switching back to the older version if something goes wrong. This ensures that if a bad update slips through, it only affects a small slice of decisions.

3. Probes make bias observable. One way is "paired testing": one submits two applications that are exactly the same except for details that hint at things like someone's race or gender, and see if they get treated differently. One can also compare approval rates against the company's own normal levels. Together, these checks can catch an unfair credit model in just a few weeks, instead of waiting years to see who actually pays back their loans. One can also check rejected applicants against credit-bureau data to see which ones got approved by someone else, and how they fared.

These are similar to the idea of decision receipts that record which rules were applied to which facts and in what sequence for every decision made by a government or public system (Srivastava, 2026).

Over time, the supervisor should also build a repertoire of its own test cases drawn from incidents, complaints and examinations across the system and run it against every material AI deployment, much as stress-test scenarios are run against every balance sheet today. This will ensure that what surfaces in one institution becomes a probe for all others. It thus allows the regulator to set its own observability layer.

Process regulation is then reserved for the harms that are systemic, correlated, or irreversible at machine speed. For such events there need to be protections such as circuit breakers that halt things before they cause damage, limits on how much the systems can do on their own, model diversity so they don't all fail the same way, and rehearsed back-up plans.

What follows for the supervisor, the board, and the customer

For the supervisor: The unit of examination shifts from the model to the deployment, and the examiner's question shifts from "show me the validation report" to "show me the behaviour": what boundaries were set, what exceptions were thrown, what overrides were exercised, how far the system drifted from its baseline. Supervisors will also need to find the intellectual clarity to avoid a wide variety of extraneous claims about regulation of AI, e.g. the push for economic nationalism which has nothing to do with market failure.

For the board: A board cannot certify systems it cannot inspect or understand. Its role is to govern the framework which includes the limits on what the system is allowed to do, and making the rules for when a decision must be escalated to a human. The board then needs to continuously check the exceptions and overall performance.

Exception-handling should also be written into policy. If a problem stays unresolved beyond a defined size or time limit, it automatically gets escalated to the board. Internal auditors should double-check that the numbers are real. The board sticks to this audited framework and does not inspect the system directly.

If a board is asked to approve something they can't understand, they will default to saying no. But if you let them govern the limits and the exceptions instead, they can say yes.

The customer: The customer becomes part of the supervisory architecture. The widening firm-to-consumer asymmetry has a structural corrective the pre-AI world lacked: the customer now has AI too. Mainstream assistants abroad have begun connecting directly to users' accounts. If product terms (rates, fees, eligibility criteria) are mandated to be structured and machine-readable, the customer's own AI does the comparing, the explaining, and the policing of mis-selling, continuously and at zero supervisory cost. The redress channel weakened by opacity is restored the same way: an adverse decision should carry its reason to the customer, and what would have had to be different for the answer to change. This restores the ability to contest. The same asymmetry that AI widened, AI-equipped customers can close, but only if regulation hands them the data.

An example

Consider a debt-collection example. When borrowers fall behind on payments, the bank has to decide how to chase each overdue account. Contact methods differ in cost. Automated SMS and IVR (the automated phone system - "press 1 to pay") are cheap; having an employee actually phone the customer is expensive. So the bank builds a model that allocates accounts: cheap automated nudges for most, and the costly human call reserved for the accounts where the model predicts that talking to a person will actually "cure" the account (get it back to paying). Human calling works better but it costs more. The model is rationing an expensive resource to where it thinks it'll pay off.

With AI, the collection systems would retrain continuously or run reinforcement-style optimisation against a live reward (cure rate per rupee spent). The model would be moving on its own, faster than the review cycle, toward a target that is a proxy for what the bank may actually want. Under the conventional "inspect-the-artifact" approach, the model is checked before deployment, signed off, and reviewed on a schedule (say quarterly or annually). However, if AI is an active optimiser, a quarterly review discovers deterioration only after it has touched thousands of accounts. It may drift toward a mis-specified target in a way no one can read off the model itself. Further, if the collections model is a shared vendor product, or several banks fine-tune the same foundation model on similar data, they all go bad the same way at the same time when borrower behaviour shifts. In contrast, in a "supervisability-built in by design" approach, a small slice of accounts is deliberately kept on the previous allocation method, running live alongside the new model. This allows comparison between the old and the new in real time.

Every decision should keep a record of a few things: which version of the AI was used, what action it chose, which group of customers it was dealing with, and how things turned out in the end. This is how it helps: Say the AI's overall success rate is going up, but one particular group is quietly getting fewer phone calls from real people and doing way worse because of it. With all that recorded, that gap shows up right as it's happening. You can compare it against what's going well elsewhere, trace it back to a specific version of the AI, and undo it that same day, because the older setup is still up and running as a backup.

The company should never have to dig into the AI's inner "thought process" to realize its behaviour has taken a bad turn. None of this replaces the normal testing that the firm would do anyway. The difference is that here, the ability to observe what's happening, pin down what caused it, and shut it down are all built right into the system from the start, instead of being things you have to go do to the AI afterward.

Conclusion

AI deployments sit uneasily in conventional mores of banking regulation. The intelligence is rented, the behaviour is probabilistic, and harms can move at machine speed. We suggest that the response should not be more inspection of what cannot be inspected, nor blind faith in outcomes that arrive too late. Efficient regulation will emerge from observability, attributability, and reversibility. Regulators should require banks to build these.

References

Board of Governors of the Federal Reserve System, Federal Deposit Insurance Corporation, and Office of the Comptroller of the Currency. "Supervisory Guidance on Model Risk Management." SR Letter 26-2. April 17, 2026. https://www.federalreserve.gov/supervisionreg/srletters/SR2602.pdf.

European Parliament and Council of the European Union. Regulation (EU) 2024/1689 (Artificial Intelligence Act), arts. 12, 26(6), and Annex III(5)(b). https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-12.

Srivastava, Manish. "Digital Governance Needs Decision Receipts." Episode 73 of Big Ideas. XKDR Forum, June 1, 2026. Podcast, video, 13:06. https://youtu.be/WFX4ITb9yok

Announcement

Position for researcher in public finance and urban governance

XKDR Forum is looking for a full-time researcher with qualifications in economics, finance, law, public policy, or an MBA to work on public finance, municipal finance, cities, urban governance, and public health finance.

About XKDR Forum

XKDR Forum is a Mumbai-based inter-disciplinary group of researchers working in the fields of regulation, courts, household and firm finance, climate change, health, public finance management and land. In these fields, the group engages in academic and policy oriented research, and advocacy.

As a researcher who joins us, you will be part of our team that builds knowledge on the working of government and its organisations from the lens of laws, regulations and institutional capacity. As a research associate at XKDR Forum, you will work on project deliverables under the supervision of a research lead. You will be expected to work in person at the office premises in Mumbai.

Eligibility

The eligibility requirements for this position are:

1. Qualifications in economics, finance, law, public policy, or an MBA.
2. At least one short-form and one long-form publication in reputed publications.
3. Quantitative skills are preferred.

Candidates with work experience of 3-5 years are preferred. You must be comfortable working in an inter-disciplinary research environment with people from varying backgrounds such as data-science, economics, public policy and law. You must be curious and passionate about research and must be willing to work on independent outputs as well as in teams.

The remuneration offered will be commensurate with your skill and experience and will be comparable with what is found in other research institutions.

Interested candidates are required to fill out this form by June 30th, 2026.

Author: Prashant Narang

Prashant Narang is a researcher at TrustBridge Rule of Law Foundation.

On this blog:

• When remedies become regulation: The Karnataka High Court's intervention in food licensing and street vending, 9 June 2026.
• What happens when arbitration deadlines are missed, 2 April 2026.
• From Statute to Zero-Cost: Section 31A and the Bombay High Court's Zero-Cost Culture, 18 November 2025.
• How Indians rank their rights: what 26 interviews tell us about Article 19 and property, 30 October 2025.
• Evaluating India's Customs Authority for Advance Rulings (CAAR) and charting a path for reform, 11 March 2025.
• The Blind Spot in Indian Arbitration: Fees, Power, and Structural Oversights, 27 February 2025.
• Executive discretion in regulating private schools in India: Evidence from Delhi, 19 February 2020.

When remedies become regulation: The Karnataka High Court's intervention in food licensing and street vending

by Prashant Narang, Aryan Pandey and Indira Unninayar.

I. When public health litigation expands into regulatory governance

On 19 September 2025, the Karnataka High Court delivered its decision in Karnataka Pradesh Hotel & Restaurants Association v. Union of India. The case began as a routine industry challenge to the Food Safety and Standards Act, 2006. The judgment oversteps statutory adjudication to engineer regulatory design. It answers a real public-health worry. But litigation like this rarely stays within the parties before the court. The Court encroached into the executive territory with no consideration of whether the state is actually capable of implementing what it now directs. Such directions tend to produce selective enforcement and compliance costs that fall hardest on those least able to bear them.

The petition arose from a 2012 directive on licensing enforcement. The judgment was delivered nearly a decade and a half later by which time, the regulatory landscape and the affected ecosystem had evolved substantially. Street vending, food delivery and the law on informal work had all changed and all bore directly on what the Court now ordered.

II. What the petition sought, and what the Court ultimately directed

Hotel and restaurant associations had challenged orders to enforce the FSS Act and its regulations. The trigger was a letter dated 13 March 2012 issued by the State Food Safety Commissioner, acting on the Union instructions, requiring all States to enforce the Food Safety and Standards Authority of India's (FSSAI) licensing and registration regime. Every Food Business Operator' ("FBOs") had to obtain a licence or registration as a condition for continuing their business.

The petitioners contended that this requirement was impractical and arbitrary, especially applied uniformly to establishments of vastly different scale and capacity. The burden, they said, fell hardest on smaller operators. They went further, asking the Court to strike down swathes of the Act and its regulations as unconstitutional.

The Court rejected these constitutional challenges in their entirety and upheld the validity of both the Act and the Regulations, noting that the Supreme Court had already affirmed the Act. It restated food safety as a public-health aim and accepted the State's claim that the rules rested on scientific and international standards.

It then issued two directions with implications beyond the immediate dispute.

1. It directed the Union Government to classify restaurants into small, medium, and large categories and to enact separate laws or frame separate guidelines for each, observing that reliance on turnover-based thresholds alone, was impractical and insufficiently responsive to differences in size and operational capacity.
2. The Court directed the State government to introduce health and safety rules specifically for street vendors and food trucks, and to establish a mechanism to ensure strict oversight of their implementation.

These directions are what give the judgment its broader regulatory significance.

III. Expanded prescriptions sans diagnosis risk over-regulation, arbitrary discretion, and regulatory incoherence.

A. New rules directed without a policy diagnosis -

The judgment's biggest gap is that it never finds that existing regulation has failed. Nor does it explain why new, vendor-specific rules are required, and whether existing processes for licensing, inspection, and enforcement have failed. It even concedes that the licensing rules already impose hygiene standards on every operator.

The FSS Act already establishes a comprehensive enforcement architecture. Section 30 vests primary responsibility in the State Commissioner of Food Safety, while Sections 36 and 38 operationalise enforcement through prescribed methods and designated officers at the district level within municipal and local jurisdictions.

The Court should have asked two questions: were existing standards inadequate, and had enforcement failed? However, the judgment neither raises nor answers these questions.

The reasoning moves from a general observation about the informality of street vending directly to remedial directions that materially reshape regulatory obligations. It does so without identifying any institutional deficiency that might have justified such an expansive remedy.

B. The Street Vendors Act framework was overlooked entirely -

The Court acts as if street vendors operate in a regulatory vacuum. They do not.

The Street Vendors (Protection of Livelihood and Regulation of Street Vending) Act, 2014 ("SVA") was specifically enacted to balance livelihoods against congestion, public health and urban order. It overrides inconsistent municipal laws and works through town vending committees ("TVCs"), surveys, and certificates of vending. The SVA is not merely a procedural architecture; it embodies a considered normative choice by Parliament, that street vendors are rights-holders, entitled to livelihood protection, meaningful participation through TVCs, and procedural safeguards before any restriction on their vending.

By directing new health and safety rules for vendors without engaging with this framework, the Court implicitly undoes that normative settlement. It treats vendors not as participants with protected rights but as subjects of fresh regulation – inverting the very premise of the statute Parliament enacted for them.

The result is regulatory incoherence and it is worth being specific about what that means in practice. Under the SVA, a vendor acquires a certificate of vending through a TVC process that must include vendor representation; this certificate is her legal entitlement to occupy a designated vending zone. Under the FSS Act, she must separately obtain a licence or registration from FSSAI, subject to turnover thresholds and hygiene standards. The Court's direction would now superimpose a third layer: vendor-specific health and safety rules with a fresh enforcement mechanism. Each of these three regimes carries its own authority, its own compliance requirements, and its own enforcement officer.

C. The Court's directions assume state capacity that does not exist -

As far back as 2020, only 47% of town vending committees had any vendor representation; seven states had not notified schemes under the SVA, and in four states no compliant TVC had been constituted at all (Narang et al., 2020).

The enforcement machinery under the FSS Act tells a similar story. As of 2021, there were only 2,531 Food Safety Officers nationally for roughly one crore street vendors, with vacancy rates between 33% and 90% across states (Mishra & Khattar, 2025). Between 2018 and 2021, fewer than 1% of food adulteration cases ended in conviction. None of this means enforcement has stopped. It means enforcement has changed. When an inspector cannot police everyone, he polices whomever he likes – and scarcity only raises the price of his goodwill.

Piling fresh directions onto this will not help; it will hurt. Pritchett, Woolcock and Andrews (2010) examined three well-funded reforms (schooling in India, budgeting in Mozambique, land titling in Cambodia) that all failed for one reason: each demanded transaction-intensive implementation, millions of scattered discretionary acts no centre can supervise. Street-food safety is the same kind of task. It is transaction-intensive (a crore of vendors, countless daily sales), discretionary (each inspector judges hygiene on the spot), high-stakes (a failed check can end a livelihood) and opaque (the encounter leaves no record). On all four counts, the very dimensions Kelkar and Shah (2022) name as the hardest for any state to master, it scores about as badly as a task can.

The sequencing is backwards, too. Early state-building, Kelkar and Shah argue, should begin with low-stakes, high-visibility tasks, short feedback loops, correctable errors – and reach for hard ones only once capacity exists. The order to keep "strict vigil" over vendors does the opposite: it escalates coercion before building the institutions that would restrain it.

This dynamic has become characteristic of the Indian regulatory ecosystem. Shah's account of the history of Indian finance documents a pattern of regulatory agencies consistently engaging in micro-management whilst lacking the state capacity to enforce their own frameworks.

High discretion combined with low capacity does not produce zero enforcement; it produces selective, rent-seeking enforcement. When inspectors are too few to visit every vendor, they must choose whom to visit and a shortage of inspectors does not dilute that discretionary power, it concentrates and rations it. The fewer the officers relative to a crore of vendors, the more valuable each discretionary decision becomes, and the higher the payment it can command.

As Rai and Shah (2015) observe, the Indian state is too often strong as in scary but not strong as in capable: it commands coercive reach without the institutional depth to convert that reach into governance outcomes. Ordering strict vigil onto a system with 90% officer vacancies in some states therefore does not produce better public-health outcomes; it produces more rent-seeking. Inspectors arrive not on a fixed schedule but whenever they are short of cash, and vague, subjective standards give them the pretext to do so (The Seen and the Unseen, Ep 18). The Court's directions thus simply widen the regulatory perimeter within which this behaviour can operate.

D. Cross-jurisdiction comparisons are persuasive only when capacity is comparable-

The judgment leans hard on foreign examples to justify a strong licensing and enforcement regime. It cites international norms to rebut the claim that the regime is impractical.

But it ignores the conditions that make those systems work. Licensing does not work in the abstract. It needs capacity, trained inspectors, predictable procedure and firm limits on discretion.

The judgment itself notes that regulators such as the United States Food and Drug Administration recognise wide variation in the size and capacity of food establishments, and that enforcement is typically carried out by local health authorities. These details matter. They determine whether regulation produces overall compliance or its very opposite by way of uneven and discretionary enforcement.

This is where the comparison breaks down. The FSLRC (2013) treats foreign models as inputs to adapt, warning against any bid to "mechanically transplant ideas from elsewhere". International standards inform; they do not, on their own, justify a domestic enforcement regime. The court inverted this. It used the FDA comparison as the justification itself, without asking whether the administrative architecture that makes those powers function exists here.

Pritchett, Woolcock and Andrews (2010) show why that architecture cannot simply be assumed to exist. As per them when governments copy institutional forms from higher-capacity settings, the laws, the agencies, the enforcement powers, without first building the administrative foundations that make those forms function, the result is the appearance of reform without its substance. It is, in their words, no reform at all.

The FDA comparison does not establish that India's enforcement regime should be intensified. It shows only that the FDA works within machinery that makes its powers function. Transplant the powers without that architecture and you import the coercion while leaving behind the restraint.

E. Non-parties bear the burden of directions issued without participation -

The High Court has not abided by one of the basic principles of natural justice, audi alteram partem – the 'right to be heard' before any orders are passed against a person, as it has not 'impleaded' and 'heard' street vendors and pliers of food trucks, before proceeding to pass directions concerning them. Yet it ordered the State to write new health-and-safety rules for them and to keep 'strict vigil' over them, without studying who they are or what they face.

The regulatory burden falls on informal workers operating under constrained economic conditions. The court treats informality as a regulatory gap to be closed, vendors operate outside the system, so the system must be extended to capture them. Shah (2026) inverts this reading. Where the state's enforcement is slow and unreliable, operating informally is not evasion of good rules but a rational adaptation to bad institutions. Vendors build workarounds precisely because formal compliance offers little protection and predictable harassment. The state then misreads the adaptation as defiance and tightens the rules, which raises the cost of formality further and entrenches the informality it set out to cure. A direction to bring a crore of vendors under "strict vigil" is the next turn of exactly this cycle.

IV. Food safety is a compelling goal, but cannot justify prescription without basis

The strongest defence of the Court's approach lies in the public interest at stake. Food safety directly impacts public health and the FSS Act itself emphasises risk management, consumer protection, and preventive regulation. The Court did not draft rules itself; it told the executive to. . Read this way, the judgment can perhaps be seen as an attempt to prompt more effective implementation of an existing legal framework.

However, that defence, has limited force if any, as the Court does not explain the reasons why such directions pertaining to street vendors and food trucks were required in the first place, and how the existing enforcement mechanisms under the FSS Act were inadequate. Without a demonstrated failure, intervention at the level of design has little to stand on.

The promise of later consultation cures nothing. Consultation after an order to make rules is not consultation about whether the rules are needed at all. Once the outcome is predetermined, the space for meaningful policy deliberation is confined to that predetermined outcome.

The Court unfortunately moved too quickly from concern to prescription, and in doing so, blurred the line between ensuring lawful administration and reshaping the regulatory architecture itself.

V. Conclusion: Prescriptions must stay focused and relevant

The judgment reflects a growing tendency: courts shifting from reviewing validity to supervising regulation, especially under the banner of public health or public interest. Such interventions may be well-intentioned. But good intentions do not substitute for institutional competence. In this case, the Court's directions go beyond correcting unlawful administration to enter the terrain of regulatory design, without any demonstrated failure of the existing framework and without hearing those most affected by the outcome.

This tendency is not confined to any single domain. As Jain and Reddy T (2025) observe, reform through judicial diktat characteristically bypasses public consultation on questions that carry complex second-order effects. The adversarial courtroom is not designed for the stakeholder deliberation that sound policymaking requires. When it substitutes for that process, the people most affected, here, street vendors and food truck operators, bear consequences that were never examined.

Lon Fuller, in The Forms and Limits of Adjudication (1978), offers a useful framework for understanding why. Fuller identified a class of problems he termed "polycentric", those where the disposition of any single issue carries implications for every other, such that pulling one strand "will distribute tensions after a complicated pattern throughout the web as a whole". In such contexts, he argued, adjudication becomes institutionally incapable, because the affected party's participation through proofs and reasoned arguments loses all meaning when no advocate "could possibly present to the tribunal the grounds that must be taken into account in the decision".

The Karnataka High Court's directions bear precisely this character. A judicial mandate to introduce new health and safety rules for street vendors does not resolve a discrete regulatory question, it simultaneously displaces an existing framework under the Street Vendors Act, imposes fresh compliance burdens on informal workers already operating at the economic margin, adds enforcement obligations to a system strained by Food Safety Officer vacancy rates and multiplies points of regulatory contact where discretion can be monetised. Each of these consequences shapes the others, and that interdependence is exactly what Fuller's framework identifies as lying beyond the proper limits of adjudication.

The cost is not only procedural. Compliance burdens imposed without the capacity to administer them do not produce better governance; they tax the everyday enterprise of people operating at the margin and dampen the very economic activity the state should want to encourage. As Shah (2026) puts it, this is the "effervescence of creativity and invention that a poor country cannot afford to extinguish."

The lesson is that remedial ambition must be matched by remedial discipline. Prescription without diagnosis, and supervision without capacity, do not produce better governance. They produce the illusion of it.

References

Bedi J. and Narang P., 2020. Progress Report 2020: Implementing the Street Vendors Act. Centre for Civil Society.

Mishra G. and Khattar J., 2025. FSS Act: Need for enforcement and accountability in India's food safety regime. Bar and Bench. 26 June 2025.

Pritchett L., Woolcock M. and Andrews M., 2010. Capability Traps? The Mechanisms of Persistent Implementation Failure. Center for Global Development.

Kelkar V. and Shah A., 2022. In Service of the Republic: The Art and Science of Economic Policy. Penguin Allen Lane.

Varma A. and Menon M., 2017. Restaurant Regulations in India. The Seen and the Unseen. 15 May 2017.

Financial Sector Legislative Reforms Commission, 2013. Report of the Financial Sector Legislative Reforms Commission. Ministry of Finance, Government of India. 22 March 2013.

Jain C. and Reddy T P., 2025. Why reform through judicial diktat is fraught with perils. Times of India. 8 November 2025.

Fuller L. and Winston K I., 1978. The Forms and Limits of Adjudication. Harvard Law Review, Vol. 92, No. 2.

Rai S. and Shah A., 2015. Going from strong as in scary to strong as in capable. The Leap Blog. 25 February 2015.

Shah A. and Varma A., 2026. Why Freedom Matters | Episode 10 | Everything is Everything. Everything is Everything. 1 September 2026.

Ahluwalia R. and Shah A., 2026. Why Firms Build Economies Ft. Ajay Shah | Growth is Good | Ep 25. Foundation for Economic Development. 27 March 2026.

Prashant Narang and Aryan Pandey are researchers at TrustBridge Rule of Law Foundation. Indira Unninayar is an Advocate-on-Record, Supreme Court of India.

Trust in the Era of the AI-Informed Customer

by Maninder Singh Juneja.

A patient grappling with a longstanding eye problem was diagnosed with MGD (Meibomian Gland Dysfunction) by an eye specialist. To make sense of the medical jargon, she photographed her meibography report, the gland scan, and uploaded it to an AI, which confirmed the diagnosis and the line of treatment. A few weeks later, wanting to avoid buying shades of cosmetics she already owned, she uploaded a photograph of her stock and of the items she was about to buy to the same AI. Instead of commenting on the shades, the AI told her the waterproof eyeliner she had used for years was blocking the pores of the gland, like wax in a drain, and making her dry eyes worse. She stopped, and this one change brought immediate relief. The AI connected across domains, unasked. Neither the physician nor the cosmetics counter had connected the two. The ophthalmologist saw the eye. The counter saw the product. AI saw the person. Neither was wrong. Each was trained to look at one domain. Every institution serves its own; AI serves the customer. This is not a failure of individuals but a structural shift with strategic implications. In this article we examine these implications.

Verification in markets

For as long as markets have existed, buyers have had to trust sellers, because there was no way to verify their word, or the effort and cost were too high. The economist George Akerlof spent many days puzzling over the nature of the vegetable retail market in Delhi. In 1970, he offered new insights into the market failure that arises when the buyer cannot verify what the seller knows. He later won the Nobel prize in economics for this work which helps us understand asymmetric information.

The market economy combats asymmetric information using substitutes, brands, professional licensing, statutory audits, regulators, and guarantees. For example, the customer trusts the bank's brand name instead of reading the fine print, or the doctor's medical degree instead of evaluating the diagnosis. Half a century of institutional architecture in financial services is built on this logic.

The use of AI has brought about a foundational shift in trust. AI has made verification in many situations quite feasible. Customers who once accepted substitutes for trust in institutions can now check a claim, compare alternatives, or challenge a recommendation on an inkling of doubt. The barriers to specialist access, cost, time, language, jargon, have all collapsed at once. The result is not an end of trust, but the pillars on which trust rests have changed.

How it impacts the brand

The brand is an informational shortcut, one of Akerlof's substitutes for verification. It stood in for the customer's inability to verify. The bank's or insurer's reputation stood in place of the customer's understanding of provider soundness, and the terms and conditions of the contract. AI changes that, but the picture is complicated, because the brand connotes two things at once: A promise of what we will do for you, and an aspiration of who you become or which tribe you join by choosing us.

The promise of what the brand will do is now more verifiable. The customer's AI checks every promise before purchase, searching the wider internet, cross-referencing user reviews and triangulating from multiple public sources. The bank that claims to be customer-first has its complaint-resolution record extracted from annual reports or instantly summarised from X. The insurer that promises easy claims has its claim-rejection ratio surfaced against peers. Promises that survive verification strengthen the brand. Promises that do not are revealed in seconds.

The aspiration is not in the product, it lives in the customer. People buy Apple products partly because Apple-ness signals something about themselves. Customers bank with a private bank not for any major service benefit but for who else banks there. People ride Royal Enfield partly because owning one says something no specification sheet captures. AI audits the promise. It cannot interpret the tribe.

The split deepens in the AI-to-AI world, where the customer's AI transacts with the institution's AI. The customer only experiences the outcomes, settlement speed, dispute rate, complaint-resolution time and median application-to-approval. It does not watch advertisements. In this perspective, brands need to invest in advertising that bolsters tribal loyalty, but advertising that is supposed to bolster the promise is now less important.

How it impacts labour

Like the brand, the professional is going to be hit by the AI wave. The professional of the old world was, by training, organised around the domain. The cardiologist gave the right answer to the heart in front of her, not to the medicine cabinet or the financial situation at home. The mutual fund agent recommended within his manufacturer's product set, not against the seven existing funds in the customer's portfolio. The AI can see the landscape comprehensively and the professional has to now compete with it.

In most domains the gap between the bottom and the top decile of professionals has been wide. AI compresses the gap from below. The advisor who tests his recommendation with AI before delivering it catches the portability clause he had not considered. The relationship manager who has the model argue the customer's case against his own pitch will close more often. Every word the professional says can now be cross verified; the smartest professionals will go up against 'unsophisticated consumers' with more respect.

In recent research, Brynjolfsson, Li and Raymond (2023) studied 5,179 customer-support agents at a Fortune 500 software firm and found that average productivity rose 14% with access to a generative AI assistant. Within this overall average, novice and low-skilled workers improved by 34%, while experienced and highly skilled workers showed minimal gain. By this reasoning, AI-powered unskilled labour will be tough competition against skilled practitioners.

The pattern is visible in India also. At one NBFC, the productivity of fresh-college LAP underwriters rose 40% with AI assistance, while experienced underwriters showed minimal gain. The bottom of the labour quality distribution comes closer to the top.

As the floor rises, the implications are twofold. First, the customer's worst experience disappears, and with it the customer's reason to switch providers. Second, the professional's competence becomes the table stakes. They have to now provide what AI cannot supply: Trust earned over time, judgement under ambiguity, the willingness to take a customer-friendly call when things are not going well with the business. These are the qualities that live on the aspiration side of the brand, and they are the real moats of professional competence.

When the asymmetry reverses

So far the shift has run one way, AI in the customer's hand against the institution. The same architecture runs the other way too. Institutions have always known what customers did. But AI interactions reveal something deeper, what the customer considered doing. The questions asked, the scenarios tested, the decisions abandoned. These are cognitive traces, and they sit closer to intent than anything an institution has had access to before. If applied to underwriting, pricing or customer acquisition, they create a new informational advantage that did not exist a year ago.

The trust consequence runs deeper than the privacy one. Behaviour is what the customer did once it was done. Intent is what she rehearsed before she was ready to be seen. A customer can absorb the knowledge that her behaviour was logged. When she learns the institution priced her, or declined her, on the strength of her question, the breach is of a different order. The first asymmetry was about information she did not have. This one is about information she did not know she was giving. This may lead the customer to stop being candid with the one tool that was working for her, because she now suspects it is also working for the firms. The independence this piece began with is the first thing she loses when the architecture turns around. We will have to face a new world of consumer protection complexity, going beyond the simpler questions of data privacy.

What then survives in trust? Not the part that rested on the customer's inability to check. What survives is what AI cannot manufacture. Judgement under ambiguity, the call no model will take responsibility for. A relationship proven over time, the banker who backed the customer through a bad cycle and was proved right. The human presence in a hard moment, the advisor who delivers difficult news with care. None of this can be read off a document, so none of it can be verified, and so none of it can be commoditised. The trust that survives is the trust that was never about information in the first place.

The institutions that Akerlof described were built on substitutes for verification. Those substitutes served a purpose. They filled a gap the customer could not fill herself. That gap is closing. What remains when the substitute is no longer needed is the thing the substitute was always standing in for. Genuine expertise, honestly applied, in the customer's interest. The institutions that had that all along have nothing to fear from the informed customer. The institutions that were selling the substitute will find, quietly and permanently, that the customer has stopped calling back.

The scarce asset is the question

If one risk is that the institution reads the customer's question, the other is that she asks a wrong question. While AI can reduce the information asymmetry, friction has not entirely disappeared. It rests on arriving at the right question to ask. Speed and convenience applied to the wrong question produce a confident wrong answer faster. The customer who compares home loans on interest rate alone misses the prepayment clause. The customer who has already decided to switch insurers asks questions that confirm the decision. The informed customer is powerful. The misinformed customer with AI is powerfully wrong.

The question itself is the unclaimed opportunity. No bank will build a question set that surfaces its own prepayment clause weaknesses. No insurer will build a question set that exposes its own claim-rejection ratio. This needs independent actors: non-profits, researchers, consumer bodies, who can make a GitHub for the questions consumers should be asking financial institutions.

What the Boardroom should debate

Every earlier wave of technology was institution first. Computing, internet, mobile was adopted and absorbed by the organisation and then passed onto the customer on their terms. AI exploded in consumers' hand, 100 million users in two months. The Board is now governing businesses where customers will know as much if not more than the organisation.

The first institutional response to AI has been operational, adding chatbots, analytical tools, dashboards, voice bots, service automation, that are good cost-saving initiatives. The strategic question is the viability of the business model itself.

What happens when the customer arrives informed? Which elements of the value proposition survive verification? Which revenue streams depend on customer ignorance or high search costs? Which promises would survive an AI audit? Which parts of the sales process assume an information asymmetry that no longer exists? These are not technology questions for the CTO. They are business-model questions for the Board.

The institutions that emerge stronger will not be those that adopted AI fastest. They will be those whose value remains after verification becomes cheap.

References

Akerlof, George A. (1970). "The Market for 'Lemons': Quality Uncertainty and the Market Mechanism". Quarterly Journal of Economics, 84(3), 488-500. https://www.jstor.org/stable/1879431

Brynjolfsson, Erik, Danielle Li and Lindsey R. Raymond (2023). "Generative AI at Work". NBER Working Paper 31161. https://www.nber.org/papers/w31161

Reuters / Similarweb (2023). ChatGPT user-growth figures. https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/

Reserve Bank of India. Annual Report of the Ombudsman Scheme. https://www.rbi.org.in/Scripts/AnnualReportPublications.aspx

Maninder Singh Juneja is a partner at True North. He serves on the boards of Pine Labs, Nivara Home Finance and Integrace, and has previously served on the boards of Niva Bupa Health Insurance, Federal Bank Financial Services and HomeFirst Finance. The author thanks participants at a talk at XKDR Forum for myriad good ideas, and Ajay Shah, Renuka Sane and Aditi Mascarenhas for comments on earlier drafts.

Chinks in market efficiency: A Melody story

by Ajay Shah and Atibhi Sharma.

The main paradigm in finance is the efficient market hypothesis which suggests that as soon as there is new information, the news is rapidly incorporated in the price. Markets are generally quite efficient, it's hard to find opportunities for supernormal returns.

But there are chinks in the armour. There are clear examples, worldwide, where market prices have been clearly wrong.

In this article, we show a recent Indian story which should be added into the Mistakes of Markets catalog. The share price of Parle Industries soared following the video of Italian prime minister Giorgia Meloni receiving a packet of Parle Products' Melody, gifted by Indian prime minister Narendra Modi. Parle Products, the maker of Melody, is an unlisted private entity. We also show some interesting global examples and offer some thoughts on understanding finance.

The "Signal" Ticker Confusion (2021)

In January 2021, after WhatsApp changed its privacy policy, Elon Musk tweeted 'Use Signal', referring to the free, open-source, and heavily encrypted messaging app. This pushed noise traders to purchase shares of Signal Advance (SIGL). The actual Signal app is managed by the non-profit Signal Foundation and is not publicly traded. Signal Advance, on the other hand, is a tiny medical detection devices manufacturer in Texas.

SIGL was an illiquid penny stock. It surged from \$0.60 to \$70.85 within three days—an 11,700% surge within three days. It took the markets six to nine months to return to pre-tweet levels.

Clubhouse Media Group (2021)

In the same month, on January 31, 2021, Elon Musk tweeted that he would be joining a room on Clubhouse, the then-viral invite-only audio chat application. The next day, shares of Clubhouse Media Group (CMGR), a penny stock entirely unrelated to the privately held Clubhouse app, surged. CMGR had recently rebranded itself as a marketing agency from a healthcare firm. The stock opened at \$10.85 on February 1, 2021, up from \$1.50 - \$2.00 from the previous days. It reached an all-time high of \$28.43 on February 15, 2021 (a near 1,400% increase from its pre-tweet baseline) and it took roughly four months to return to its pre-tweet level.

Zoom Video vs Zoom Technologies (2019-2020)

When Zoom Video Communications (ZM) filed for its IPO in 2019, investors rushed to buy shares, accidentally purchasing the shares of a defunct mobile phone parts manufacturer called Zoom Technologies (Ticker symbol: ZOOM), a penny stock trading at \$0.005. The stock surged 54,000% to around \$5. This confusion was not a one-off incident, as a similar trend was observed during the lockdown in the pandemic of 2020 when ZOOM went up 1,800% to \$20.90 until the US Securities and Exchange Commission physically intervened, suspending trading for Zoom Technologies for 10 business days and forced a ticker symbol change to ZTNO.

Bombay Oxygen Investments (April 2021)

During the second wave of COVID-19, when there was a shortage of medical oxygen in India, noise traders started looking for oxygen in the market. They landed on Bombay Oxygen Investments, a Non-Banking Financial Company that had exited the oxygen manufacturing business in August 2019 and had received RBI's registration certificate on December 31, 2019 for the same. The stock surged 131.3% in under 12 trading sessions, rising from Rs 11,025 on March 31 to Rs 25,500 intraday on April 20 and then, fell by ~ 50% to 12,700 in August of the same year.

L. G. Balakrishnan (2025)

In 2024, shares of LG Balakrishnan & Bros (a manufacturer of automotive chains) surged to a 52-week high because noise traders mistook it for the upcoming IPO of the consumer electronics giant LG Electronics India. The mispricing lasted only a day as investors realised their mistake.

Pan-Homophonic Events

Zhang et al. (2026) documented a new phenomenon of pan-homophonic events, where confusion between linguistics, trending keywords and stock names triggers sudden market volatility, specifically for a Chinese technology firm named Chuan-da-zhi-sheng. Since "Chuan-pu" is a loose and phonetic translation of Trump in Mandarin, noise traders phonetically interpreted the company's name to mean "Trump wins big". This unrelated traffic software stock essentially became a trading proxy for US political events. In 2016, when Donald Trump won the US presidential election, shares of the company surged 7.6% in a single day and then again in 2024, following major turning points in the US election cycle, the same stock repeatedly hit its maximum 10% daily upper circuit limit.

Ticker Confusion and the Limits of Arbitrage

Balashov and Nikiforov (2019) documented the systematic nature of these mix-ups. Investigating 254 pairs of stocks, they found that erroneous trades account for roughly 5% of all trading turnover in the smaller "shadow" companies. A classic example is Ford Motor Company (Ticker: F). Investors systematically assume its ticker is 'FORD' - which is actually the trading symbol for Forward Industries, a micro-cap manufacturer of carrying cases for medical devices. Similarly, a paper by Rashes (2001) studied the mass confusion between MCI Communications and a completely unrelated fund with the ticker MCIC. They found that while the co-movement between the two similarly-named stocks is statistically significant, it is not something that arbitrageurs can easily exploit because shorting an illiquid shadow company is costly.

Parle Industries

The Parle Industries episode is the latest entry in this ledger. Parle Industries is not a confectionary giant; it is a micro-cap company involved in infrastructure development, real estate, and paper waste recycling. Prior to the viral Modi-Meloni video, its market capitalization hovered around Rs.360 million, the price of a few apartments in Bombay. It was a penny stock with a share price of about Rs.5 and an average trading volume of 20,000 to 60,000 shares daily in the 6 months prior window.

• 9:30 AM (IST) - The Baseline: Market opened in India. Parle Industries opened at INR 4.95.
• 10:00 AM (IST) - The Event: Italian Prime Minister Giorgia Meloni uploaded a video to her official Instagram account and then her X account with the caption, "Thank you for the gift."
• 10:00 AM to 3:35 PM (IST) - Amplification: The video went viral. On X, the hashtag #Melodi trended, and the reel became the most viewed reel on PM Meloni's Instagram account. Simultaneously, Google Trends intraday data showed a spike: searches for "Parle share" and "Melody".
• 3:30 PM (IST) - Noise trading: By the market close, Parle Industries was locked into a 5% upper circuit closing at INR 5.25. It closed at a volume of 857,248 shares.

The stock hit the upper circuit for five consecutive trading sessions. BSE historical data shows that from May 21 through the end of the month, the stock's delivery percentage reached exactly 100%. The noise traders were taking delivery, believing it was worth holding this for multi day horizons.

Market Efficiency and the Role of Liquidity

These are examples of how prices can go wrong, exposing failures in market efficiency.

A better interpretation of market efficiency comes from focus on how clever people could exploit the mistakes of the noise traders. Here, we see the problems of financial market completeness (can you take an opposing trade?) and financial market liquidity (is the size of your winning trade big enough to matter?). A market inefficiency that is not exploitable will not be readily solved by the market. With small cap penny stocks there are no single stock derivatives that rational traders can short. In India, stock lending does not work so it is not possible to short sell and profit from the mistakes in the price. To the extent that better financial economic policy increases liquidity, it will, in turn, increase access to the correct tools for trading (single stock derivatives and stock lending). As a result, these problems will be diminished.

These problems are a reminder of the difficulties of small capitalisation stocks. Financial market trading works extremely well for large firms. We may perhaps apply a thumb rule in India of a minimum point of a market capitalisation of Rs.10 billion. But we do wrong to assume it is equally useful and equally effective for small firms. There is a certain social justice instinct in India, where we like to bring the glory of stock market listing to small firms, thinking that we are giving a helping hand to a weak firm. We need to be more cautious in the usefulness of this approach.

Financial markets are a remarkable information processing system. It is easy to disrespect the drama that is ceaselessly afoot. What is going on is that millions of clever people have been harnessed to constantly look at the world and make prices. These prices are the commanding heights of the economy and shape the resource allocation. Markets are not perfect, they are the best aggregation of what humans can figure out based on their self-interest.

Bibliography

Parle Industries' upper circuit to Signal's 5,100% surge: 5 mistaken stocks that triggered market frenzy, Surabhi Pandey, Moneycontrol, 20 May 2026.

#Melodi trends on X as PM Modi gifts Melody toffees to Italian PM Giorgia Meloni, DH Online, Deccan Herald, 20 May 2026.

190 Million And Counting: Meloni's Melody Moment With PM Modi Is Mega Viral, Abhinav Singh, NDTV, 21 May 2026.

Publicly Listed Zoom Video Communications: Traders Buying Zoom Technologies, Jonathan Garber, Markets Insider, 18 April 2019.

Want to Invest in the Zoom IPO? Make Sure You Buy ZM, Not ZOOM, Minda Zetlin, Inc.com, 18 April 2019.

Traders mistakenly invest in Clubhouse Media Group after Elon Musk tweets about a separate, private app with the same name, Natasha Dailey, Business Insider, 2 February 2021.

COVID-19: Bombay Oxygen shares up 256%; it doesn't even make oxygen, Business Today, 20 April 2021.

Massively Confused Investors Making Conspicuously Ignorant Choices (MCI-MCIC), Michael S. Rashes, The Journal of Finance, Vol. 56, No. 5, 2001.

How much do investors trade because of name/ticker confusion? Vadim S. Balashov and Andrei Nikiforov, Journal of Financial Markets, Vol. 46, 2019.

Quantifying the Linguistic Complexity of Pan-Homophonic Events in Stock Market Volatility Dynamics, Yunfan Zhang, Jingqian Tian, Yutong Zou, Xu Zhang, and Xiao Cai, Entropy vol. 28, no. 1, 12 January 2026.

Mistaken Identity: LG Balakrishnan Shares Surge as Investors Confuse It for LG Electronics India, Nishanth Vasudevan, Economic Times, 15 October 2025.

The authors are researchers at XKDR Forum. The authors would like to thank Susan Thomas, Amrita Agarwal, Aditi Mascarenhas and Jay Kulkarni for their valuable feedback and discussions on this piece.

A Market Failure Framework for Evaluating Public Sector Undertakings

by Arjun Krishnan.

Commentators and investors often judge companies, including state-owned firms, by their profitability. While profit is a useful metric for private firms focused on generating returns for owners, applying the same standard to state-owned enterprises is problematic. Many lament that India's Public Sector Undertakings (PSUs) incur losses, assuming that losses signal failure and profits signal success. However, this assumption misjudges the real purpose of PSUs. Although some evaluation frameworks expand beyond profit, few explicitly align performance criteria with the specific market failure that the enterprise was established to address. This article argues that assessing PSUs solely on profitability is misguided, and their success should be measured by how well they address the public purpose for which they were created.

This article proposes a two-part framework for evaluating PSUs. The first part poses an ex-ante question about purpose. A PSU is justified only when it aims to correct a market failure that less intrusive instruments cannot correct. The second part poses an ex-post question about performance. Evaluators should then judge a justified PSU on two dimensions: efficiency and effectiveness. Efficiency measures how productively an enterprise converts resources into outputs. Effectiveness captures whether the PSU actually corrects the failure it was created to address. Balance sheets cannot serve as a reliable proxy for either dimension on their own.

Ex-ante: when is a PSU justified?

The justification for any PSU must begin with market failure. When markets function well, they allocate resources efficiently, and the state has no grounds to intervene. Economists identify four situations where markets fail to do so. First, externalities arise when a cost or benefit of an economic activity falls on an unrelated third party. Positive externalities lead to underprovision, and negative externalities to overproduction. Second, public goods are non-excludable and non-rivalrous. Firms cannot easily charge users, and private markets typically underprovide them. Third, information asymmetry occurs when one party to a transaction knows more than the other, distorting decisions and reducing market efficiency. Fourth, market power arises when limited competition allows firms to raise prices or restrict output below socially optimal levels.

A market failure on its own does not justify a PSU. There are three additional tests. First, scale: how many people are affected, and by how much? A localised information asymmetry in a niche market is different from one that excludes millions from credit. Second, persistence: is the failure temporary and self-correcting, or structurally durable? Markets sometimes endogenously mitigate their own failures through competition, reputation, or contracting. Exogenous forces such as technological innovation or institutional adaptation can have the same effect. Non-state mechanisms such as industry associations, cooperatives, or third-party certifiers may emerge to address coordination problems or information asymmetries without government ownership. Even classical public-good cases have been addressed without state ownership. Coase's (1974) account of English lighthouses is a canonical illustration: what was treated in classical economics as a pure public good requiring state provision was, in fact, supplied for centuries by Trinity House, a private body that collected dues from ships at port. The presence of such adaptive mechanisms weakens the case for a PSU. By contrast, failures that persist despite opportunities for institutional adaptation present a stronger case for public intervention.

Even when a market failure is large-scale and persistent, the state has many ways to respond. The third test, then, is instrument choice: Is ownership the right way to address this failure? The state can regulate, tax, subsidise, or contract with private providers. Ownership is one of the most costly options. Ownership exposes the exchequer to operating losses, creates a vehicle vulnerable to political capture, and softens the budget constraint in ways regulation and subsidy do not. The case for ownership has weakened with experience. Publicly owned natural monopolies in many sectors turned out to deliver less output for a given level of inputs than the textbook treatment had suggested. Regulatory practice has grown more sophisticated, with sector-specific knowledge and administrative law tools that did not exist when many PSUs were created. Public-private partnerships have further narrowed the cases for state ownership, with private firms providing goods and services under contracts that set market structure, pricing, and quality. For every PSU, the central question is why the problem could not be addressed through one of these alternatives.

India operates 291 Central Public Sector Enterprises across sectors as varied as petroleum refining, power transmission, hotel management, and defence manufacturing. The policy debate about this universe has been conducted in terms of profitability. How many are loss-making? What do aggregate losses cost the exchequer? These questions are downstream of a prior one: which market failure, if any, justifies each enterprise. Those that address no market failure have no business existing and should be sold off to buyers or wound down, with assets liquidated. Those that aim to correct a market failure face a harder question: how well do they perform in addressing the failure they are expected to correct?

Ex-post: efficiency and effectiveness

Two questions emerge for judging how well a PSU is performing. The first is efficiency. Efficiency is the ability to derive the greatest possible output from a given quantity of financial, physical, and human resources. The second is effectiveness. Since a PSU is justified only for market failures, we need to evaluate whether it, in fact, corrects the failure it was created to address. A PSU can be efficient at producing what a competitive market would produce anyway, or effective at reaching its target population at three times the cost a regulated private operator would charge. Both outcomes are undesirable. In the first case, the PSU adds little social value. In the second, it imposes unnecessary costs to achieve a legitimate public objective.

Experience with direct public provision over the last half-century has weakened the case for PSUs on efficiency grounds. Publicly owned natural monopolies in many sectors exhibited poor x-efficiency, producing less output for a unit of input than private firms. The Ministry of Road Transport and Highways reports that the revenue-to-cost ratio for the 58 reporting undertakings fell to 63.6% in 2021-22, that state cabinets blocked fare revisions, and that the resulting losses reached Rs 30,192 crore in aggregate. The mobility problem the SRTUs were created to address is real, but the case for state ownership of the operator is much weaker than the case for state involvement in the sector through options like subsidies.

The regulatory and contracting alternatives to ownership have grown more capable over the same decades. The regulatory state now possesses sector-specific tariff and quality regulation, administrative law procedures for rule-making, and incentive-compatible contracting techniques (Laffont and Tirole, 1993). Public-Private Partnership (PPP) models have spread across sectors once considered the natural home of direct provision. Iossa and Martimort (2015) show that bundling construction and operation into a PPP can be efficient when build quality materially lowers operating costs. This structure is common in roads, water systems, and many public utilities. Even classical public goods, including urban streets and water supply, are now routinely constructed and operated through PPP agreements that specify quality and pricing. The Government of India's disinvestment policy lists market imperfections and public purpose as criteria for retaining a PSU in public hands. The government excludes profitability as a criterion.

For certain types of goods, state ownership may be the preferable option. When contract terms cannot specify aspects such as quality, the case for ownership over contracting strengthens (Hart, Shleifer, and Vishny, 1997). A private operator paid to deliver an output will cut costs along whatever margins the contract leaves unspecified. Where quality is one of those margins, the cost saving comes at the consumer's expense. A private prison contractor's contract may specify calorie counts and dietary variety, but regulating food quality or taste may prove impossible. These savings flow to the contractor while the welfare loss falls on inmates. Direct public ownership is preferable in such settings precisely because the public manager's weaker incentive to cut costs leaves the unspecified quality dimensions intact.

In addition to efficiency, effectiveness needs to be judged. Effectiveness measures whether the PSU is correcting the market failure it was created to address. To illustrate the difference, consider a state-owned bus operator tasked with providing transport connectivity to remote rural areas. An efficient operator minimises the resources needed to run the service. An effective operator ensures that the targeted rural communities are actually connected. A PSU may succeed on one dimension while failing on the other.

The effectiveness criteria take different forms across failure types because the welfare yardstick differs. The market power row needs some additional explanation. A monopolist with declining average costs cannot price at marginal cost without losses. Ramsey-Boiteux pricing sets the loss-minimising alternative: markups above marginal cost should rise as demand elasticity falls, placing the heaviest charges on users whose consumption is least price-sensitive. A markup on inelastic demand reduces output the least and destroys the least surplus per rupee of revenue. A political cross-subsidy structure follows a different logic, allocating markups across user groups by political weight rather than by elasticity. The effectiveness test for a PSU that disciplines market power, therefore, asks whether its markup structure approximates Ramsey-Boiteux rather than political cross-subsidy.

Table 1 sets out the effectiveness criterion for each market failure, with the less intrusive instrument serving as the comparator.

Table 1: Effectiveness criteria for PSUs by type of market failure

Market failure	Market problem	Role of PSU	Effectiveness criterion	Less intrusive instrument
Externalities (positive)	Producers cannot capture the full social benefit, so the private market undersupplies relative to the social optimum.	Produce at a level that accounts for spillovers private producers ignore, or finance investments whose social returns exceed appropriable private returns.	Is the targeted output being delivered, and is the additional supply above the private optimum sufficient to close the externality gap?	Production subsidies, tax credits, intellectual property protection, advance market commitments.
Externalities (negative)	Private producers impose costs on third parties they do not bear, so the market overproduces relative to the social optimum.	Produce at a level that internalises external costs private producers would otherwise externalise.	Has the targeted reduction in harm been achieved, and are emissions per unit of output below the unregulated counterfactual?	Pigouvian tax, tradable permits, command regulation.
Public goods	Private producers cannot exclude users from a non-rivalrous good, so the market undersupplies or fails to supply.	Provide the good where private cost recovery is impossible or inefficient.	Is the service reaching the target population, and is coverage approaching the welfare-maximising level?	Contracting with private providers under a public service obligation.
Information asymmetry (seller knows more)	Private sellers hold information about quality that buyers cannot observe, so low-quality goods crowd out high-quality ones.	Enter the market and disclose costs, quality, and pricing as a benchmark that private sellers would otherwise suppress.	Has the PSU's presence made quality observable to buyers and sustained transactions that would otherwise have unravelled?	Mandatory disclosure regulation, third-party certification, independent benchmarking authority.
Information asymmetry (buyer knows more)	Buyers hold private information about themselves that sellers cannot verify. Sellers respond by raising prices, rationing, or withdrawing supply.	Offer service to groups private firms avoid because they cannot distinguish high-risk from low-risk customers.	Is the PSU enrolling the high-risk groups private markets exclude, and is the share of high-risk individuals covered higher than under the private counterfactual?	Risk-pooling mandates, mandatory insurance schemes.
Market power	Private firms price above competitive levels or restrict output below the social optimum.	Compete to discipline private pricing. In a natural-monopoly case, price at the welfare-optimal level.	Is the PSU pricing closer to marginal cost than an unregulated monopolist would, and does the markup structure approximate Ramsey-Boiteux rather than political cross-subsidy?	Competition law and antitrust enforcement, sectoral price regulation, separation of monopoly network from competitive services.

A PSU can fail on either dimension or both, and each case calls for a different response. A state-owned bus operator that abandons remote routes for crowded urban corridors is efficiently delivering something the market can deliver. Efficient delivery of a service that the market would have provided is no justification for state ownership. A PSU that effectively corrects a market failure but does so at an unjustified cost may generate more welfare loss through waste than welfare gain from correcting the failure. Reform or contracting out may be the appropriate response.

Soft budget constraints

PSUs operate under what Kornai (1998) called a soft budget constraint. A private firm that runs at a persistent loss is likely to go bankrupt. A PSU does not, because the PSU expects the state to cover the shortfall. The expectation of rescue weakens the discipline that revenues and costs would otherwise impose on managers.

A PSU pursuing a legitimate mandate should not show sustained accounting losses on its own books. Where pricing reflects deliberate policy, including selling output below cost for welfare reasons, the resulting deficit is a transfer from the treasury to the consumer. The right place for that transfer is the government's expenditure account, recorded as an explicit subsidy and matched by income on the PSU's accounts. Accounting separation keeps the cost of social policy visible in the budget, where parliamentarians can scrutinise it, rather than in an enterprise's operating accounts.

The Food Corporation of India illustrates what happens when this discipline breaks down. The Corporation procures grain at minimum support prices set by the Cabinet, stores it, and supplies it to ration shops at prices well below procurement cost. The shortfall is intended to be transferred to FCI as a food subsidy from the union budget. However, for long stretches, the government did not transfer the full subsidy in time, and FCI raised debt, much of it from the National Small Savings Fund, to bridge the gap. This practice was especially prevalent between 2016 and 2021. The losses that appeared on FCI's books reflected the failure to transfer the subsidy on time. The government used FCI's balance sheet to delay recognition of expenditure that should have appeared in the budget.

The existence of PSUs can then soften budget constraints in two ways. First, if a PSU knows it can rely on transfers from the treasury to cover any shortfall, its managers have less incentive to contain costs than managers of a private firm would. As a result, the PSU's budget constraint is softened. Second, PSUs like FCI soften the budget constraint of the government that created them. Since the enterprise absorbs costs that should have appeared in the budget, the state's social spending is understated. Once this practice exists, the headline profit or loss of a PSU carries less information. A loss-making PSU may be one that delivered the mandate but did not receive the subsidy. A loss-making PSU may also be wasteful. Which case applies cannot be understood based on the profit and loss account. Persistent losses on a PSU's books are therefore a useful diagnostic. They suggest either operational inefficiency or off-budget accounting through the PSU's balance sheet.

The framework, so far, treats PSUs as faithfully pursuing their mandates. They often do not. Even an enterprise with a legitimate market failure justification is run by people with their own interests. Governments are themselves made up of self-interested actors, and political objectives can capture an enterprise created to address a market failure. These government failures show up in the performance criteria above: prices far from welfare-optimal levels, investments that do not deliver the promised social benefits, or operations that consume more inputs than the next-best instrument would have required.

Government failure also shapes how the state manages exit from enterprises that have outlived their justification. A framework that identifies when a PSU has no reason to exist is useful only if exit decisions follow honestly from that assessment. Chakrabarty (2023, page 9) finds that around 43% of India's disinvestment proceeds between 1991 and 2022 involved no actual transfer to private hands. Shares were transferred between public entities, and the proceeds were counted towards the disinvestment target without any change in underlying ownership. Even where the case for a PSU has lapsed, political incentives corrupt the exit process and sustain enterprises that should not exist.

Applying the framework

Consider three applications of the framework. First, the electricity transmission network exhibits natural-monopoly characteristics, with high fixed costs and declining average costs over the relevant output range. State intervention to address market power is justified ex-ante, whether through regulated private ownership or direct public ownership of the grid operator. Power Grid Corporation of India is a PSU that runs the inter-state grid. From the standpoint of allocative efficiency, tariffs should be set close to marginal cost. Because the marginal cost of transmission is below the long-run average cost, such pricing would generate a structural revenue deficit. A deficit of this kind would not necessarily signal operational inefficiency. It can reflect a deliberate tariff policy that expands access and maximises network use, and it would be justified where it represents the least-cost route to the social objective when compared with direct transfers or alternative subsidy mechanisms. India does not price transmission at marginal cost. Tariffs are set under the CERC Tariff Regulations, 2024, which use a cost-plus framework. For new transmission projects, Regulation 30(3) provides a base return on equity of 15%, along with recovery of interest costs, depreciation, interest on working capital, and operating and maintenance expenses. The PSU's operational record is strong: the transmission system was available 99.85% of the time in FY24 across 1,77,699 circuit kilometres carrying around half of India's inter-state electricity. Whether the cross-subsidy implicit in cost-plus regulation is the least-cost route to network expansion, or whether a direct transfer would deliver the same access at lower fiscal cost, is a question worth asking. Power Grid no longer builds new lines by default. The regulator auctions each new project to the lowest bidder, and Power Grid competes for these contracts alongside Adani, Sterlite and Tata Power. Private bidders win a growing share. Whether the legacy network would also be cheaper in private hands is a separate question. The framework's verdict on Power Grid therefore turns on an empirical question: whether the cost-plus regulation of the legacy network delivers cheaper transmission than competitive procurement would.

Second, the India Tourism Development Corporation (ITDC) runs the Ashok Group of Hotels. ITDC was established in 1966 to develop tourist infrastructure, including hotels. The Taj, Oberoi, ITC, Lemon Tree and Marriott chains, among others, operate across the segments and price points ITDC serves. Hotels are not a public good. Private operators can charge customers, competition is adequate, and there are no externalities, information failure, or market power problems that require the state to own a hotel chain. The market failure test fails at the first step, so no ex-post analysis is needed.

A third example concerns the post-independence wave of public investment in heavy industry and infrastructure. The standard defence of state ownership in this period rested on the absence of capital markets: long-term finance was scarce, and only the state could mobilise it. Bhagwati and Desai (1970) contested this claim, arguing that private capital existed and that the licensing regime was producing the shortages it purported to remedy. Whatever the merits of the original argument, India's capital markets have since deepened. A second defence rests on positive externalities through learning effects and supply-chain spillovers, where social returns exceed what a private investor can appropriate. Underinvestment results from this appropriability gap, even when capital is available. The defence still has limits. Where production subsidies, intellectual property protection, or advance market commitments can close the gap, ownership is the costlier instrument. A surviving PSU founded on these grounds must show that such alternatives remain inadequate.

India's early integrated steel plants at Rourkela, Bhilai, Durgapur, and Bokaro are concrete cases. The plants were justified on grounds that private firms could not raise the long-term capital required, and that they would generate large downstream spillovers through skilled labour, supplier networks, and engineering capabilities whose full social value private investors could not capture. The appropriability gap was real, and thin capital markets compounded the problem by raising the cost of private investment. Both conditions have since changed. India's capital markets have deepened, project finance has matured, and Tata Steel and JSW Steel have built modern integrated capacity at scale. The spillovers that public investment was meant to generate now flow through the private steel industry instead. SAIL, the operator of the original plants, produces around 15% of Indian steel and is profitable. Profitability does not save it from the framework's test. Where private operators produce the same steels at a comparable scale, the market failure has been resolved, and continued public ownership no longer has a justification.

The way forward

Profit is the wrong measure for judging a PSU. It speaks neither to whether the enterprise should exist nor to whether it is doing what it exists to do. For each of India's 291 central PSUs and more than a thousand at the state level, the question is whether a market failure persists, whether ownership is the cheapest way to address it, and whether the enterprise actually does so. Some profitable PSUs would fail this test. Some loss-making ones would pass.

References

India: Planning for Industrialization: Industrialization and Trade Policies Since 1951 by Bhagwati J and Desai P, 1970, Oxford University Press for OECD Development Centre.

On the Management of Public Monopolies Subject to Budgetary Constraints by Boiteux M, 1971, Journal of Economic Theory, 3(3), 219 to 240.

The Lighthouse in Economics by Coase R H, 1974, Journal of Law and Economics, 17(2), 357 to 376.

The Proper Scope of Government: Theory and an Application to Prisons by Hart O, Shleifer A and Vishny R W, 1997, Quarterly Journal of Economics, 112(4), 1127 to 1161.

The Simple Microeconomics of Public-Private Partnerships by Iossa E and Martimort D, 2015, Journal of Public Economic Theory, 17(1), 4 to 48.

In Service of the Republic: The Art and Science of Economic Policy by Kelkar V and Shah A, 2019, Penguin Allen Lane.

The Place of the Soft Budget Constraint Syndrome in Economic Theory by Kornai J, 1998, Journal of Comparative Economics, 26(1), 11 to 17.

A Theory of Incentives in Procurement and Regulation by Laffont J-J and Tirole J, 1993, MIT Press.

Allocative Efficiency vs. "X-Efficiency" by Leibenstein H, 1966, American Economic Review, 56(3), 392 to 415.

State versus Private Ownership by Shleifer A, 1998, Journal of Economic Perspectives, 12(4), 133 to 150.

Arjun Krishnan is a consultant at the Centre for Civil Society, a Delhi-based think tank. He thanks Sourya Banerjee for the early conversations that inspired this article, Jayana Bedi for her thoughtful feedback during its drafting, and an anonymous referee whose comments considerably sharpened the argument.