Why India's largest companies cannot say where their CSR goes, and how a five-rung test would show whom it reaches
Abstract
Since 2014, India has required its largest companies to spend 2 percent of their profits on corporate social responsibility, and the amount is now substantial: the National CSR Portal records Rs 40,131 crore (US$4.80 billion) spent in 2024-25. Because the law names social ends, including ending hunger and poverty and reducing the inequalities faced by socially and economically backward groups, we can judge its money only by who received it and what changed for them. The filings that account for it answer a different question: where the money went, and for most of the money, they do not even answer that. This working paper tests the record against a five-rung ladder: where the money landed, who it passed through, who received it, what position they held relative to the rule that selected them, and whether it made a difference in their lives. For the 357 companies that spent at least Rs 10 crore between 2014 and 2020, Rs 55,857 crore (US$8.25 billion) in all, only 11.8 percent of the money names a district, and 302 of the 357 companies record a district for less than half of what they spent. Of the 55 companies that do record where, 26 fail a test of reach to the tail end, and 10 of those reach aspirational districts but not the tribal-majority districts within them. No filing reaches the four higher rungs. The paper argues that the gap is not a shortage of money or goodwill but a design choice in what the law asks companies to record, and it sets out how the Tail-End Assessment would close it.
Keywords
corporate social responsibility; Section 135; impact assessment; aspirational districts; Scheduled Tribes; intermediaries; disclosure; India
Interactive map
Where CSR lands: interactive map
Suggested citation
Ramprasad, V. 2026. Where, but not who: Why India's largest companies cannot say where their CSR goes, and how a five-rung test would show whom it reaches. Tailwinds Working Paper 2026-03, Tailenders Analytics, v0.1, 12 October 2026. https://doi.org/10.5281/zenodo.23307292
Working papers share research in progress to invite comment; this version has not been peer reviewed. Comments are welcome through tailenders.org. © 2026 Tailenders Analytics. Licensed under CC BY 4.0: share and adapt with attribution.
Corporate social responsibility is defined by its impact on society. The activities that Schedule VII of the Companies Act, 2013 allows companies to fund, including eradicating hunger, poverty, and malnutrition; promoting education; and measures to reduce inequalities faced by socially and economically backward groups, change people's lives, not places on a map. The test of CSR money is therefore social: how society, and which part of it, has been changed by it. That test cannot be run without knowing who received the money. Where tells an auditor that a rupee was spent; who tells the public whether it was spent on the people the law had in mind. Only once the who is known can the outcome the law names be evaluated. The CSR Rules already turn on who: since 2021, they exclude from CSR activities benefitting a company's own employees. A rule that excludes some beneficiaries cannot be enforced, let alone evaluated, without knowing who the beneficiaries were.
A district names a place. It does not name a person, a position, or a change. India's CSR regime, the first in the world to make corporate social spending a statutory obligation, has built an accounting system that can locate a rupee, when it bothers to, and cannot follow it any further. The result is a public record that tells a board what it spent and tells the people at the tail end of the distribution nothing at all.
This paper makes three claims. First, the record runs out at the first rung: the largest companies, those Rule 8(3) obligates to assess their impact, cannot say where most of their CSR went. Second, where the record does run, it shows reach that stops short of the tail: money that arrives in an aspirational district and misses the tribal-majority districts within it. Third, and most consequential, no filing touches the four rungs on which equity and effectiveness turn: the intermediary that carried the money, the households that received it, their position relative to the rule that chose them, and whether their lives changed. These are not failures of measurement technique. They are what the law chose to ask.
Every benefit that reaches a person climbs five rungs, and each rung has a question that can be answered or left blank.
| Rung | Question | In the CSR record today |
|---|---|---|
| 1. Where | In which district and block did the money land? | Optional in effect; 11.8% of the money of the 357 largest spenders names a district |
| 2. Through whom | Which agency or foundation carried it? | Agency registered since 2021; not publicly linked to the money |
| 3. Who | Which households received it? | Not captured in any filing |
| 4. Position | Could they shape it, know the rule, complain, and be heard? | Not captured |
| 5. Difference | Did their lives change, against a comparison, a year on? | Rule 8(3) assessments above thresholds, reported to boards |
The rungs are ordered, and the order matters. A rupee that cannot be located cannot be traced to an intermediary; an intermediary that cannot be named cannot be asked who it served; a beneficiary who cannot be identified cannot be asked whether the rule that selected her was published, whether she could complain, or whether anything changed. The ladder is therefore also a diagnosis: wherever the record stops, everything above it becomes invisible, and the invisibility compounds.
The analysis uses the Ministry of Corporate Affairs' project-level CSR data for 2014 to 2020, as mirrored by the India Data Portal: 36,664 records of company, district, development sector, and amount spent, covering 553 to 663 companies per year. The file includes only the largest spenders, representing between 49.9 percent and 91.9 percent of the national total, depending on the year, which makes it the appropriate population for this question: these are the companies whose scale brings them under Rule 8(3), the provision of the Companies (CSR Policy) Rules that requires an independent impact assessment where a company's average CSR obligation is Rs 10 crore or more.
Districts were matched to Census 2011 (381 exactly, 31 by spelling, 76 through a logged table of renamed and post-2011 districts, and one record excluded as an error) and to NITI Aayog's list of 112 aspirational districts. A district is tribal-majority if Scheduled Tribes constitute at least half its population in Census 2011. The design was registered and checksummed before any company-level statistic was computed. Implementation choices were logged before the run, and the script reproduces every number in this paper.
Amounts are given in Indian rupees with US dollar equivalents. One crore is ten million rupees and one lakh is one hundred thousand. Spending is converted to dollars year by year at the Federal Reserve's annual average exchange rate for the calendar year in which each financial year begins (release G.5A; FRED series AEXINUS): Rs 60.9953 per dollar in 2014, 64.1073 in 2015, 67.1572 in 2016, 65.0659 in 2017, 68.3734 in 2018, 70.3777 in 2019 and 74.1429 in 2020, with totals summed after conversion. National totals for 2022-23 to 2024-25 use the averages for 2022 (78.5792), 2023 (82.5708) and 2024 (83.6566). Statutory thresholds, which apply today, are converted at the 2025 average of Rs 87.1468: Rs 10 crore is about US$1.15 million and Rs 1 crore about US$114,749.
Of the Rs 55,857 crore (US$8.25 billion) spent by the 357 companies with at least Rs 10 crore of CSR over the period, Rs 6,610 crore (US$982 million), or 11.8 percent, names a district; across all 693 companies in the file the share is 12.4 percent. The rest is recorded as 'Not mentioned', as a national fund, or as spread across states. The median company records a district for 15.6 percent of its money, and 302 of the 357 record one for less than half.
This is not missing data in the ordinary sense. The filing permits the blank, so it is a legitimate answer, and companies that use it are complying. The consequence is structural: for the companies whose CSR is large enough to require an impact assessment, the public record cannot support even the first question an assessment would ask. An assessor can verify that a project happened; nobody outside the company can check where its money went, let alone whom it reached.
The pattern is consistent with what practitioners report about the national total. Crisil Foundation's analysis, as reported, finds that backward districts received 12 percent of the CSR spend it examined for 2023-24, and analyses of the portal attribute the concentration to headquarters and plant locations. The first proviso to Section 135(5) asks companies to give preference to the local areas where they operate; the Ministry has told companies to follow it 'in letter and spirit', and has also told Parliament that the emphasis is 'only directory and not mandatory'. Read together, the proviso explains why money stays near plants, and the permissive filing explains why the record cannot show how far it strays.
Fifty-five companies record more than half their money in a district, and for them the reach test can be run. For each, the share of its recorded money landing in aspirational districts is compared with those districts' share of the population in the states where the company itself chooses to spend; within aspirational districts, the test then asks whether tribal-majority districts receive their share. A company fails where the gap exceeds 5 percentage points, and its 90 percent bootstrap interval lies below zero.
Twenty-six of the 55 fail. Ten of the 26 fail only at the second level: they reach aspirational districts and miss the tribal-majority districts within them. Thirty-two of the 55 recorded no money at all in any aspirational district over seven years.
The figures that follow come from public, open data: the CSR disclosures that companies filed themselves with the Ministry of Corporate Affairs, as published on the National CSR Portal and its mirror. They illustrate a pattern; they do not rank companies, and they describe 2014 to 2020 only. Cases such as these show its shape. Nava Bharat Ventures recorded 92.7 percent of its district-located money in aspirational districts, 70.0 points above their population share, and still fell 12.5 points short at the tribal-majority level. Vedanta recorded 29.4 percent, 12.9 points above the population share, and fell 26.4 points short within it. JK Paper recorded 35.3 percent and passed at both levels. Reaching the aspirational district is not reaching the tail; the tail is inside the district, and the record has to follow the money that far to see whether it arrived.
What would overturn a company's verdict is stated in advance. Any one of the following would: a later filing, from 2020-21 onward under the revised forms, that records the location of its money more fully; money recorded at state level or as 'PAN India' that in fact reached aspirational or tribal-majority districts; an error in the matching of its districts to Census 2011; or a project whose beneficiaries live outside the district in which it is recorded. Each company can file a correction, and a corrected record would be rerun and published.
The aspirational districts received 5.0 percent of all district-recorded money in this file. That is above the roughly 2.5 percent Sattva reports for all CSR over 2014 to 2023, a figure built on a different population and denominator, and it is a consistency check rather than a comparison. What it cannot do is reassure: a share computed on the 12.4 percent of the file's money that names a district says nothing about the 87.6 percent that does not.
Parliamentary answers say the current CSR filing does not capture beneficiary-wise details of CSR projects, and the Government does not maintain completion status separately. The record ends at rung one by design, and four rungs follow from that choice.
Through whom. Since 2021, agencies implementing CSR projects must register with the Ministry, and companies report the registration number, yet no public route links the money to the agency that carried it. This is the rung where capture lives. In Telangana, across 212 smallholder households and more than 60 civic, state, and market institutions, state and NGO actors held disproportionate influence over resource distribution, while smallholders were largely unable to influence the policy process; benefits were as vulnerable to capture from powerful institutions as from local elites (Ramprasad 2021, World Development). An intermediary is not a conduit. It is an institution with rules, and those rules can be tested: whether the governed can enter its decisions, whether its eligibility rule is published, whether its grievance route answers.
Who. Companies and their implementing agencies hold beneficiary lists. Aggregate counts by caste, tribe, gender, age, disability, land, and tenure require no personal data and would answer the second question at almost no cost. Nothing asks for them.
Position. Whether a household could shape a project, knew the rule that selected it, and could complain and be heard is the difference between a beneficiary and a recipient. Induced participation is vulnerable to elite capture and is rarely monitored in ways that allow learning (Mansuri and Rao 2013); the case evidence on CSR lists the failure to involve intended beneficiaries among the obstacles to development (Frynas 2005).
Difference. Rule 8(3) requires impact assessment only above a size threshold, for projects of Rs 1 crore (about US$114,749) or more, and only a year after completion; the reports go to the board. Nothing requires a comparison, a tail sample beside the average, or a check that the change lasted. An assessment built this way can certify that a program delivered outputs. It cannot tell a board whether the people at the end of the distribution are better off.
The reach test used a fixed threshold of 5 points, and the run exposed its flaw. Twenty of the 29 companies that passed recorded no money in any aspirational district; they passed because aspirational districts make up a small share of the population in the states where they work, so no 5-point gap could arise. Under a threshold that scales with the population share, 41 of the 55 fail. That second figure was computed after the run and is reported as a sensitivity, not a finding. The Tail-End Standard v0.2 replaces the fixed threshold with one that scales: a gap fails where it exceeds the smaller of 5 points and half the reference share. A test that cannot fail a company with nothing in the tail is not a test of the tail, and saying so is part of the method.
None of what follows requires new money. Each step asks the record to climb one more rung.
The Tail-End Assessment is built to do these five things on a single project and to state, in one rating statement, which rungs it reached. It treats the statutory impact assessment as the floor, not the ceiling.
The questions this paper asks of CSR do not require anyone's personal data, and the method is designed so that it never needs it. That restraint is deliberate and rests on both law and ethics. What follows sets out the rules that govern each type of information the ladder touches, how this paper and the Tail-End Assessment comply with them, and where the law is still settling. It is a statement of method, not legal advice, and it reflects the law as of October 2026; when a provision is described from published summaries rather than the Gazette text, that is stated.
Company filings. Documents that companies file with the Registrar under the Companies Act, 2013, including CSR disclosures in Form AOC-4 and the annual CSR report in the Board's Report, are open to inspection by any person on payment of the prescribed fee under Section 399. The Companies (Corporate Social Responsibility Policy) Rules, 2014, as amended in 2021, require the CSR policy and the projects approved by the board to be disclosed on the company's website (Rule 9), and require impact assessment reports under Rule 8(3) to be placed before the board and annexed to the annual report on CSR (described here from published summaries of the amended Rules). Every company-level figure in this paper comes from these public filings, through the Ministry's own portal and its mirror.
Company confidentiality. Contracts between companies and implementing agencies may carry confidentiality clauses, and project budgets, internal evaluations, and partner assessments may be commercially sensitive. The Right to Information Act, 2005 reaches public authorities as defined in Section 2(h), which include government companies but generally not private ones, and even for public authorities it exempts commercial confidence and trade secrets whose disclosure would harm a third party's competitive position (Section 8(1)(d)), subject to a larger public interest. The Tail-End Assessment uses only public filings or information a client authorizes in writing, and it never seeks a company's confidential material through a public-records route on behalf of another party.
Personal data of beneficiaries. Beneficiary lists held by companies and implementing agencies are personal data. The Digital Personal Data Protection Act, 2023 governs their processing, and the Digital Personal Data Protection Rules, 2025, notified in November 2025, bring most obligations into force by May 2027 (reported commencement dates; to be checked against the notification). The Act exempts processing for research, archiving, or statistical purposes under Section 17(2)(b) only where the data are not used to take any decision specific to an individual and the processing meets the standards the Rules prescribe in their Second Schedule; the exemption is conditional and is read narrowly. Information in a public authority's hands that is personal and unrelated to any public activity is also exempt from RTI disclosure under Section 8(1)(j).
Aadhaar. Many beneficiary lists are keyed to Aadhaar numbers. Section 29 of the Aadhaar (Targeted Delivery of Financial and Other Subsidies, Benefits and Services) Act, 2016 bars sharing core biometric information for any reason and limits sharing and use of other identity information to what the Act and its regulations allow and to the purposes disclosed to the individual. The Assessment never requests, stores, or publishes an Aadhaar number.
What the Tail-End Assessment does. It asks for the who rung in aggregate only: counts and shares by caste, tribe, gender, age, disability, land, and tenure, never a list of names. Where fieldwork samples households, it follows the Tailenders Privacy and Confidentiality Policy and the Tail-End Protocol: informed consent in the respondent's language, the right to decline any question, no payment or benefit conditional on answers, names and identifiers stored apart from responses, identifiable field data destroyed within six months of the final report unless a contract requires longer for audit, and no household-level data ever supplied to the company, the agency, or any administering body the respondent may need to contest. Findings travel toward the governed, never against them.
What this paper publishes. Company names and figures as they appear in public filings; district aggregates; no data about any individual. The companion map shows the same public figures. A company that believes a figure is wrong may file a correction, which is logged and answered.
The data run from 2014 to 2020, cover only the largest spenders, and come from a mirror of the Ministry's portal; filings for 2020-21 onward, under the revised forms, may record location more fully, and the Ministry has been asked for them. Six post-2011 aspirational districts are mapped to their 2011 parents, so their population shares are approximate. Census 2011 populations are dated. The reach test measures where money landed, not what companies intended, and the local-area proviso explains some concentration near plants. The four higher rungs are reported as unobservable because the record does not reach them, not because the paper has measured them; measuring them is fieldwork, and it is the work the Assessment exists to do.