AI in Waste Management: Building Independent Verification into Delhi’s WasteChain
If you are interested in applying to GGI's Impact Fellowship program, you can access our application link here.
1. Executive Summary
India’s waste management system runs on self-reporting. Producers, recyclers, processing plants and municipal bodies declare what they collected, processed and disposed of. Almost nothing checks these declarations against what actually happened.
This gap between what is reported and what actually happens is not a small data problem. It is the central weakness in India’s waste and circular-economy policy. It is also the reason a decade of investment has not been able to solve the waste management problem in cities such as Delhi.
Delhi produces roughly 11,862 tonnes of waste every day. Of this, a CPCB (Central Pollution Control Board) affidavit to the National Green Tribunal put processing at only 64 per cent with the rest going to the Bhalswa and Ghazipur landfills, while other official figures report a higher share. Installed processing capacity, built up mainly through waste-to-energy incineration, stands at 7,641 tonnes a day.
Independent inspection found a gap between what these waste-to-energy plants report and what they actually emit, in 2020 and again in 2025-26, even after CPCB moved the plants to a lighter-touch, self-certified category in 2025. The Solid Waste Management Rules, 2026 now require all reporting to go through one national portal. But the rules do not say how these figures will actually be checked.
This paper traces where the gap comes from, draws on how India and other countries have addressed it, and proposes an AI-based verification system that checks every reported figure against an independent measurement each month and sends every mismatch to a named officer.
2. How the numbers are made
Before turning to how these figures are reported, it helps to see how the waste itself actually moves, from where it starts to where it ends up.
Every instrument that does exist in this chain is held by the party whose performance it measures. The weighbridge, the only reliable measurement in the whole journey, sits at the very end and is operated by the party being paid.
A weighbridge is a large industrial scale set into the ground at the entrance of a facility. A loaded truck drives onto the platform, is weighed, tips its load inside, then drives back over on the way out and is weighed again. The difference between the two readings is the tonnage delivered.
There is no single system for reporting waste figures in Delhi. Four different channels operate side by side.
3. Why the gap persists
Delhi's reporting gap doesn't persist because the city lacks data. It has weighbridges, GPS-tracked vehicles, continuous emission monitors, and courts that have repeatedly ordered figures to be produced, and it has been generating readings from all of these for years. The gap survives regardless, because six separate failures sit underneath it at the same time, each answering a different question about how the system was built, and each with its own evidence trail in Delhi's own records.
• Measurement design: how is the data actually measured, and why does that design fail?
Delhi's installed waste-to-energy capacity is 6,550 tonnes a day, spread across four plants - Okhla, Ghazipur, Bawana, and Tehkhand; on top of 558 tonnes a day from 257 decentralised compost pits; and 293 tonnes a day of material recovery facilities, adding up to the 7,641 tonnes a day of total installed capacity1. Every one of these plants is required to monitor its own emissions continuously and report the readings to CPCB and DPCC, but the plant buys, installs, and controls that monitor itself; nobody else's instrument is in the loop.
This isn't an oversight specific to Delhi. It's how environmental self-monitoring was designed
nationally: a lighter-touch model that asks the regulated party to report on itself rather than
requiring a third party to verify it, because independent verification was assumed too costly to
mandate everywhere.
In September-October 2020, the National Green Tribunal ordered a joint CPCB-DPCC inspection of Delhi's three operating waste-to-energy plants. At each one, the plant's own monitor and CPCB's physical inspection disagreed, and the plant's numbers were consistently more favourable. Okhla's monitor looked close to compliant while CPCB found dioxin and furan levels 890 per cent over the standard limit; Ghazipur came in 170 per cent over; Narela-Bawana was 390 per cent over, while also running above its permitted waste intake. Each plant drew a ₹5 lakh fine in 2021, with no order to fix the monitoring itself.
The same pattern resurfaced six years later. In January 2025, CPCB reclassified these plants into its lighter-touch "Blue" category, relying more on self-certification and less on independent inspection. A September 2026 review tested that classification using CPCB's own scoring method and found Delhi's four plants actually score 100 out of 100 on the Pollution Index, Red-tier territory, not Blue. At Ghazipur, the plant's monitor still didn't match CPCB's own readings: dioxin and furan levels 270 per cent over limit. Six years on, the same problem persists: the readings still do not match.
• Incentive design: why would anyone report accurately?
Collection contracts pay for tonnage collected, not for how well the waste is sorted, because weight is easy to measure and sorting quality isn't. Therefore, contractors get paid the same whether they sort properly or not, and there's no financial reason to do it right.
Further along, EPR (Extended Producer Responsibility) is the rule requiring companies to pay for recycling the packaging and products they sell, and the certificates that prove recycling happened can be sold in the open market. So, if someone inflates how much they recycled, that inflated number turns straight into money. And where there's any enforcement, it's cheap to shrug off: Okhla was fined ₹5 lakh per plant and additional ₹25 lakh by the tribunal, but neither fine came with an order to fix the actual monitoring. So, paying up was simply cheaper than fixing anything.
• Institutional design: who is actually responsible for Delhi's data, and why does the
fragmentation survive anyway?
On paper, the chain is simple: facility operators report to their local body, local bodies report to DPCC and the state Urban Development department, and DPCC consolidates a single annual report and submits it to CPCB on a standard form, Form-V, by law, every July6. So DPCC is the body legally responsible for producing Delhi's consolidated number; CPCB is a receiver of state- level submissions, not a primary collector.
The problem sits upstream of DPCC. Delhi's waste-handling bodies are themselves split three ways by territory, not function: the Municipal Corporation of Delhi covers most of the city, the New Delhi Municipal Council runs a separate system for the Lutyens' Delhi government enclave answerable to the Union Home Ministry, and the Delhi Cantonment Board does the same for the military cantonment area, answerable to the Ministry of Defence.
A 2022 investigation into exactly this fragmentation found that a formal coordinating role does exist on paper, the state Urban Development Secretary is designated "Director, Local Bodies" with the power to align these agencies, but functions in practice as "a rubber stamp". And CPCB's own account of the data once it reaches this point names three specific defects: it is often collected "not from the point of origin of waste but from secondary locations such as treatment facilities, landfills, dumpsites," filed with "partial and incomplete information at various levels," and undermined by "unavailability of a system of weighing" at the operator level6
DPCC is answerable for consolidating the number. Nobody is answerable for whether the number
that reaches that authority was ever checked against anything independent before it got there.
• Capacity design: is anyone actually staffed to do this work, and why not?
Even a well-designed mandate fails without people to carry it out. DPCC has run at under 60 per cent of its sanctioned strength for years: 153 of 344 posts stood vacant as of November 2024, an RTI query showed 192 vacant the year before, and the agency was below half-strength as far back as 2020.
This isn't unique to Delhi; CPCB's own reporting shows roughly half of India's state pollution control boards struggling with comparable vacancy rates, which suggests the cause is structural. Regulatory staffing has never kept pace with what these departments are asked to enforce, rather than a one-off local failure. Reconciliation work is staffing-intensive by nature, and the institution meant to perform it has never had the people with the right skills.
• Systems design: does any technology connect the journey end to end, and why doesn't
one exist yet?
The Solid Waste Management Rules 2016 that governed Delhi's waste reporting for a decade
required local reporting but never mandated a shared digital architecture to hold it, so each
authority built, or didn't build, its own separate record-keeping. Before the 2026 Rules, India's waste reporting ran on what one review called "fragmented, non-standardised reporting," not a connected system.
The new centralised CPCB portal is the first real attempt to fix this design gap, and it is still
untested. Until it proves itself, each stage of the waste journey, household, collection, transfer, transport, and processing, keeps its own separate record, if it keeps one at all, with nothing built to stitch them into a single traceable loop.
• Accounting design: what actually gets counted, and why is a whole category left out?
The informal sector, waste pickers and small aggregators, recover an estimated 15-20 per cent of Delhi's waste at several points along the chain, and none of it appears in any official return. This is a design gap, not a data gap: the formal reporting system was built around contracted, registered quantities, and waste pickers were never brought into that registration structure, so there was never a field in any form for their tonnage to go into. As a result, every official waste count that the city produces leaves out an amount nobody has measured, by design, not by accident.
4. Best practices in Independent Verification
Every stage of this chain could, in principle, swap a declared figure for a measured one. The table below shows four examples, and how far each goes.
Country | What the Country does | What Delhi can take from it |
South Korea | One electronic record follows industrial waste from the factory, through the transporter, to the disposal site, with RFID tags, QR codes and GPS logging each handover - for Industrial waste only. | A single identifier carried across every handover — the role the Waste Reporting ID would play in Delhi. |
South Korea — RFID food waste bins | Household food waste is weighed by the bin at the moment it is thrown away, and the household is charged on that measured weight. | The figure comes from a meter rather than a form. This is the principle behind instrumenting Delhi's collection points. |
Japan — JWNET electronic manifest | The generator, the transporter and the disposer each enter their own record of the same consignment into one shared system, which flags reports that are overdue. Industrial waste only; household waste is not covered. | More than one party reporting on the same consignment, so no single account stands unchecked. |
European Union — Area Monitoring System | Every area-based farm subsidy claim is checked against satellite imagery, and the claimant must be told of a finding before any penalty is applied. | Independent observation through use of satellite imagery can be used to test a declaration, with a right of reply — directly applicable to Delhi's three landfill sites. |
Sources: Japan Industrial Waste Information Center, Electronic Manifest System for Industrial
Waste in Japan (July 2023) — 308,160 subscribers and 38.5 million e-manifests a year, household waste not subject to the manifest system; Korea Environment Corporation and UN Office for Sustainable Development, Solid Waste Data Generation for Sustainable Development, Solid Waste Data Generation through Digital Transformation: country case of Korea12; Commission Implementing Regulation (EU) 2022/1173, Article 10.
Indore, a city of about 25 lakh across 85 wards generating roughly 1,115 tonnes of waste a day,
reached 100 per cent door-to-door collection across every ward by the end of 2016 and now sends only 5-6% of its waste (the inert residue) to sanitary landfill. The city was able to achieve this without a single waste-to-energy plant; it relied on source segregation, composting, material recovery and a 550 tonne-per-day bio-CNG facility, the largest in India.
What matters here is how Indore tracks its performance. Indore runs a command-and-control centre that tracks every collection vehicle by GPS, digitises routes to confirm coverage is
complete, optimises them against historical data, evaluates driver performance and pulls in data
from third-party systems. That gives the city - a daily view of which wards and which vehicles are falling short, and the ability to act on it while it still matters, rather than discovering it in an annual return.
Indore's advantage is not better technology or bigger plants, it is that somebody sees the data
every day.
5. The solution: Satyapan - an AI verification system for Delhi's waste management
SATYAPAN (Hindi for ‘verification’) is a single system, an artificial-intelligence model and the portal it runs on, that gives Delhi, for the first time, an independent check on every figure its waste chain produces. It issues one identifier to every unit in the chain, collects one standard return, draws in measurements from the instruments and agencies that the reporting party does not control, compares the two every month, and routes every discrepancy to a named officer with a deadline. It is designed to be built into the national portal that the Solid Waste Management Rules, 2026 now require.
Verification in Delhi's waste chain rests on five things, and they work only together. Every unit needs one identifier, used on every form at every level, so records about the same plant or vehicle can be matched automatically. Returns need to come in monthly, so a comparison can be made at the same pace the waste moves. Each figure needs a second measurement from an instrument or agency the operator does not control. The waste pickers who recover a fifth of the city's material need to be counted, so the city's totals add up. And every mismatch needs a named officer with a deadline to act on it.
The sections below set out what the AI model does (5.1), the filing architecture it runs on (5.2), the instrumentation it requires along the waste journey (5.3), how the informal workforce is brought inside it (5.4), and the rules that govern the model’s operating system (5.5).
5.1 What the system does
Six modules. Four do the analysis. Two make the results usable. No human team could do this work at this speed or scale. The AI models read, retrieve, and rank. A separate, rule-based system does all the actual arithmetic. So, if a model misreads something, it shows up as a visible, fixable error, not a final decision.
Module | What it does |
1. Read | Reads the actual paperwork each month, CPCB portal filings, scanned documents, court affidavits, and plant operating permits, and turns it all into one consistent set of numbers, even when the format changes or a new contractor takes over. |
2. Reconcile | For each site, checks the number reported against an independent source, like a weighbridge or an electricity meter, and flags any gap that's too big to be normal. Also checks that a plant's reported tonnage, the heat value of that waste, and the electricity it actually produced all line up, so an inflated tonnage number gets caught. |
3. Observe | Uses satellite and drone images to measure how much waste is actually sitting at a landfill, and to catch dumping that was never reported. No one has to file anything for this check to happen. |
4. Rank | Puts every flagged mismatch in order, by how large the gap is, how often that site has had problems before, and its compliance history, so inspectors go first to the sites that matter most. |
5. Assist | This lets an officer ask plain questions, like what did this site report, what does the independent record show, which rule does this gap break, and the system pulls up the evidence and drafts the notice. Operators can also check their own filing and fix a genuine mistake before any penalty starts. |
6. Record | Keeps one running record for each site, so MCD, DPCC, CPCB and the Ministry are all working from the same number, and publishes a monthly list of what matches and what doesn't. |
5.2 The filing architecture it runs on
The closest working model in Indian administration is the income-tax system, which standardises the whole filing process rather than only the final check: one identifier per filer, one standard return, one statement of third-party data the filer never types, and a comparison of the two before anything is settled. Three of the four pieces exist in Delhi's waste chain today under different names, on different schedules, at different offices. Standardising them, and adding the one that is missing, is the administrative core of the system.
Element | What it does |
One ID per unit (W-1) | Every plant, landfill, transfer station, vehicle fleet, recycler and registered waste picker gets one ID, used on every form, at MCD, DPCC, CPCB and the ministry. |
One standard return (W-2) | Each unit files one return a month: tonnage received, processed and sorted by category, and energy output. Same fields, no matter which authority is looking. |
One number the filer can't touch (W-3) | Weighbridge readings, metered electricity export, inspection results, sample checks and landfill surveys go in directly from the instrument or agency that owns them to the Ledger portal. |
An automatic check (W-4) | Every month, the two numbers are compared. Anything that doesn't match gets sent to one named officer, with a deadline. The unit is told and gets a chance to explain or fix it before any penalty. |
5. Assist | This lets an officer ask plain questions, like what did this site report, what does the independent record show, which rule does this gap break, and the system pulls up the evidence and drafts the notice. Operators can also check their own filing and fix a genuine mistake before any penalty starts. |
One shared record | A single running record all four authorities can see, and a public page showing whether the figures matched that month. |
5.3 Instrumentation along the waste journey
A reconciliation engine is only as good as the independent measurements available to it, and those measurements have to be created at each point where waste changes hands. The table below shows what makes that possible at each stage, from the household bin to the certificate issued for a recycled tonne.
Stage | What is installed | What it lets you check |
Household and colony | Weighing or RFID bins must be installed at colony level and then at district level also. | Whether waste is actually being segregated and in what quantities and frequencies. |
Collection | GPS and load sensors on every vehicle that is collecting the waste from these RFID bins placed at colony / district level. | Whether routes were completed and how much was actually lifted — and, over time, which colonies generate the most, so collection effort can be directed where the volume is. |
Informal recovery | Handheld logging at registered aggregation points, against the worker's own ID. Informal waste workers are formalised into the system as employees. | Recovered tonnage that appears nowhere today, which is what closes the city's mass balance. |
Transfer station | Imaging and periodic sampling of incoming loads is extremely crucial. | The real composition of the waste, set against what the ward reported. |
Transport | An electronic record that links the vehicle, where the waste was picked up, and where it was dropped off, so you can confirm the waste collected is the same waste that actually arrived at the facility. | Whether a load reached the site it was sent to. |
Processing | Weighbridge at the gate, metered electricity export and stack monitors, each recording automatically and generating returns that must be filed. | Whether the tonnage claimed fits the energy produced and the emissions released. |
Landfill | Drone and satellite surveys on a fixed schedule. | Whether the site is shrinking as fast as the records say. |
Recycler and EPR | Chain-of-custody record tied to each certificate. | Whether a certificate has a real consignment behind it. |
5.4 Bringing the informal workforce inside the system
An estimated 15-20% of Delhi's waste is recovered by waste pickers, saving the municipality at least ₹6 lakh a day, yet none of it is captured in any return, portal, or record. This is not only a fairness gap but an accounting one: if a fifth of the material is left out of the count, the numbers can never be reconciled, and every comparison run by the system is distorted by that absence.
The same identifier system used elsewhere is applied here. A Waste Reporting ID is issued to each registered waste picker. When material is sold at a registered aggregation point, the sale is logged against that ID by the aggregator, not by the worker or the municipality, and an independent record is created in the process. The recovered tonnage is then credited back to the ward it came from.
That identifier is paired with formal employment. Sorting and recovery is essential work for the city, yet it is currently carried out without a contract, without protection, and without any official record of what is handled.
Formal municipal posts are created for this work, each carrying a job ID, fixed pay, protective gear, insurance, and a route into the collection contracts the city already awards. In this way, the people already performing this work, most without formal qualification, are brought into the system as employees rather than left outside it.
5.5 Governance and scope
Three rules would keep the system fair and enforceable. Behind all three, every confirmed mismatch goes to a named officer at DPCC with a fixed period to settle it, so that a finding leads to an action.
• A published tolerance band - A gap must cross an agreed threshold before it is flagged, so
rounding and ordinary measurement error do not fill the queue.
• Notice before consequence - A unit sees the discrepancy and has a set window/ timeline to correct or explain it before any penalty begins, as the European Union requires for its own satellite checks also.
• A physical audit behind the automated one - Units with repeated or large mismatches go
onto an unannounced inspection list. A check that always runs the same comparison can be
learned; the surprise visit is what sits behind it and would be required for such units.
6. Conclusion
The real value of independent verification lies beyond compliance, in what it allows a city to change. Delhi's landfills continue to grow, and what accumulates there burns, leaches and releases methane into a city that already struggles to breathe. Waste Management sits at the junction of two of Delhi's hardest problems, air quality and climate, and what is still missing is an independent measure of how much is generated and where it goes. Every target the city sets - to divert more waste, to reduce what is landfilled, to close a site - is easier to pursue, and to demonstrate, against a baseline that has been independently established. That is what verification offers: it allows a reduction to be shown rather
than asserted.
This is where Artificial Intelligence earns its place in the waste management. The AI model working as a reconciliation engine would show patterns that would tell Delhi how to reduce waste that already exists in the city's own records: which settlements generate what, how composition shifts by season and density, where segregation holds, which routes and facilities perform. What is required is a system that reads them together, so the learning they hold can be put to use. Measured consistently, those records become evidence, and evidence makes reduction a policy that can be aimed rather than estimated.
A model of this scope has not yet been applied across every stage of a city's household waste, in such consolidation, as per the available evidence. Certain levers of the model have been implemented in certain cities, but a model that combines each lever so that the data arrived and the patterns generated from them could be useful for future readiness is the need of the hour. Therefore, it is recommended to test the suggested AI model as a pilot and build the case as it runs.
India is committing substantial capital to waste infrastructure under a framework that has only just become digital, and the returns on both will rest on performance that can be independently established.
Meet The Thought Leader
Laboni is a mentor at GGI and is currently working at The Bridgespan Group as a Senior Associate Consultant. She takes interest in socioeconomic development issues, public policy, and equity across different vectors of gender, caste, class, and ability, which in turn fuelled her transition from working at a global bank to the social sector. She is an Urban Fellow from the Indian Institute for Human Settlements, Bangalore and has a bachelor's degree in Economics from St. Stephen's College, University of Delhi.
Meet The Authors (GGI Fellows)
Amarpreet Kaur Chawla is a Policy Manager at PwC India with over 9 years of experience in policy advocacy, government relations, stakeholder engagement, and social impact. Having worked with organisations like the Confederation of Indian Industry (CII), Indian Direct Selling Association (IDSA), and Samarthan, she brings a strong understanding of public policy, institutional partnerships, and industry-government collaboration. With a focus on shaping policy discourse and driving systemic change, she is passionate about governance, entrepreneurship, and building impactful public-private partnerships.
If you are interested in applying to GGI's Impact Fellowship program, you can access our application link here.
References
1. Delhi Economic Survey 2025-26, Planning Department, Government of NCT of Delhi.
2. The Hans India - https://www.thehansindia.com/news/national/delhi-processes-only-64-pc-of-daily-waste-rest-dumped-at-bhalswa-ghazipur-landfills-cpcb-tells-ngt-1035540
3. Centre for Financial Accountability, December 2022. https://www.cenfa.org/the-national-capitals-experience-with-waste-to-energy/
4. Centre for Financial Accountability, September 2026 https://www.cenfa.org/breathing-poison-the-100-100-pollution-score-the-cpcb-doesnt-want-you-to-see/
5. Press Information Bureau.
6. Central Pollution Control Board / Ministry of Statistics and Programme Implementation, July 2023.
7. The Print, April 2022. https://theprint.in/opinion/8-agencies-are-to-blame-for-delhis-waste-mountains-4-ideas-can-demolish-them/926089/
8. 9. The Patriot, November 2024. https://thepatriot.in/reports/delhi-fight-against-pollution-hit-by-staff-shortage-61403
Down To Earth, 16 March 2026
Comments