Cutting Theft and Shrinkage in Indian Factories with Cameras

Shrinkage — the stock that goes missing between the receiving gate and the dispatch dock — eats an estimated 1–2% of material throughput at a typical Indian plant, roughly ₹50–100 lakh a year on ₹50 crore of stock. Camera analytics targets the theft slice, about half to two-thirds of that loss, by cross-checking what the camera physically saw at the gate and dock against your own dispatch paperwork.

Published 16 July 2026.

Most owners treat shrinkage as a cost of doing business, a number the auditor reconciles once a year. That is exactly why it stays expensive. This page puts India-specific figures around the problem and, using a documented BSE-disclosed factory theft, shows why the real value of analytics isn't spotting a 2 a.m. loiterer. It's being the one witness your own staff cannot bribe.

How big is factory shrinkage in India?

Plan for 1–2% of material value lost to shrinkage at a mid-size Indian plant handling loose, portable or high-value stock: copper offcuts, yarn, fasteners, pharma packaging, FMCG cartons. Hard manufacturing-specific data is thin, so that band is a reasoned synthesis of measured retail rates plus what factory theft cases actually show — not a surveyed figure. Measure your own reconciliation gap to replace it.

The retail anchor: the Global Retail Theft Barometer put India at the highest shrinkage rate in the world, 2.38% of sales (July 2010–June 2011 data), ahead of Russia at 1.74% and Morocco at 1.72% (via IFSEC Insider). The same series reported an India cause split of roughly 45% external, 23% employee, and 23% admin/process error (those sum to ~91%, the rest supplier fraud); the study was later discontinued, so treat it as directional. More recently, Trent disclosed FY24 shrinkage rising to 0.41% of sales, up from 0.22% (Business Standard, June 2024).

Key figures

Why factory theft is an inside job: the Goodyear case

Vendor pages assert that internal theft dominates on a factory floor and leave it there. Here is the proof they don't use. In November 2024, Goodyear India found a physical shortage of 4,571 tyres, about ₹3.91 crore, at its Ballabgarh plant. An Ernst & Young investigation, disclosed to the BSE in July 2025, traced the bulk of it to third-party warehouse workers, loading manpower and security guards acting in collusion: truckloads of tyres left the plant on unidentified trucks with no record at the warehouse, the material gate, or the weighbridge. The suspicious-message trail alone pointed to 4,057 tyres across 20 dates between July and December 2024 (Goodyear India material-event disclosure; Autocar Professional).

Read that pattern closely, because it defines what a factory control actually has to beat:

• The theft was slow and repeated — 20 dates over six months, not one smash-and-grab.
• It was covered by paperwork — the goods moved in daylight, on-shift; the fraud was the absence of a record.
• The guard was not the control — the guard was compromised. One security person admitted taking kickbacks to keep the trucks out of the register.

That is the theft that hurts an Indian plant, and it is nothing like a retail shoplifter. Which means most of the "security analytics" sold on this keyword is aimed at the wrong target.

Employee vs external vs paper: what cameras actually catch

Split the loss by cause before sizing what analytics can claw back, because cameras only touch part of it. The shares below are our synthesis of the GRTB split adjusted for factory context — on a shop floor the internal share runs higher than retail's 23%, because staff know the blind spots and the reconciliation lag, and Goodyear is exactly that pattern.

Cameras realistically address the theft slice, call it half to two-thirds of total shrinkage. Anyone promising to erase all shrinkage with lenses is selling you the paper-error slice they can't touch.

The detection that matters: reconciliation, not loitering

Every vendor page publishes the same checklist — intrusion, loitering, tailgating, perimeter breach. Those retail-style detections are the weakest possible controls against the Goodyear pattern: colluding insiders moving goods in daylight, on-shift, covered by falsified paperwork. A loitering alert never fires on a truck that is supposed to be at the dock.

The high-value factory use case is reconciliation cross-check: match what the camera saw at the gate, dock and weighbridge against the goods-receipt note, weighbridge slip and dispatch record, then flag the truck that left the frame carrying goods but left no record. That single check is what would have surfaced Goodyear's 20 undocumented dispatches months before the annual count. Indian vendors — Agrex AI, viAct, Katomaran — run detection on your existing RTSP feeds and advertise figures like a 62% cost reduction for warehouse deployments (Agrex AI); those are vendor-reported, not independently audited, so demand a measured before/after in your own pilot.

Real-time matters here for a different reason than in retail. A shoplifter is a one-shot event where alert latency is everything. Insider factory theft is slow, repeated and paperwork-covered, so the win is not a faster alarm. It is closing the reconciliation-lag window, the months of invisibility between audits that the colluders depend on.

The contrarian ROI: the control your staff can't switch off

Every ROI-focused vendor page runs the same labour-arbitrage pitch: analytics lets you deploy fewer guards. That argument is weak, and the Goodyear file shows why — there, the guard was the control, and the control was bought for a kickback.

Sell the sharper truth instead. At an Indian plant a guard is not just a cost line, he is a corruptible node, and manned registers are the first thing an insider ring compromises. Camera analytics is the tamper-evident, un-bribable witness: an event trail that cannot quietly be not logged. That reframes ROI from "save on guards" (contested) to "the one control your own staff cannot switch off" (unique). The money still works either way:

Recoverable shrinkage — a plant moving ₹50 crore of material a year:

The one-third reduction is a placeholder, not a measured deterrence effect; swap in your own pilot before/after. Even so, a fraction of a percent of throughput dwarfs what the software costs.

The guard-cost reference point. Manned guarding is the default anti-theft control, and it is not cheap once statutory costs are counted. A compliant unskilled guard runs ₹24,000–25,000/month in Delhi including PF, ESI and statutory bonus (Knighthood, 2025). A single round-the-clock post needs about 3.5–4.5 guards: 168 hours a week over ~48 hours per guard is 3.5, plus weekly-off and leave relief under the Occupational Safety, Health and Working Conditions Code, 2020. That puts one 24/7 post near ₹1.0–1.15 lakh a month (the top end includes the agency's 6–10% service fee), roughly ₹12–14 lakh a year for a single gate. Analytics doesn't fire guards; it makes them and their registers auditable. Deferring one added post funds a multi-camera deployment on its own.

Point cameras at the exits and the evidence

Theft leaves through a handful of choke points: loading/unloading docks (goods and vehicles meet here — the highest-risk zone), finished-goods and high-value stores, the scrap yard and weighbridge (a classic under-weighing and swap-out collusion point), and gates and perimeter.

"Runs on your existing cameras" is true and every vendor says it, but it hides the catch: most factory cameras are aimed wrong for theft evidence. A dock theft needs the goods, the person and the vehicle in one frame to stand up against a dispatch record; a gate needs a readable plate, not a wide yard shot. Existing cameras only help if they point at the exits and the evidence. That aiming gap is what Mama closes: record a short phone walkthrough of the floor, docks and yard, and it returns a floor plan plus a camera-placement plan — which sightlines each camera needs to survive a reconciliation cross-check — without waiting on a site survey.

Do it legally: worker video is personal data

Anti-theft monitoring points cameras at your own staff, and Indian law is now specific about it. This section is the compliance checklist most vendor pages skip:

• DPDP Act, 2023. Video of workers is personal data. You need a defined purpose (loss prevention/safety), notice and signage, a retention limit and restricted access (official text, MeitY). Employee "consent" is weak footing given the employment power imbalance, so lean on a clearly stated legitimate purpose plus proportionality rather than a signed form.
• No audio without consent. Recording sound alongside video is a separate exposure; keep analytics cameras video-only unless you have explicit consent.
• No cameras in private zones. Restrooms, changing rooms and locker areas are off-limits.
• Covert cameras are a crime. Secretly recording, especially of women, is punishable as voyeurism under Bharatiya Nyaya Sanhita Section 77 (which replaced IPC Section 354C from 1 July 2024) — signage isn't only courtesy, it's protection.
• BIS/STQC hardware mandate. Any new cameras bought in India from 1 April 2026 must meet the BIS/STQC Essential Requirement ER:01 for CCTV (Gazette 9 April 2024; ER:01 effective 9 April 2025; no sale of non-compliant units from 1 April 2026) (BIS CRS guidelines). Specify compliant hardware for any expansion.

Do the compliance groundwork before you deploy, not after an incident.

FAQ

What shrinkage rate should an Indian factory assume? Plan for 1–2% of material value; manufacturing-specific data is scarce, so this is derived from measured retail rates (India peaked at 2.38% of sales in the Global Retail Theft Barometer; listed retailers now report 0.4–0.5%) plus documented factory theft cases. Measure your own reconciliation gap for a real number.

Is most factory theft internal or external? On a factory floor the internal share usually runs higher than retail's ~23%, because staff know the blind spots and the reconciliation lag. The 2024 Goodyear Ballabgarh case is the pattern: warehouse workers, loading crew and security guards colluded to move 4,000-plus tyres out with no gate, warehouse or weighbridge record.

Can cameras really recover the money, or just record it? They recover it by closing the reconciliation-lag window: cross-checking what the camera saw against the GRN, weighbridge slip and dispatch record surfaces the systematic gap colluders rely on between annual audits. That addresses the theft slice, roughly half to two-thirds of shrinkage, not administrative error.

Will analytics let me cut security guards? Usually it reallocates rather than eliminates them, but the stronger reason to deploy it is that a camera trail is tamper-evident where a manned register is not — in the Goodyear case a guard was paid to not log trucks. Given a 24/7 post costs ~₹12–14 lakh/year, deferring one added post can fund a multi-camera deployment.

Is it legal to monitor my own workers on camera in India? Yes, for a legitimate purpose like loss prevention, but worker video is personal data under the DPDP Act, 2023 — you need signage/notice, a defined purpose, a retention limit and restricted access. Keep cameras out of restrooms and changing areas, avoid audio without consent (covert recording is punishable as voyeurism under BNS Section 77, formerly IPC 354C), and buy only BIS/STQC ER:01-compliant hardware from 1 April 2026.