In plain English: this is what we collect, what we lose, why we lose it, and the specific actions that recover most of it. Every number traces back to actual data — not estimates.
across 4,899 active retailers · 12 months
Excise + sales revenue, FY 2025-26
2.97% of collection · 4 distinct causes
Conservative estimate from the same dataset
at a 5% capture rate · achievable in the first year
Models + recommendations are ready
For every ₹100 the state collects in excise revenue, about ₹2.97 is bleeding out — across four distinct causes. We don't need to fix all of them perfectly. Capturing just 5% of the leak puts about ₹44 Cr back into the budget every year, with technology that already exists.
Four causes, ranked by size. Each one has a clean, executable fix.
Several retailers show monthly sales patterns that diverge from their peers in the same district — abnormally low velocity, missing categories, or spikes inconsistent with local demand. Some are operational reasons, but a measurable portion isn't.
51% of total leak
Distilleries are allotted production quotas in proof-litres but consistently lift less than allotted. The unlifted volume is unaccounted-for inventory — it should be in the system but isn't.
38% of total leak
Retail outlets within 50 km of the Telangana, Karnataka and Tamil Nadu borders see consistent revenue softening. Cross-state price differentials make these corridors economically attractive for small-volume diversion.
12% of total leak
Documented seizure cases since 2019 — but seizures only count what enforcement caught. The actual counterfeit volume on the shelf is materially higher.
0% of total leak
Ordered by recovery size · each one builds on data we already have
A 4-week feedback loop between allotted vs. lifted will surface gaps before they accumulate into 6% of unaccounted volume.
The compliance model already separates 4,899 retailers into compliant / watch / flag tiers using their own monthly velocity. Inspectors get a ranked list each month, not a thousand-row spreadsheet.
The corridor model spots three behaviours that signal in-transit diversion: long detours, off-route deviations, and unusual dwell times. Real telematics replaces synthetic data once trucks are equipped.
A camera + the existing Label Shield model catches counterfeit holograms in under 100 ms per scan. Enforcement teams currently catch what they happen to inspect; this catches everything passing through pilot outlets.
Recovery estimates assume the stated capture % of the corresponding leak. Combined, these four actions recover roughly ₹44 Cr / year at a conservative 5% blended capture rate — the headline number you saw above.
Equivalent public-spend at standard AP government rates
Unit costs are approximate AP government rates and are illustrative — they trade off precision for relatability. Adjust on the Revenue Gains page if you want different anchors.
Every figure on this page is computed from the live AP Excise dataset: 4,899 retailers, 1.27 million sales rows, 42 distilleries, 1,373 approved labels, 430 recorded seizures. Patterns are surfaced by AI models that you can audit on the Methodology page; raw data is browsable on Datasets. When a model relies on synthesized input (as Lock 2 GPS and Lock 3 holograms currently do for the POC), it's called out explicitly — including on the recommendations above.