The most telling sentence in India's smart metering programme was spoken in the Lok Sabha this April, when the Union power minister clarified that prepaid smart meters are not compulsory and are not being imposed on anyone. With that, the largest metering rollout in the world formally became a matter of consent. And consent is currently being withheld. Of 20.33 crore meters sanctioned under the Revamped Distribution Sector Scheme, 5.73 crore are installed, with 7.24 crore in place nationally across all schemes; the deadline has slipped from March 2026 to March 2028. Procurement explains part of the gap. The rest is households saying no, in protests from Bihar to Goa, and most consequentially in Gujarat, where anger over inflated bills left the state at about 38.79 lakh installations against a sanctioned target of over 1.6 crore.
It is tempting to file this under misinformation. I think that is a misdiagnosis. Look at the bargain from the household's side of the wall. The meter makes measurement exact, and in prepaid mode it makes supply contingent on maintaining a balance. Both of these serve the utility. Neither obviously serves the household, and a suddenly higher bill is exactly what exact measurement feels like to someone who was measured loosely for years. A household that refuses this bargain is not superstitious. It is waiting to be offered something in return.
The frustrating part is that the something already exists. It is sitting, unread, on utility servers.
The meters themselves work. They report consumption, voltage and events at 15-minute intervals or finer, in volumes no army of meter readers could have produced. What has not kept pace is the ability to act on any of it. A discom's SCADA and network systems see the grid well down to the feeder and, increasingly, the transformer. Below that, at the level of the street and the individual connection, the smart meter is the only sensor that exists, and it is the newest, least integrated part of the estate. Ask a question that ought to be trivial, whether a given street is actually without power or its meters have merely stopped reporting, and the answer must be assembled across the three to five separate head end systems a single utility may run, procured from different vendors on different communication technologies. Utilities in far wealthier countries are wrestling with the same problem. Meanwhile the household does what households have always done: waits a while, then make calls to customer care. Senior discom officials, quoted by Down To Earth in January, put it plainly, acknowledging that even advanced utilities currently use less than a quarter of the data their meters generate.
A Smart Meter Operations Centre is the unglamorous fix. The name suggests a command room; the reality is a layer that reads from the systems a utility already owns and reconciles them into one live view of consumers, assets and crews. Its first job is answering the street question reliably, without anyone opening five screens. Its second is acting on the answer. The example I consider most important is not an outage at all: If a prepaid disconnection is about to be applied to a connection running medical equipment or a water pump, the system should stop and ask a human being first. The sharpest fear behind the resistance to prepaid metering is automated disconnection at the worst possible moment. A utility that can demonstrably prevent it is not showing off a feature. It is keeping a promise.
{{/usCountry}}A Smart Meter Operations Centre is the unglamorous fix. The name suggests a command room; the reality is a layer that reads from the systems a utility already owns and reconciles them into one live view of consumers, assets and crews. Its first job is answering the street question reliably, without anyone opening five screens. Its second is acting on the answer. The example I consider most important is not an outage at all: If a prepaid disconnection is about to be applied to a connection running medical equipment or a water pump, the system should stop and ask a human being first. The sharpest fear behind the resistance to prepaid metering is automated disconnection at the worst possible moment. A utility that can demonstrably prevent it is not showing off a feature. It is keeping a promise.
{{/usCountry}}Rules alone will not get there, because rules cover only what someone anticipated. This is what AI-first actually means, and it is an architectural choice rather than a slogan. Today the operator queries and the system reports. In an AI-first centre the inference sits on the data path itself: the system proposes what is happening and what should be done, and the operator supervises and overrides. The silent-meter question shows why this is genuinely a modelling problem and not a threshold. The evidence is scattered across systems, the historical record of past outages was itself built from complaint calls, and the costs are lopsided, since a crew sent to a communications fault is a crew the genuinely dark street waits longer for. The International Energy Agency estimates, globally, that AI-based fault detection can cut outage durations by 30 to 50%. The number is plausible precisely because so much of an outage's duration today is spent working out where it is.
Prevention matters as much as response. When a distribution transformer fails, every household behind it goes dark, often for hours. Figures discussed at a Central Electricity Authority standardisation meeting last October put the national failure rate near 10%, but the spread is the real story: Kerala at 1.9%, some states above 20, the best performers below 0.5. A 20-fold gap between operators of identical equipment is not physics. It is maintenance and information, and the warning signs: sustained overload, unbalanced loading, voltage excursions, are already in the data.
Uttar Pradesh's 1912 helpline logged 95.11 lakh complaints in the year to April 2026, by UPPCL figures reported in Millennium Post, including ₹8.87 lakh bill revision requests and ₹2.32 lakh about smart or prepaid meters themselves. UPPCL resolved 98.88% of them, which is a genuine achievement and an expensive one. Every resolution is a call, a ticket, often a vehicle, paid for by a sector that returned its first collective profit only last year, ₹2,701 crore in FY25, against accumulated losses of ₹6.39 lakh crore. And a resolution rate measures whether a ticket closed, not whether the problem returns next month. The cheapest complaint is the one that never happens: the anomalous reading investigated mid-cycle instead of disputed after the bill, the theft flag checked against transformer-level data before an inspector knocks on an innocent door, the chronic voltage problem fixed before it quietly kills the fridge, which today never becomes a complaint at all and, therefore, never gets fixed.
None of this works on the data as it stands. In one detailed CEEW feeder study, only about 60% of consumers were correctly mapped to their transformers, so the first task of any intelligent operations layer is usually helping the utility repair its own model of its own network. And because 15-minute data reveals when a house is empty and when it earns, utilities holding it are likely to qualify as data fiduciaries under the data protection rules phasing in through 2027. These systems belong inside the utility's own environment, which is the position the ministry of power's India Energy Stack already takes.
The scheme and its bidding documents govern acquisition thoroughly; metering providers carry contractual obligations on read success and data availability. Nothing governs what happens after the data arrives. A utility can receive every reading it contracted for and still issue estimated bills, still learn of outages from its call centre, and breach nothing, because none of it is measured. The ministry has set the direction, with its conference on AI in distribution, its stated intent to mandate meter interoperability, and the Energy Stack itself. The funding incentive has not caught up, and with roughly three-quarters of sanctioned meters still to be installed, most of the money is still yet to be spent. The single sharpest metric to reward is also the simplest to state: what share of outages did the utility detect before the first consumer called? No vendor can deliver it alone: it is hard to game, and it measures, almost directly, whether the meter is working for the person it sits on.
That, finally, is the answer owed to the household. Power restored before anyone has to make a complaint. A bill that is right the first time it’s sent. No inspector on a blunt flag. And where state commissions bring in time-of-day pricing, a meter that lets a household lower its bill rather than merely be billed more precisely. The meter is not the reform. What we do with what the meter sees is the reform: and three-quarters of the programme is still to come. There is time to make the box on the wall worth saying yes to.
(The views expressed are personal)
This article is authored by Udit Poddar, co-founder & CEO, WorkOnGrid.