412,000 first-generation AMI meters, fourteen years in the weather, and thousands of them are lying. Watch a full fleet forensics sweep: 6,100 silently under-registering meters found from interval shapes alone ($3.1M/yr walking out the door), 11,300 mesh orphans burning truck rolls, and the meters' own voltage data catching three failing transformers nobody had on a list. Then the question every metering VP faces: where does AMI 2.0 go FIRST? Not oldest-first, value-first: every neighborhood scored on failure rate, outage exposure, DER density, and revenue risk, sequenced into three waves where Wave 1 captures 54% of the benefit for 22% of the spend.
| Without | With GridCORTEX | Δ |
|---|
| Without | With GridCORTEX | Δ |
|---|
A fictional utility has 412,000 first-generation smart meters that have spent fourteen years in the weather, spread across eight named areas (NORTHGATE 58,000 meters, CEDAR PARK 47,000, LAKELINE 41,000, MIDTOWN 64,000, RIVERSIDE 55,000, EASTPORT 38,000, SOUTH YARD 49,000, BAYSHORE 60,000). Each meter records usage every 15 minutes, plus the voltage at the house, which adds up to 14 billion readings a year. The problem: aging meters fail quietly. Some read low, which means the utility delivers electricity it never bills for. The simulation runs an 8-week forensic sweep of the whole fleet. In weeks 1 and 2, every meter is compared against its neighbors on the same local transformer, in the same weather, so a meter that drifts stands out. Early finds scroll through the feed: 240 BAYSHORE meters reading 3 to 6% low with a known hardware-fault fingerprint, and 1,900 MIDTOWN meters with missing readings in a failing-radio pattern that the billing software has been quietly smoothing over for years, because filling gaps is its job. The finished scan totals the leak: 6,140 under-reading meters averaging 4.1% low, worth $3.1 million a year in unbilled electricity; 11,300 meters that lost their wireless connection, forcing 9,400 drive-out manual reads a year; and 480 meters suspected of being stuck entirely.
Three recommendations go to the metering engineering team, and nothing happens until they approve. Week 2: the revenue recovery program: a replacement list ranked by how much money each faulty meter is leaking, 14 new radio collectors to reconnect the cut-off meters, and corrected back-bills only where the drift can be proven from the readings. Annualized value: $3.1 million recovered plus $840,000 in avoided truck trips. By week 3 the first 800 replacements are recovering money at a $0.4 million yearly pace and the disconnected-meter count is down 3,100. Weeks 3 and 4 are the twist: the meters' voltage readings turn the fleet into a 412,000-point sensor network for the grid itself. Three clusters of homes show voltage dipping together, and each cluster maps to a single neighborhood transformer: T-4471 in RIVERSIDE (214 customers), T-8809 in BAYSHORE (186), and T-2213 in MIDTOWN (151). The pattern matches overheating insulation 4 to 8 weeks before the transformer fails. The sweep also flags 61 homes with loose wiring connections at the meter socket, the kind that starts fires. The second approval dispatches planned transformer replacements and same-week safety visits. T-4471 is swapped in a scheduled 2-hour window, and the teardown confirms it was weeks from failing.
Week 6 asks the question every metering executive faces: the utility is about to spend $126 million on next-generation meters, so which neighborhoods get them first? The software scores every area on meter failure density, outage exposure (where new meters' instant outage alerts help most), adoption of solar, batteries, and hourly rates (where detailed readings matter most), revenue risk, and network health. The third recommendation is a value-first plan: Wave 1 goes to BAYSHORE and RIVERSIDE (115,000 meters, 22% of the budget, 54% of the total benefit, paying for itself in 19 months); Wave 2 covers NORTHGATE, MIDTOWN, and SOUTH YARD; Wave 3, the quiet areas where the old meters still work fine, waits 30 months. The simulation ends at week 8 with the board packet: $3.1 million a year in leaks found, $840,000 in truck trips avoided, 3 failing transformers caught, 61 fire hazards fixed, and every number traceable back to the raw meter readings behind it. The closing tiles: revenue recovered $2.9 million a year and climbing at close (against the $3.1 million leak), Wave 1 capturing 54% of the benefit for 22% of the spend, and failing transformers caught 3 of 3, versus 0 of 3 with one failing live.
Every system did its job and nobody read the whole story. The billing software validated the drifting meters' readings and sent the bills out low. The meter database stored the voltage data and moved on. The collection system kept collecting. So the 6,140 silent meters keep leaking $60,000 a week, the trucks keep driving out to read disconnected meters, and transformer T-4471 fails at evening peak: 214 customers dark for 6 hours, an oil spill to clean up, and emergency overtime, while the warning sat unread in the data. The new-meter rollout goes oldest-first, spending $28 million of Wave 1 on the quietest streets, and the regulator phases the $126 million budget because the business case is a vendor brochure.
The software scans billions of readings and compares every meter against its weather-matched neighbors, which is how it catches the ones reading low. It separates tampering from hardware faults by their distinct fingerprints, so the right team goes to each address. It reads the voltage channel the utility already owned and finds three dying transformers and 61 fire hazards. Metering engineers approve every list, and the rollout plan goes to a steering committee. The upgrade is sequenced by value instead of by age: Wave 1 lands where the failures, outages, and revenue risk actually are, capturing 54% of the benefit for 22% of the spend and paying for itself in 19 months, with the measured results from the old fleet serving as the regulator's evidence that the new spend pays.
| KPI | Without GridCORTEX | With GridCORTEX | Delta |
|---|---|---|---|
| Under-registering metersmeters reading low, so customers get billed for less electricity than they used | 6,140 leaking silently | found & replaced | $3.1M/yr recovered |
| Mesh orphansmeters that lost their wireless connection, so someone must drive out to read them | 11,300 · 9,400 truck rolls/yr | re-homed via 14 collectors | $840K/yr in trips avoided |
| Failing transformersneighborhood transformers whose voltage patterns showed they were weeks from dying | T-4471 fails at peak, 6 hrs | 3 caught, planned windows | the voltage data was the clue |
| Loose-neutral / hot socketsdangerous loose wiring at the meter socket, a known house-fire risk | found by fire dept | 61 flagged, same-week visits | fires prevented, not investigated |
| Back-billing basisthe evidence for correcting past under-billed accounts | none; no evidence | proof from the readings themselves | defensible if challenged |
| Theft vs hardwaretelling meter tampering apart from honest equipment failure | one angry bucket | separated by signature | right teams to right doors |
| Wave-1 targetingwhich neighborhoods get the new meters first, and why | oldest meters first | BAYSHORE + RIVERSIDE by value | 54% of benefit for 22% of spend |
| Wave-1 paybackhow long before Wave 1's savings cover its own cost | unmeasured | 19 months, self-funding | a case the CFO signs |
| Prudence casethe evidence shown to the regulator that the $126M is being spent wisely | vendor brochure math | measured results from the old fleet | the commission approves |
| Reliability benefitwhether the new meters' instant outage alerts land where outages actually happen | spread thin | aimed at the outage-prone areas | shorter outages where it counts |
| Quiet neighborhoodsareas where the old meters still work fine | re-metered first | deferred 30 months, still working | capital respected |
| Fleet reporthow the health of 412,000 meters gets reported | assembled quarterly by hand | continuous, traceable to raw readings | built for auditors |
| Revenue recovered / yr (headline tile)unbilled electricity actually being recaptured per year by the close of the program | $0 | $2.9M annualized at close | the leak, reversed |
Two real exposures move. Troubleshooters stop being sent to hang and retrieve recorders on hunches, and transformers that were going to fail in service get changed out on a planned outage instead of failing hot, which is when oil, fire, and an unplanned energized job all arrive together.
Counted in units you already track:
Truck roll hours come back to troubleshooters and meter techs, and analysis hours come back to the distribution engineer who currently writes queries by hand.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Field labor and vehicle | truck rolls avoided x crew hours per roll x your loaded crew rate, plus miles avoided x your fleet cost per mile |
| Emergent to planned conversion | your own cost premium of an emergent changeout over a planned one x the changeouts you convert |
| Overtime and callout | after hours callouts avoided x your callout minimum and overtime multiplier |
| Customer equipment claims | your average paid claim for equipment damage from sustained voltage problems x the claims avoided by finding the condition first |
| Deferred replacement | your unit cost per distribution transformer x the units you defer once measured loading shows they have headroom |
You pay for the scoped engagement that builds and runs this and the compute to process the full meter population rather than a sample, for the integration into your MDM, your GIS, and your EAM, the enterprise asset management system, and for your engineers to validate the anomaly queue against a season of known outcomes. The honest prerequisite is meter to transformer connectivity data. If your GIS does not know which meters sit under which transformer, fixing that is the first project.
Payback is dominated by avoided truck rolls and by the emergent to planned conversion on transformer changeouts, because both are in records you already keep. Treat customer claims as upside.
Most of what this finds is fixed at a desk. A multiplier correction, a missing service point, or an account that silently stopped billing is a customer information system change and nobody drives anywhere, which is the honest and unglamorous part of the value. Where a field visit is required it is a targeted one, so the confrontation risk that comes with any theft investigation lands on the smaller set of cases the evidence actually supports.
Counted in units you already track:
The loss analyst gets the monthly close back and spends that time working causes instead of assembling the balance.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Recovered unbilled revenue | accounts restored to correct billing x their corrected monthly amount x months forward, plus your allowed back bill period x your own recovery rate |
| Analyst labor | energy balance close and exception review hours avoided x your loaded analyst rate |
| Field investigation efficiency | unproductive investigation visits avoided x your fully loaded cost per field visit |
| Loss reduction | the portion of unaccounted for energy you attribute to a fixable cause x your average retail rate, using your attribution and not ours |
| Audit and reconciliation | hours your finance group spends today reconciling billed revenue to energy delivered x your loaded rate |
You pay for the GridCORTEX analytics service, for integrations to your meter data management system, customer information system, mapping system, and distribution transformer records, and for your own analyst time to agree the attribution rules and validate the first corrections before anything posts. The data quality work on meter to transformer to service point mapping is the real cost, and it is yours.
Payback is usually led by accounts restored to correct billing, because that is money you can audit inside one cycle, with theft recovery behind it and loss attribution last. Model the desk corrections first, since they are the fastest and the least contested.
This is one of the few cases where the safety claim is direct rather than indirect. An energized conductor that protection cannot see is a public contact hazard and an ignition source, and detecting it in milliseconds at the device is the difference between isolating a section and patrolling a feeder looking for smoke.
Counted in units you already track:
Patrol and troubleshooting hours come back to the field, and waveform analysis hours come back to protection engineering.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Patrol labor and vehicle | patrol hours avoided x your loaded crew rate, plus miles avoided x your fleet cost per mile |
| Ignition exposure | your own insured and uninsured exposure per ignition event x the share of ignitions you believe device level detection would have caught, a share you set, not us |
| Shutoff scope | your cost per customer hour of a public safety power shutoff x the customer hours removed when de energization can be scoped to a section instead of a feeder |
| Restoration labor | outage restoration hours avoided x your loaded crew rate, where targeted isolation replaces a full feeder lockout |
| Protection engineering | waveform analysis hours avoided x your loaded protection engineer rate |
This one has hardware in it and we will not hide that. You pay for the edge compute units at each instrumented node, for the installation, including the outage or the live work to mount and wire them, for the communications bandwidth to carry alerts, for the platform, and for periodic model refresh. Your protection engineers also need time to tune alert thresholds through a season, because an alert stream your shift supervisor stops trusting is worth nothing.
Payback on the labor alone is slow, because patrol hours are not that large a number. In fire risk territory the case is carried by ignition exposure and shutoff scope, and those depend on your own risk numbers, so build the case with your risk group in the room from the start.
What is this, exactly? It is AI software: intelligent agents and models built and delivered by SoftServe, running on NVIDIA accelerated computing. It is not a hardware appliance and it does not replace the systems you run today. It deploys in your own cloud or on your premises, connects read-only to your existing systems, and recommends; your people approve every action, starting in shadow mode until it earns trust.
A population-scale analytics engine that reads every meter's interval data as a diagnostic sensor; distribution engineers get a ranked anomaly queue: transformers showing early failure signatures, phase imbalance, and degraded secondary service. The demo above uses synthetic data; everything below describes what the real deployment needs from your organization.
| Your system | Typical products | How we connect |
|---|---|---|
| Metering (AMI head-end and meter data management) | Itron, Landis+Gyr, Aclara; Oracle or Itron meter data systems | database replica refreshed nightly |
| Geographic Information System (GIS) | Esri ArcGIS Utility Network, GE Smallworld | read-only API |
| Outage Management System (OMS) | GE PowerOn, Oracle NMS | scheduled file export (CSV or CIM XML) |
| Customer Information System (CIS) and billing | Oracle CC&B, SAP IS-U | database replica refreshed nightly |
| SCADA historian for transformer loading | AVEVA PI System, AspenTech eDNA | historian mirror (one-way feed) |
| Recloser and switch controllers (distribution automation devices) | SEL, Eaton Cooper Power, G&W | event stream (read-only) |
| Advanced Distribution Management System (ADMS/DMS) | Schneider EcoStruxure ADMS, GE Vernova PowerOn | read-only API |
Runs in your own cloud account on GPU instances, from a nightly read-only replica of the meter data store, with no connection to the head-end's control functions and no meter commands. It starts in shadow mode, scoring history against known failures before anyone acts on a prediction.
The Approve button you just clicked in the demo above is the real workflow. This is what it looks like on the screen of a distribution engineer in the GridCORTEX console:
Approve creates a draft inspection work order in the EAM via its API with meter evidence attached, in pending status for the planner to release; the anomaly is written as an annotation on the transformer record in GIS. GridCORTEX never controls any field device.
Detection is automatic from AMI interval feeds; engineers can mark a finding field-confirmed or false with one click.
Interval reads arrive from the AMI head-end every 4 to 24 hours depending on your network; each anomaly card shows the newest read behind it.
Lives as a ranked anomaly queue in the GridCORTEX console feeding the EAM; confirmed failure signatures push to email. The console runs in a browser beside your existing screens on day one; embedding into your own systems is a roadmap step once the read-only phase has earned trust. Approve, Modify, and Decline are all captured in an audit trail your compliance team can pull, and GridCORTEX never blocks or overrides anything in the systems you run today.
The fair question from any VP of Metering or Grid Mod: "We have a head-end, an MDM, and a VEE process, what's new here?" Here's the honest answer.
When someone asks "what did it actually calculate?", this is the list. In the simulation these factors drive the storyline; in a pilot they are computed from your head-end, MDM, GIS, and outage history.
Presenter's one-liner: "Fourteen years of meters, fourteen billion intervals a year, and nobody was listening. The forensics found $3.1M walking out the door, three transformers about to fail, and eleven thousand meters we were reading by truck. Then it sequenced AMI 2.0 by value instead of age, Wave 1 pays for itself in 19 months and funds the rest. The meters were talking the whole time."