AMI network health monitoring that finds the one failing device and fixes it before the bill run
Comms failures fixed before the bill run, so customers get a real read, not an estimate.
See one case, screen by screen ↓On a radio network, meters pass their reads hop by hop through each other and through pole-top routers to collectors, which backhaul them over cellular or fibre. When one router loses its power tap and its battery runs out, the head-end raises a “meter not responding” alarm for every meter behind it. The analyst starts the shift with an alarm list, an MDM validation report and a spreadsheet of non-communicating meters. Each shows one layer; none shows that 412 meters share one parent, or what that means for Thursday's bill run.
So the fault surfaces days later, as missing billing reads at the cycle cut-off. Every meter without a read at cut-off gets an estimated bill — a likely call, a possible rebill and, repeated, a complaint. And the obvious fix is the expensive one: meter-level orders send a technician to each house for a fault one crew can fix on one pole. Today's read rate is known; the read rate at the cut-off is not, so priorities are set by alarm count, not by bills at risk.
Estimated split for one router or collector fault, in days, by hand and with the solution.
Seven specialist agents watch the reads, find the failing device and forecast the cut-off; the analyst confirms and the supervisor dispatches.
Lumina Utilities reads 1,214,600 smart meters and water endpoints over a radio network. On Monday 12 October at 21:40, router R-4471 on pole 22-118, Harrow Fields Road, Westmere, loses its power supply. Here is what happens next, screen by screen, in the working solution.
Nadia owns read rates for cycles 12–20. The map draws the Harrow Fields part of the network as it really routes: collector C-031 at Westmere substation, 13 routers, and every meter and water endpoint as a dot coloured by read success. Network read success over 24 hours is 99.62% against a 99.5% target. The overnight alarms read 1,287 → 6 incidents, and the next cut-off at risk is cycle 14, Thursday 15 October.
“412 without a read behind R-4471 · 41 re-routed”
The scrubber replays Monday 21:00 to Tuesday 06:40. R-4471 reports AC power lost at 21:40, battery low at 01:58, and its last heartbeat at 02:14. From then on amber spreads out through the meters behind it, and by 03:10 41 have turned blue as they find a new route through R-4468. Beside the map, the cycle 14 cut-off forecast falls from 99.8% to 96.8%, below the 99.5% line.
“If nobody acts, 435 meters on this route have no read at cut-off: 318 electricity and 94 water behind R-4471, and 23 single meters”
The root-cause finder walks the routing tree. 453 meters share R-4471 as their last parent, and they went silent 02:11–02:14, inside 3 minutes of its last heartbeat. Its parent R-4466 and neighbour R-4468 are healthy. The remote fixer pinged it three times at 06:28: no answer. The outage system shows feeder WM-11 energised, and field history shows the battery last replaced in March 2019, rated 6 years. Root cause at 94% confidence, named at 06:31.
“Feeder WM-11 energised, no outage at pole 22-118 — so the fault is the router's own power tap”
A Corvane LR-900 pole-top router on firmware 5.4.2, installed June 2015, on a pole power tap from feeder WM-11. Its event timeline runs from the lost AC power to the outage check. Of the meters behind it, 150 are direct, 200 two hops out and 103 three hops out. Radio neighbours: R-4466 at −78 dBm, healthy; R-4468 at −84 dBm, which took 41 meters; R-4470 at −93 dBm, too weak to re-route. The May 2024 inspection noted the power tap corroded.
Ping, on-demand reads and a work order are one click from the device.
The alarm grouper folded the 1,287 alarms by shared parent device, time and event pattern: 838 of them are R-4471's. Collector C-117 at Ridgeway Depot is dropping its Northwave Mobile backhaul 01:00–05:00, but its reads are buffered — with Callum Reid. On 286 Oakhurst meters, firmware 7.2.1 drifts the clock past 90 seconds; the remote fixer sent a time sync and 279 recovered. 133 Millbrook water endpoints from a 2014 batch have low batteries. The read watcher found 1,749 single meters with missed reads; 1,186 answered an on-demand read at 06:05.
“A root cause below 80% confidence is shown as unexplained”
The next eight route cut-offs, each with the read rate the cut-off forecaster expects at 23:59 on the day and the incident that changes it. Cycle 14 Westmere: 96.80%, 435 estimated bills, 23 if R-4471 is fixed by 15:00 today. Cycle 17 Millbrook: 98.70%, 155 bills, 22 once the batteries are replaced on Monday 19 October. The assumptions are printed under the table.
“Re-registered meters back-fill their stored reads within 2 hours; buffered collectors hold 72 hours; scattered single meters at each route's 30-day average.”
She reads the evidence on the rail and confirms. The root cause turns green, and the work order drafter starts writing WO-26-18842. Nothing goes to the field yet: the order still needs the field meter supervisor's approval, and the 412 meter-level orders are held.
“Root cause confirmed · the Work order drafter is writing WO-26-18842”
Nadia sets “If R-4471 is fixed by” to 15:00 today. Meters re-register and back-fill their stored reads within 2 hours, so the Westmere cut-off goes back to 99.8% with 23 estimated bills. Fixed by Wednesday 12:00 it would leave 29 meters unread; by Thursday 18:00, 141; not fixed, 435.
“Fixed by 15:00 today · 99.8% at cut-off, 23 estimated bills”
WO-26-18842: replace the power supply and battery on R-4471, pole 22-118, Harrow Fields Road. Crew 7, Ewan Talbot, today 13:00–15:00, with an LR-900 power supply unit and a 12 V battery pack. The root-cause finder's evidence goes with it. Marisol approves and it goes to crew 7's tablets. Today's orders avoid 507 meter visits, 1,476 this month.
“412 ‘meter not responding’ orders, one per address, are held. They close when R-4471 is back and each meter reads.”
The report writer drafted it at 06:35 from the night's reads, incidents and forecasts: read success, the six incidents and their actions, what it means for billing and for the field. When Nadia signs, it goes to Hamid Qureshi in billing operations, Callum Reid in telecom, Marisol Ventura in field meter services and the AMI operations team.
“Draft · not sent. The Report writer never sends without your signature.”
The same dark router, seen by the five people who own it — what their morning looked like, and what it looks like now.
Seven specialist agents watch the reads, find the failing device and forecast the cut-off; the analyst confirms and the supervisor dispatches.
Takes the head-end's read and communication statistics every 15 minutes and keeps read success per meter, router, collector, area and billing route. Flags a meter amber after 2 missed intervals and red after 24 hours.
Folds the head-end's alarms and meter events into incidents by shared parent device, time and event pattern. A group smaller than the minimum stays as single meters.
Walks each incident's routing tree to the deepest device all affected meters share; checks its events, its parent and neighbours, pings it, checks the outage system for its feeder and reads its field history. Names the failing device with a confidence and its evidence.
Forecasts the read rate at each billing route's cut-off, with and without each fix, and the estimated bills at risk.
Sends the remote actions it is allowed — pings, on-demand reads, time sync and route refresh — in order of least disruption, and logs every command and its answer on the incident.
Drafts one order per failing device with the evidence, part, pole and slot, and holds the meter-level orders behind it.
Drafts the daily network health report at 06:35 with every number linked to its source.
Confirms each cause; supervisor approves dispatch; engineer approves collector work. The agents propose; a named person decides.
Ask about any device, area or cut-off, or say what to do next. Every answer points to its records.
Why is the Westmere cut-off forecast 96.8%?
Because router R-4471 on pole 22-118 went dark at 02:14. 453 meters route through it; 41 found a new route through R-4468, and 412 (318 electricity, 94 water) have no route, so no read for Thursday's cut-off. With the route's usual 23 single meters, that is 435 estimated bills of 13,600: 96.8% against a 99.5% target. Feeder WM-11 never lost supply, so it is the router's own power tap, not a power cut.
Which incidents put the most bills at risk this week?
Four of the six: R-4471 in Westmere, 412 bills on Thursday's cycle 14; the Millbrook water batteries, 133 bills on cycle 17, covered by Monday's batch order; router R-3982 in Larkspur, 57 bills on cycle 18; and 38 unregistered meters at Saltmarsh Quay on cycle 19. Ridgeway's collector C-117 drops in and out, but its reads are buffered for 72 hours, so cycle 16 is safe if the carrier fixes it by Thursday.
Time sync the Oakhurst meters still failing
Sending. 7 of the 286 Oakhurst meters on firmware 7.2.1 still drift after the 06:22 sync. A time sync is on the list of remote actions I may take without asking; a firmware rollback is not, and stays with Callum Reid.
Send the R-4471 work order to Marisol
Done. WO-26-18842: replace the power supply and battery on R-4471, pole 22-118, Harrow Fields Road; crew 7 (Ewan Talbot), today 13:00–15:00; parts in van stock. The evidence goes with it. I have held the 412 meter-level orders, so nobody visits the houses. It needs Marisol Ventura's approval before it reaches the crew.
13 screens from the working solution, on its sample data. Pick one to see it large.
Read targets, grouping, what the agents may do on their own, and who decides. Every change is saved as a new version.
| Setting | Default | Choose from |
|---|---|---|
| Daily read success target | 99.5% | Raised or lowered in steps; areas and routers below it are flagged on the map and the heatmap |
| Billing read target at cut-off | 99.5% | Raised or lowered in steps; the cut-off forecaster warns when a route will close below it |
| Missed reads before a meter turns amber | 2 | Consecutive 15-minute intervals, raised or lowered in steps |
| Alarm grouping window | 15 min | Alarms inside this window with a shared parent become one incident |
| Smallest group | 10 meters | Fewer meters than this stay single meters |
| Root-cause confidence floor | 80% | Below it the incident is shown as unexplained, with its evidence |
| Check the outage system first | On | On · Off — a device behind a real power cut is not a network fault |
| Approver for field orders | Marisol Ventura | Marisol Ventura · Rosa Lindqvist |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample network data · “target” = the design goal, measured in the live solution · “estimated” = our arithmetic: $1.2M = about 4,800 meter visits a year merged into device orders × $250, the demo utility; the split of a fault's days is an estimate · “typical” = published figures (truck roll $250–$500, S&C Electric via Ventavid). Context: missing end-of-cycle reads drive estimated bills (SmartGrid.gov PECO smart-grid metrics; NETSCOUT, Assuring quality performance of AMI). Lumina Utilities is a fictional demo utility; the people, places, devices and orders named on this page are characters and sample data in the working solution.
It is the work of keeping smart meter reads arriving: tracking read success by meter, router, collector and area, and finding and fixing the network devices that stop them. AMI Network Health Monitoring does it with seven agents that watch every read, fold the head-end's alarms into incidents, name the failing device and forecast the read rate at each billing cut-off. The analyst confirms the cause and the supervisor approves the dispatch.
The alarm grouper folds alarms by shared parent device, time and event pattern. The root-cause finder then walks the routing tree to the deepest device the silent meters share, checks its events, parent and neighbours, pings it, checks the outage system and reads its field history. On the sample night it named router R-4471 behind 412 meters at 94% confidence.
The cut-off forecaster forecasts the read rate at 23:59 on each route's cut-off day, with and without each fix, and the estimated bills at risk, so priorities follow bills at risk, not alarm count. In the sample case, fixing R-4471 by 15:00 brings Thursday's Westmere cut-off from 435 estimated bills back to 23.
Yes. The demo network carries 842,300 electricity meters and 372,300 battery-powered water endpoints, with read rates shown for each. It also catches water-specific faults, such as a 2014 batch of endpoints whose batteries fell below 3.1 V and began transmitting every 4 hours instead of hourly.
Only the ones on its allowed list: ping, on-demand read (up to 3 an hour per meter), time sync for meters and route refresh. A collector reboot waits for the telecom engineer, a firmware push waits for the engineer and a change ticket, and connect or disconnect is never done by an agent.
It drafts one order per failing device and holds the meter-level orders behind it while the device's order is open. In the sample case one order to pole 22-118 replaces 412 meter visits; today's orders avoid 507. Our estimate for the demo utility is about $1.2M a year — 4,800 meter visits × $250.
Yes. The AMI operations analyst confirms every root cause and signs the daily report, the field meter supervisor approves every field order, and the telecom network engineer approves collector and firmware actions. The agents never edit a read, close a head-end alarm or dispatch a crew.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your network inventory, head-end exports, billing cycle calendar, dispatch rules and runbook — and it goes live once every quality gate has passed. We will walk you through it on your own reads and alarms first.
Every held bill released inside its cycle, each estimate with its method and source.
1-min demoEnergy Theft DetectionAn estimated $8.4M a year recovered, with 3 in 5 site visits finding theft.
1-min demoMeter Event Management18,214 alarms become 23 actions, and every safety alarm reaches a person in 5 minutes.
1-min demoWe'll run AMI Network Health Monitoring on a sample of your own head-end reads and alarms.