By Sagar Shankaran, Founder of CallSphere
Why a nightly per-account check against its own consumption history became economic in 2026, and what it does to August high-bill complaints and leak credits.
Key takeaways
An 8,400-connection water system on hourly AMI collection produces 201,600 meter reads a day. Over a month that is roughly six million individual records landing in the head-end — Sensus FlexNet, Neptune's portal, Badger Beacon, Itron, whichever one the district bought when it did the changeout. Ask the utility billing supervisor how many of those reads get looked at, and the honest answer is: one threshold looks at them. Continuous flow for 24 or 72 hours, depending on how the vendor set it up on install day. Everything else passes straight through to billing untouched.
That threshold is not useless. It catches the stuck toilet flapper and the irrigation valve that failed open. But it only ever answers one question, and it answers it late. A continuous-flow flag is just a meter that never reads zero for a set number of hours — which means a leak that stops for twenty minutes a day is invisible to it, and so is every other kind of trouble a meter can tell you about.
Meanwhile the exception report that lands on the billing clerk's desk on the second of the month has 340 lines on it, she has four days to get bills out the door, and she works the top of the list until she runs out of time. That is not sloppiness. That is arithmetic.
The reads come out of the head-end and import into Tyler's Munis or Incode, Springbrook, Caselle, Central Square or Harris NorthStar. The system throws exceptions for high consumption, zero consumption, negative consumption and estimated reads. The clerk sorts by dollar impact, calls the ten biggest, sends the meter tech to check three registers, and releases the rest so the bills mail on time, because the late fee schedule and the shutoff calendar in the rate ordinance are keyed to the mail date.
Then August happens. Irrigation season runs the residential bill up two to four times, the tiered rate structure amplifies it, and the front counter starts ringing on the ninth. Every one of those calls runs fourteen minutes: the customer disputes the read, the CSR pulls the interval graph, they find the pool fill or the sprinkler clock set to every day instead of every other day, and somebody explains the leak adjustment policy — usually one credit per customer per rolling twelve months, at the wholesale rate, with a plumber's receipt required.
The utility eats it twice, in staff time and in the credit. A share of that water never gets billed at all, which surfaces a year later in the AWWA M36 water audit and makes the non-revenue water number the board asks about every budget cycle.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
Here is the check the district always wanted: take every account, every night, compare last night's 24 hourly reads against that same account's own two-year history adjusted for weather, its class, its meter size and whether the irrigation account next door moved the same way — then say in one sentence what changed and what somebody should do about it.
In 2025 that was not a serious proposal at this size. A real read-through on eight thousand accounts every night, with an explanation attached, cost more than the water it would have saved. What changed is price: running strong models fell roughly tenfold from 2025 into 2026, and doing the high-volume first pass on hardware in the building rather than over the internet runs about ninety percent cheaper again. The check was not invented in 2026. It got cheap enough to run on everybody, every night, instead of on the top ten lines of an exception report.
flowchart TD
A["AMI head-end drops last night's hourly reads"] --> B["Every account checked against its own 24-month history"]
B --> C{"Flow never drops to zero for 18 hours?"}
C -->|Yes| D["Leak letter queued, door hanger printed for the route"]
C -->|No| E{"Jump versus the same week last year, weather adjusted?"}
E -->|Yes| F["Held for the meter tech: register, tamper, or irrigation clock"]
E -->|No| G["Read passes clean into Munis for billing"]
D --> H["Billing supervisor works the queue at 8 a.m."]
F --> H
H --> I["Customer called before the bill mails, not after"]
6:00 a.m. The overnight run finishes. Out of 8,400 accounts, 61 are flagged. Not 340 — 61, because the check is comparing each account to itself rather than to a flat threshold, so the retiree who uses 1,800 gallons a month and suddenly uses 4,100 gets caught while the family of six at 12,000 gallons in July does not get flagged just for being a family of six.
8:10 a.m. The billing supervisor opens the queue. Each line has a plain sentence: "Account 44-1027, Elm Street, flow has not returned to zero since Thursday 11 p.m., about 190 gallons a day above baseline, consistent with a running toilet." Nineteen get leak letters and door hangers. Eleven look like irrigation clocks and get a courtesy call. Six show the read pattern of a register that stopped and started — those go to the meter tech with a work order, because that is meter accuracy, not customer use.
9:30 a.m. Four commercial accounts show 2 a.m. flow that never used to be there — a restaurant with a failed ice machine solenoid, a car wash whose reclaim system is off. Both get a phone call. Both would otherwise have opened the August bill, called angry, and asked for an adjustment the ordinance does not allow on commercial service.
Assumptions: 8,400 accounts, hourly reads, run every night. Assume a full check on one account for one night costs six hundredths of a cent — an illustration, not a quote, but in the right neighborhood at 2026 prices with the first pass running on a box in the building. Assume 120 high-bill complaint calls in August at 14 minutes each, a CSR loaded at $31 an hour, 38 leak adjustments averaging $61 in credit, and that catching leaks two to five weeks earlier cuts both by about half.
| Item | Basis | Monthly |
| Nightly check, all accounts | 8,400 x 30 x $0.0006 | $151 |
| High-bill calls handled | 120 x 14 min x $31/hr | $868 |
| Leak adjustment credits | 38 x $61 | $2,318 |
| Estimated reduction at 50% | calls + credits | $1,593 |
| Net in the peak month | $1,593 less $151 | $1,442 |
That is one month of one line item. It ignores the water itself, which is the bigger number at any system buying wholesale, and it ignores the meter-register catches, which are pure recovered revenue rather than avoided cost. Prove it the boring way: run the check for ninety days without acting on it, then compare what it flagged against what actually turned into a complaint call or an adjustment. If it did not see them coming, do not buy it.
It cannot see a leak on the utility's side of the meter. Distribution losses — the weeping joint under the county road, the 4-inch that has been feeding a storm drain for two years — never touch a customer meter. That is still leak detection surveys, acoustic loggers, district metered areas and night-flow analysis, and no amount of read-checking substitutes for it.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
It will also embarrass you if you let it write to the customer without a person in the middle. A vacant house that just sold, a new pool, a returning snowbird and a broken service line all look similar for the first two days, and an automated "you have a leak" letter to a customer who just filled a pool ends up read aloud at a council meeting. Every letter should pass a person's eyes, even at four seconds a line.
And it does not fix a bad meter population. If a third of the fleet is twenty years old and under-registering, the check makes that visible and you have a capital problem, not a software problem — a budget conversation with the board, not a quick win.
No, it sits on top of it. Keep the vendor's continuous-flow flag — it is fine at what it does. The added value is everything the flag cannot express: intermittent leaks, register failures, tamper patterns, commercial equipment running overnight, and a plain-English sentence per account so the clerk does not have to interpret a graph on every line.
Less, but not nothing. With monthly reads you can still compare each account against its own history and the weather, which beats a flat threshold and shrinks the exception report. The real gain arrives with interval data — one more reason to put the changeout in the capital improvement plan.
Billing owns the queue, operations owns the meter work orders that come out of it. The failure mode is when it belongs to nobody, the flags accumulate, and in month four the queue has 900 lines in it. Put a name on it in the job description and put the count in the monthly board report.
It mostly makes the policy cheaper to run. Most ordinances credit at some reduced rate when the customer produces a repair receipt within a set window. If the utility notified the customer three weeks earlier, the leak volume is smaller, the credit is smaller, and the argument at the counter is shorter because there is a dated letter in the file.
Pull last August's read data and last August's complaint log, and check them against each other by hand for a hundred accounts. How many of the customers who called angry were visible in the interval data two, three or four weeks earlier? At most systems the answer is uncomfortable, and that number is the entire business case — no vendor demo required.
One connected note: the calls this generates all arrive at the front counter, and they arrive in a cluster the week the letters land. CallSphere builds AI voice and chat agents that answer the utility's customer line and web chat around the clock — taking the account number, explaining what the leak letter means, booking the meter tech visit, and handing anything about a bill dispute or a shutoff to staff. It does not read your meters or calculate your adjustments. It keeps a two-person office from losing August to the phone.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
How water systems use a cheap-model first pass and a strong-model escalation so nobody drives out at 2 a.m. for a SCADA alarm that already cleared itself.
A monthly drone flight over the standpipe, lagoon berms and booster station catches the deficiencies a sanitary survey would otherwise find three years late.
Non-English callers wait longest on the 2am utility line. What 70-language live translation does to dispatch time, interpreter cost and the Title VI record.
How a 4,200-connection water system uses Claude Cowork or ChatGPT Work to draft its Consumer Confidence Report, with a source cited for every number in it.
170 portal logins a month for utility and vendor invoices. What an overnight agent run does to the AP clerk's first week, the late fees and the leaks.
Cloned voices target municipal accounts payable. The callback, letterhead, dual-approval and prenote procedure that protects a city's vendor banking changes.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI