The wind picks up at 6:47 on a Thursday evening in July. A feeder trips, and nine thousand customers go dark in the same minute. In the office, the evening shift is two customer service reps and a supervisor - a perfectly reasonable staffing level for a normal Thursday, which this no longer is. Within four minutes every trunk is busy. Caller number two hundred and something is listening to hold music. So is the woman on Maple Street whose husband's oxygen concentrator just went silent, and the man who is calling because a line is down across his driveway and still sparking. The queue doesn't know which callers those are. Neither does anyone else.
Outage day is a queue-math problem, not a staffing problem
On a normal day, your call arrival rate and your answer rate are in balance, give or take a lunch rush. On outage day, the arrivals don't spread out - they detonate. A single feeder event puts thousands of households into the dark at the same instant, and a large share of them reach for the phone within the first quarter hour. Industry write-ups on storm operations describe surge multiples many times baseline volume; your own storm logs will give you your exact number, and it is almost certainly a number no human roster can meet.
The queue math is unforgiving. When calls arrive faster than agents can clear them, the line doesn't grow gracefully - it grows without bound, and then abandonment kicks in. And abandonment on outage day has a special property: nobody who abandons is done. They call back in ten minutes, which means your arrival rate now includes the retries, which makes the queue longer, which creates more retries. The surge feeds itself. If you're a regulated utility, there's a second cost: the answer-speed and abandonment metrics your commission watches don't pause for weather, and one bad storm can wreck a month's service-quality report.
You cannot staff a human call center for a load that arrives a few days a year at many times your normal volume - and the good news is you don't have to. Overtime, mutual-aid call centers, and overflow BPOs all try to solve the surge by adding humans, which means paying year-round for capacity you need a handful of days, or paying storm rates for agents who don't know your territory. The surge isn't a staffing problem. It's a sorting-and-answering problem, and most of the sorting is trivial.
Most outage calls are the same call
Pull the recordings from your last storm event and listen to twenty calls. Almost all of them are one of three questions: Do you know my power is out? When will it be back? Did my report actually register? For everyone on the same feeder, the honest answer is identical - yes, we know; here's the estimate; yes, you're counted. It is the most repetitive work your phone line ever does, which is exactly why making a human say it four hundred times while the queue grows is the wrong use of your humans.
But buried in that queue of identical calls is a small number that are nothing like the others. The downed wire that's still energized. The tree resting on a line over a school sidewalk. The smell of gas where a service was pulled off a house. The caller on home dialysis or oxygen whose outage is a medical event, not an inconvenience. These calls cannot wait in position 87 behind status checks - and on a hold-music line, that's precisely where they wait, because hold music treats every caller the same.
The bulk of the outage-day queue is one repeated question; the risk is the handful of calls that aren't - and the first job of your phone line is telling them apart in seconds. Any answering setup that can't ask "are you calling about a downed line, a safety hazard, or a power outage?" on the first ring, and act differently on the answer, isn't triaging. It's just holding.
Silence is what makes the surge worse
What customers do when the utility doesn't answer is well documented, and none of it helps you. They redial, inflating your queue. They call 911, pushing your outage into the emergency dispatch system. They flag down your line crews, who now spend restoration time doing customer service in a bucket truck. And they post - and in a municipal utility's case, the person reading the post is often on the council that approves your budget.
The satisfaction research says the same thing from the other direction. J.D. Power's utility studies have found, year after year, that proactive communication during outages is one of the strongest levers on customer satisfaction - customers who hear from their utility during an event rate it dramatically higher than customers left in the dark twice over. You don't need the study, though. You already know it from the shape of your own storm days: the second hour of a surge is angrier than the first, and the difference isn't the outage. It's the silence.
Every call you don't answer comes back as two more calls, a 911 diversion, a flagged-down crew, or a post the mayor reads - answering once, immediately, is the cheapest version of the conversation. An answered call with a real restoration estimate ends the loop. An abandoned call starts one.
What surge-proof answering actually looks like
The fix is not more hold music with a better apology, and it's not an IVR that reads a county-wide outage banner to a caller who wants to know about their street. Coverage that actually absorbs a surge has to do what your best storm-desk veteran does, at whatever volume the storm produces:
- Answer every call on the first ring, at any concurrency - the four hundredth simultaneous caller gets the same instant pickup as the first, in their language.
- Look up the address and answer the actual question - confirm the outage is known, give the current restoration estimate from your outage management system, and confirm the caller's report is logged.
- Log the reports you don't have yet - clustered new reports are how nested outages get found; every captured address sharpens the picture your operations team is working from.
- Triage hazards in the first ten seconds - downed wire, gas smell, medical-equipment dependency - and pull those calls out of the queue immediately, to your dispatcher or on-call human, with everything already collected.
- Say the same thing on every channel - the website chat that "check our website" points to should be working from the same outage data as the phone line, not a contact form.
This is the shape of what an AI agent built for utilities and municipal lines does: a voice agent on the phone number and a chat agent on the website, both grounded in your outage data and your escalation rules. The point isn't replacing your storm desk - it's making sure every status call is answered instantly and every hazard call reaches a human in seconds, no matter what the queue looks like. Your people keep the judgment calls and the restoration work. The agent keeps the four hundred identical conversations.
Start small: the one-storm test
You don't need to re-architect your contact center to find out whether this holds up. Run the cheapest possible experiment: put an AI agent on overflow only. Your team answers as they always do; the agent picks up the calls that would otherwise hit hold music - after-hours, and whenever the queue passes the depth you choose. Nothing else about your operation changes.
Then wait for weather. After your first real event, pull three numbers: how many calls the agent answered that would have been abandoned (compare against your last comparable storm), how many hazard and medical calls it escalated and how fast, and what your repeat-call rate did. Set the results against the cost - with per-resolution pricing, that's one line of arithmetic, not a capacity-planning exercise. If the numbers don't clear your bar, switch it off and you've spent one storm finding out.
If they do, widen the funnel next storm season. Setup takes minutes, and the first event where your abandonment rate stays flat while nine thousand customers call at once will tell you more than any vendor claim. The weather doesn't schedule around your shift roster. Outage day is coming either way - the only question is whether your phone line is ready for it, or whether Maple Street gets hold music again.