Social media listening for utilities that gets hazard posts to the right desk and possible outages to the control room in minutes
Outage and billing posts triaged and answered on-brand.
See one case, screen by screen ↓A street goes dark, a wire hangs off a pole by a school, a bill doubles: the first report is often a post on X, a Facebook comment or an Instagram story, not a call. A fault on the low-voltage side of a distribution transformer may never operate a monitored device, and smart-meter last-gasp messages on power-line communication can take from minutes to over an hour to come through. Today one to four people watch a keyword stream and the utility's own pages, copy outage posts into an email or a chat to the control room, look up the outage map by hand and write replies from a document of approved phrases.
Under that load, three posts from one place look the same as three posts from three places, so the outage desk hears late or not at all. Hazard posts wait in the same queue as praise, and night and weekend cover is thin. Replies go out slow or off-policy: a restoration time the outage desk has already changed, a public request for an account number. And the billing complaint on X is never counted with the same complaint on the phone, so Voice of Customer never sees it.
Estimated split for one neighbourhood outage on an evening shift, in minutes, by hand and with the solution. The outage controller's own check takes about as long either way.
Six agents read, place and triage every public post and raise possible outages; the social care agent approves every reply.
Tuesday 13 October 2026, Calder Bay. Hana Okafor runs the evening shift for the fictional Lumina Utilities, which serves electricity and water across eight districts. Here is what happens, screen by screen, in the working solution.
Since 06:00 the agents have read 1,286 posts; 214 are about a Lumina service and 41 were set aside in the last hour — jokes, adverts, other businesses named Lumina, automated accounts and duplicates. Every post that names a street or district is a pin on the Calder Bay map, coloured outage, hazard, billing, water or thanks. Two outage posts from Kestrel Park have just come in, and the rail is watching them.
“2 outage posts in 1 min 53 s, 290 m apart. One more within 10 minutes and 1.5 km raises a possible outage.”
Mila Kovac posts a photo on X: “Wire hanging off the pole on Ferry Lane by the school, it’s sparking.” It is rated urgency 99, and the hazard spotter pages the emergency desk at 18:32:40 with the post, the photo and Ferry Lane on the map — before any reply is approved. Hana posts safety reply 4.1 at 18:33 in one click.
“Please keep everyone at least 10 metres away and do not touch the wire or anything it is touching.”
The post page shows what the agents read: Power Services, Safety Hazard, Ferry Lane in Old Mill, 860 followers and 31 reposts, and why the urgency is 99 — intent 50, words 30. The photo is read too: an overhead wire hanging low from a wooden pole, sparks at the end, a school sign in view. The emergency desk dispatched crew E-14 at 18:34, and the feeder OM-03 section was isolated remotely at 18:36.
“Safety reply 4.1 posted by Hana Okafor · one click — the only reply that may skip review”
Joel Abernathy, 2,140 followers: “my bill went from $142 to $389 this month and I’ve been on hold 40 min.” Hana moves him to a private message under guideline 4.4. With his consent, the account matcher finds three estimated reads, July to September, because the meter could not be reached; $247.60 of the bill is the catch-up. The case goes to Rafael Duarte in billing, with the option to spread the catch-up over 3 months.
“Please send us a private message with your name and address (not your account number here)”
Owen Talbot on Instagram: “Kestrel Park totally dark. Plover Rd, Tern Close, all of it.” Three posts from three people in 3 min 50 s, within 600 m: the pins merge into one, near Plover Road substation PR-0417, which serves 186 customers on feeder KP-11. The network check shows no known outage, no planned works, no alarm on feeder KP-11, the next smart-meter alarm batch due at 18:55 — and no phone calls yet.
The cluster watcher never declares an outage: “The outage controller confirms or dismisses.”
Hana sends the alert to Theo Marsh in the control room with the posts, the map, PR-0417 and the network check. Kestrel Park Bakery joins the pin at 18:47. At 18:48 Grace Odum writes that her father is on oxygen at home; the hazard spotter sends her post to the priority services team for a call within 30 minutes. Theo takes a remote reading of the substation.
“Grace Odum joins the pin: father on home oxygen · Hazard spotter → priority services team · 18:49:10”
Theo confirms incident INC-26-10418: a fuse operated on the low-voltage side of PR-0417, 186 customers off, crew 2 on the way, back by 20:30. The map tints Kestrel Park, the pin shows five posts, and the rail lists the priority customer. Only now does the reply drafter write the outage replies, because the restoration time must come from the outage desk.
“Back by 20:30 · from the outage desk”
The reply drafter writes one reply for each of the five posts and one area post for Kestrel Park, from approved reply 4.2 — and 4.7, the priority customer reply, for Grace Odum — with the incident reference. Each draft shows its checks: tone matches guideline 1, restoration time 20:30 is the outage desk’s, no account details asked in public. Hana approves all six.
“Nothing is posted until you approve it.”
The first phone call about Kestrel Park reaches the contact centre at 18:52. The outage’s time line puts the posts, the alert, the desk, the replies, the calls and the meter alarms on one axis: the alert came 7 minutes before the first call, 10 minutes before the smart-meter alarms at 18:55, and the controller confirmed 4 minutes after the alert. No network alarm ever came — the fault was on the low-voltage side.
“7 min — raised before the first phone call (18:52)”
Lena owns version 7 of the response guidelines: how Lumina sounds, the never rules, the reply clocks, who each post goes to and the approved replies. She tests it on last week’s 412 public replies. Two gave a restoration time the outage desk had already changed during the Saturday storm, and one asked for an account number in public; version 7 would have stopped all three.
“2.2 Give a restoration time the outage desk has not given.” — a never rule
The same evening, seen by the five people who own it — what their shift looked like, and what it looks like now.
Six agents read, place and triage every public post and raise possible outages; the social care agent approves every reply.
Pulls public mentions, comments, tags and place-keyword posts from X, Facebook, Instagram and the community forum; decides whether each is about Lumina and useful; writes a one-line summary and sets aside jokes, ads, bots and posts about other companies, with the reason.
Classifies each post with the Voice of Customer vocabulary — Main service, Sub service, Subject, Service quality, Aspect, Voice type, Sentiment — scores urgency 0–100 and reach, and finds the street, landmark or district named and its nearest substation.
Matches hazard words and reads photos — wire down, sparking, burning smell, open chamber, discoloured water, medical equipment at home — and sends the post to the emergency, water quality or priority services desk with the approved safety reply.
Groups outage posts by place and time; at 3 posts from different people within 1.5 km and 10 minutes it checks the outage map, planned works, feeder alarms, meter alarms and calls, names the nearest substation and its customers, and raises a possible outage.
Drafts on-brand replies from the approved reply for each post’s intent, in plain words with the customer’s first name, takes every outage fact from the outage desk record, and checks each draft against the never rules.
In a private message, after the customer agrees, matches name and address to one account and summarises the last bills, reads (actual or estimated) and payments for the care team.
Approves every public word; the outage controller confirms outages. The agents propose; a named person decides.
Ask about a post, a place, an outage or a reply. Every answer cites its source.
Why was Kestrel Park raised as a possible outage?
Three people posted about Kestrel Park in 3 min 50 s — Dani Price at 18:41, Ruth Mbeki at 18:43 and Owen Talbot at 18:45 — all within 600 m. That meets the rule of 3 posts within 1.5 km and 10 minutes. The network had no alarm on feeder KP-11 and no known outage, and the nearest equipment is Plover Road substation PR-0417.
What did we tell Joel Abernathy about his bill?
Hana moved him to a private message, as guideline 4.4 requires. With his consent the account matcher found three estimated reads, July to September, because the meter could not be reached; $247.60 of the bill is that catch-up. Rafael Duarte in billing is setting up a 3-month spread.
How fast did hazard posts reach the desks today?
All five inside the 5-minute target. Ferry Lane wire down: 1 min 40 s to the emergency desk. Fenwick brown water: 1 min 30 s to water quality. Ashgrove burning smell: 2 min 45 s, the longest. Quay Road chamber cover: 2 min 30 s. Grace Odum’s father on oxygen: 40 s to priority services.
What are customers posting about billing this week?
After outages, billing is the biggest topic: 241 Billing Issue posts and 112 about high usage since Wednesday. Most trace to one cause already in Voice of Customer — bills that stay on estimates when the meter cannot be reached — which gained 38 posts of evidence this week. App log-outs while paying are new: 17 posts, proposed as a new challenge.
13 screens from the working solution, on its sample data. Pick one to see it large.
The possible-outage rule, urgency, who may post and who covers the night are settings. Every change is a new version, approved by corporate communications.
| Setting | Default | Choose from |
|---|---|---|
| Possible outage: outage posts | 3 | from different people · version 4 of the rule, approved 22 Sep |
| Possible outage: within | 1.5 km and 10 minutes | of each other |
| Skip when | Known outage or planned work | on the outage map; posts about planned work get the works notice |
| Send to | Outage desk · control room | with the posts, the equipment and the network check |
| Urgency: intent | Up to 50 | hazard 50 · outage 40 · complaint 24, plus words and photos up to 30 and reach up to 25 |
| Urgency: vulnerable customer | +30 | medical equipment, life support |
| One-click posting | Safety replies 4.1, 4.3 only | everything else needs approval by the social care team |
| Night cover | Sam Whitlock · 22:00–06:00 | hazards and outages only |
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 data · “target” = the design goal, measured in the pilot · “estimated” = our estimate · “typical” = published figures: ESFI analysis of OSHA data (overhead power lines in 42.8 % of workplace electrical fatalities); ACEEE, citing a 2015 national survey (customers the primary source of outage notification at over half of US utilities, SCADA at less than a third); Sprout Social Index 2025 (73 % expect a reply on social within 24 hours or sooner). Smart-meter reporting delays: CLOU. People, Lumina Utilities, Calder Bay, its places, posts, accounts and incidents named on this page are fictional — characters and sample data in the working solution.
Social media listening for utilities reads every public post that names the utility or one of its places, sets aside the noise and acts on the rest: hazards to the right desk, outage posts to the control room, every other post an on-brand draft reply. Social Media Listening does this with six agents, and the social care team approves every public word.
The outage cluster watcher groups outage posts by place and time. When 3 posts from different people arrive within 1.5 km and 10 minutes, it checks the outage map, planned works, feeder and meter alarms and calls, names the nearest substation and raises a possible outage. In the demo it was raised 7 minutes before the first phone call; the outage controller confirms or dismisses it.
The target is 5 minutes from the post to the desk, day and night. The hazard spotter listens for words and photos such as wire down, sparking, burning smell, discoloured water, cover off or on oxygen, and escalates before any reply is approved. In the demo a sparking wire by a school reached the emergency desk in 1 min 40 s.
Yes. Brown or discoloured water goes to the water quality desk with safety reply 4.3, a chamber cover off in the road goes to the water network desk, and low-pressure posts are matched to planned works and answered with the works notice. In the demo, brown water in Fenwick reached the water quality desk in 1 min 30 s.
No reply asks for account details in public. A customer who posts an account number is asked to delete it and send a private message instead. In a private message, with the customer’s consent, the account matcher finds the account and summarises bills, reads and payments for the billing team; it never changes an account.
Yes. The agents never post on their own and never declare an outage. A named member of the social care team approves every reply, only the safety replies can be posted in one click, the outage controller confirms or dismisses every possible outage, and corporate communications approves every change to the guidelines.
Yes. The channels and place keywords, the possible-outage rule (posts, distance and minutes), urgency weights, who may post, one-click posting, night cover, escalation desks and the response guidelines are all settings, saved as new versions and approved by corporate communications.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your response guidelines, escalation contacts, service and subject list, place names and a sample of past posts — and it goes live once every quality gate has passed. We will walk you through it on your own posts first.
Every voice read, and each recurring problem named with an owner and a fix.
1-min demoClosed-Loop Feedback ManagementNo unhappy customer left without a call back, and every recovery measured.
1-min demoComplaint Root Cause AnalysisEach complaint trend proved with system data, owned, and fixed at the cause.
1-min demoWe’ll run Social Media Listening on a sample of your own public posts from a past outage or storm.