Water quality complaint investigation that traces every brown-water call upstream to the valve, flush or burst behind it, and drafts the flush, the replies and the regulator notification
Brown-water calls traced to their cause in minutes, the flush and every reply drafted.
See one case, screen by screen ↓Taste, odour and discoloured-water complaints arrive one at a time: a call note, an app report with a photo, an email, a chat. Each one gets advice and a callback promise. The cause sits in other systems. A crew closed a boundary valve at 09:10 and the work order says nothing about water quality. The network model shows the flow on a 1962 cast-iron trunk main turning round at 09:12. A turbidity probe alarms at 10:20. A water quality scientist has to join the calls to the event by hand, across the contact centre system, work orders, GIS, the laboratory and the control system.
So the cluster is seen late, often after 20 or 30 contacts, when the street has already filled the queue. The wrong cause is easy to pick: a burst nearby gets the blame, even when it is downstream and started after the first calls. Every customer still needs the right advice and a reply. A reportable event has a same-day clock to the regulator. And the main that ran brown after valve work in April and August runs brown again.
Estimated split for one discolouration event of about 20 contacts, in working hours, by hand and with the solution.
Six agents read, cluster, trace and draft every complaint; the water quality scientist confirms each cause.
Tuesday 13 October 2026, the south network of Lumina Utilities. At 09:10 a crew closes valve V-2217 on Brackwater Road for a new connection at Orchard Rise. At 10:05 the first call comes in from 14 Millbank Road in Harrow Vale. Here is what happens next, screen by screen, in the working solution.
Each contact is classified in the customer's own words: brown water, brown with a metallic taste, black specks, chlorine taste, white and cloudy. The reader finds the property, its supply zone and the main that feeds it, and marks second contacts. Joan Pettifer at 14 Millbank Road calls again at 10:56. Black specks at 9 Kestrel Park Road from the bathroom basin only are put down to a worn tap washer, not the network.
Joan Pettifer, second call: “Still brown after running the kitchen tap for twenty minutes like I was told.”
Natalie Brooks at 88 Fenwick Street says her son had a stomach ache after drinking a glass that morning. The complaint reader passes it to Grace at 10:39. Grace calls back at 10:52: the boy is well, no fever. She advises bottled water for drinking until the flush is done, and a doctor if symptoms return. It is noted on the event, and the contact's reply card says it was handled by a person.
Agent limit: “Illness mentions go to a person within 5 minutes” · “Never says the water is safe.”
At 11:20 Leila opens her home, a control-room view of the network: reservoirs, trunk and distribution mains, valves, hydrants and probes, with every water quality contact plotted at its supply point. Today there are 35: 3 clusters and 3 single contacts. A banner reads TM-HV-07, flow reversed since 09:12, 0.9 m/s westward. Under the map, a time strip shows the morning's network events, contacts per 10 minutes and turbidity at probe TS-14.
“Cluster · flagged 10:21, at the third contact · WQE-2026-041 · Discoloured water, Harrow Vale · 23 contacts from 21 properties”
Leila presses Trace upstream. The tracer takes the flow at 10:05 from the network model: Harrow Vale has been fed from the east along TM-HV-07 since 09:12. It walks upstream along the 450 mm cast-iron main, lighting each pipe on the map, and tests every event in the 6 hours before the first contact. The standpipe on hydrant H-0931 scores 8 %. The Kestrel Hill reservoir outlet switch scores 5 %.
“What reversed the flow? Valve V-2217, closed 09:10 · first contact 55 min later (model: 48–62)”
Closing V-2217 for the Orchard Rise connection put Harrow Vale's whole supply onto the 1962 main at 0.9 m/s against its usual direction, and lifted iron deposits: 410 µg/l at 14 Millbank Road, 6.1 NTU at TS-14. All 23 contacts are on the reversed main, and it is the fourth event on this main in 12 months. Against it: no water quality note on the work order. The burst on Cobb Lane is ruled out.
Burst on Cobb Lane at 10:42: “Downstream of the cluster, not upstream” · “Started after the first 6 contacts.”
Leila confirms the cause at 11:24. The action planner has drafted FO-3318: a unidirectional flush of 2.6 km of TM-HV-07, east to west with today's flow, from H-1044 at 12:30 through H-1068 and H-1081 to H-1102, at 40 l/s each until turbidity is below 1 NTU for 10 minutes. Crews S2 and S3 under Samir Qureshi; H-1081 drains to Mill Brook, so a dechlorination mat is required. Tom approves at 11:31.
“The planner drafts; only the network control operator can approve a flush. No agent operates a valve or a hydrant.”
The reply writer has drafted one message to the 1,840 properties in the affected area, by SMS, the app and the website's incident page, in English and Arabic, and a reply to each of the 21 properties that contacted Lumina. Joan's reply apologises for the second contact, offers a call that afternoon and the cost of cleaning stained laundry. Leila approves the area message and 20 replies; Natalie Brooks has already had her call.
Area message: “It is iron from older pipes. We are flushing the main 12:30–16:00. Please avoid washing machines until then.”
The action planner asked for follow-up samples at the same three taps. Before the flush, S-77412 at 14 Millbank Road read iron 410 µg/l and turbidity 5.6 NTU, and S-77413 at 3 Ashgrove Close read iron 260 µg/l, against Lumina's limits at the tap of 200 µg/l and 4 NTU. After the flush, S-77420 to S-77422 read iron 70, 90 and 60 µg/l and turbidity 0.4, 0.5 and 0.3 NTU, all within limits, so the event can close.
Agent limit: “Never closes an event while a sample is above a limit.”
23 contacts from one zone in 24 hours is over the threshold of 20, and iron at 14 Millbank Road is above its limit, so the event is reportable by 17:00 the same day. The action planner drafts the notification: the event, its scale, the confirmed cause, the results, the health report, the actions and the prevention asked of network planning. Dr. Adeyemi signs it at 16:10.
Health: “One customer reported a child's stomach ache; called back at 10:52, the child is well. No other health reports.”
The event case shows every cause tested, the six actions with who did each and when, and a timeline of people, agents and the network. The case recorder closes WQE-2026-041 and its 23 contacts with the cause. The note to the work planner asks that valve work on TM-HV-07 is closed slowly, over at least 15 minutes, with a low-velocity pre-flush from H-1044, and that the main goes to the mains renewal review.
“Fourth discolouration event on TM-HV-07 in 12 months, three after valve operations.”
The same brown-water morning, seen by the five people who own it: what their day looked like, and what it looks like now.
Six agents read, cluster, trace and draft every complaint; the water quality scientist confirms each cause.
Reads every call note, app report, email, chat and web form; classifies the taste, odour or colour; records which taps, when it started and what the customer asked; finds the property, its supply zone and the main that feeds it.
Every 5 minutes, groups open contacts in the same supply zone within 500 m and 2 hours; at the third contact raises a cluster, counts the properties downstream on the same mains and alerts the water quality scientist on duty.
Takes the flow directions from the network model at the time of the first contact and walks upstream from every supply point in the cluster. It tests each valve operation, flush, burst, standpipe, source or dosing change and probe reading in the window: on the flow path, inside the modelled travel time, matched by the probes, seen before on this main.
Drafts the unidirectional flushing order with the current flow (hydrants in sequence, flow, turbidity target, crews, drains and pressure checks), the follow-up samples at the same taps, the regulator notification when the event is reportable, and the note to the work planner.
Drafts each customer's reply and one message to every property in the affected area, in plain words: what happened, what we are doing, what to do, and when we will be back in touch. For a second contact it apologises and offers a call.
Keeps the event timeline from every record; when after-flush samples are within limits and the people have signed, closes each contact with its cause and actions and writes the closure record and event report.
Confirms every cause; the operator approves the flush. The agents propose; a named person decides.
Ask in plain words, or say what to change. Every answer points to its records.
Why is Harrow Vale running brown?
Most likely valve V-2217, closed at 09:10, reversed the flow in TM-HV-07. The Orchard Rise connection closed it, and the network model has Harrow Vale fed from the east from 09:12. All 23 contacts sit on that main, and the first came 55 minutes later, inside the modelled travel time. Iron is 410 µg/l at 14 Millbank Road.
Which customers have called twice?
Two: Joan Pettifer, 14 Millbank Road, still brown after running the tap for twenty minutes, and Dawit Bekele, 3 Ashgrove Close, whose washing machine stained a load. Both replies now offer a call this afternoon and the cost of cleaning.
Is the Harrow Vale event reportable?
Yes. 23 contacts from one zone in 24 hours is over the threshold of 20, and iron at 14 Millbank Road is above the 200 µg/l limit. The notification is drafted with the cause, the results and the flush. It is due by 17:00 today, and Dr. Ruth Adeyemi signs it.
Make valve work on TM-HV-07 slow-close from now on
Done. New rule: “Any valve operation on a main with 2 or more discolouration events in 12 months is closed over at least 15 minutes, with a water quality note on its work order.” TM-HV-07 qualifies: 4 events in 12 months, 3 after valve operations. It applies to work orders raised from now on.
13 screens from the working solution, on its sample data. Pick one to see it large.
How contacts are grouped, how far the trace looks, when the regulator is told and who approves what are settings, chosen by the water quality team.
| Setting | Default | Choose from |
|---|---|---|
| Contacts to flag a cluster | 3 | Any number · the cluster is raised at this contact |
| Within distance · within time | 500 m · 2 h | Metres of each other along the network · hours between the first and the last |
| Same supply zone only | On | On · Off — when on, contacts in two zones are linked by a person |
| Look back · trace depth | 6 h · 8 km | Hours of events tested before the first contact · kilometres upstream along the flow, or to the source |
| Cause weights | 30 · 30 · 30 · 10 % | On the flow path · timing against the modelled travel time · reach and probes · history on the main |
| Reportable event | 20 contacts in one zone in 24 hours | Any number of contacts · plus an illness report linked to a cluster, and any sample above a limit at the tap (each On · Off) |
| Regulator notification deadline and signer | 17:00 the same working day · Dr. Ruth Adeyemi | A time on the same working day · signed by the head of water quality |
| Approvers | Flushing: Tom Achterberg · area messages over 500 properties: Leila Haddad | Only the network control operator approves a flush · the water quality scientist approves large area messages |
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, to be measured in the pilot · “estimated” = our estimate · “typical” = published figures (US EPA National Enforcement and Compliance Initiative on drinking water — 43 % of community water systems violated at least one standard in FY2022, mostly monitoring and reporting; United Utilities PR24 representation on C-MeX — industry C-MeX 81.6 in 2020-21, 75.7 in 2023-24). Regulatory context: drinking-water event notification duties in England and Wales (DWI); customer contacts about water quality, a common performance commitment in Ofwat PR24; US EPA customer complaint surveillance guidance. Limits at the tap used in the demo: iron 200 µg/l, manganese 50 µg/l, turbidity 4 NTU. Lumina Utilities, its network, the people, customers, addresses and records named on this page are fictional — characters and sample data in the working solution.
Software that takes taste, odour and discoloured-water complaints from the moment they arrive to a confirmed cause, a flush, a reply for every customer and, when the event is reportable, a signed regulator notification. Water Quality Complaint Investigation adds six agents that read, cluster, trace and draft; the water quality scientist confirms each cause.
The upstream tracer takes the flow directions from the network model at the time of the first contact and walks upstream from every supply point in the cluster. It tests every network event in the window on four things: is it on the flow path, does the first contact fall inside the modelled travel time, do the contacts and probes agree, and has it happened before on this main. Each cause is ranked with the evidence for and against.
The cluster watcher groups open contacts every 5 minutes by supply zone, distance and time, and raises a cluster at the third contact, by default within 500 m and 2 hours in one zone. It counts the properties downstream and alerts the scientist on duty. In the sample case the cluster was flagged at 10:21, at the third of 23 contacts.
The complaint reader passes any illness mention to a person within 5 minutes and flags it on the event; that rule is always on. No reply or area message says the water is safe to drink before signed results are in. An illness report linked to a cluster can also make the event reportable.
Yes. The action planner checks the reportable rules you set (in the demo, 20 contacts from one zone in 24 hours, an illness report linked to a cluster, or any sample above a limit at the tap) and drafts the notification with the cause, results, health reports, actions and prevention. The head of water quality signs it; in the sample case, at 16:10 against a 17:00 deadline.
That is the design goal: 40 % fewer repeat contacts, to be measured in the pilot, by telling the whole affected area first. The reply writer drafts one message to every property downstream and a reply to each customer who contacted you, and Patterns shows which mains keep causing events so the fix can be made there.
Yes. The agents have read-only access and never operate a valve or a hydrant, dispatch a crew or send a message without approval. The water quality scientist confirms the cause and approves replies, only the network control operator approves a flush, and the head of water quality signs every notification.
The Agentic Solution Engine builds and deploys it from your requirements and documents: your water quality event procedures, reply wording, limits at the tap and a sample of past contacts and network events. It goes live once every quality gate has passed. We will walk you through it on your own complaints first.
Leak crews sent where the loss is certain, not where a meter is in doubt.
1-min demoWater Quality Compliance MonitoringExceedances caught the hour they happen, and low chlorine seen two days ahead.
1-min demoLeak Detection AnalyticsSend crews to real leaks, not false alarms.
1-min demoWe'll run Water Quality Complaint Investigation on a sample of your own past water quality contacts and network events.