Complaint root cause analysis that proves each complaint trend with system data, prices it and puts the fix on an owner's desk
Each complaint trend proved with system data, owned, and fixed at the cause.
See one case, screen by screen ↓Complaints are resolved one at a time. Each one is closed, a goodwill credit is paid, and the category says what the customer felt — billing, supply, reconnection — not the process step that failed. Voice of Customer can name the challenge (“reconnection delayed after payment”), but the payments, reconnection orders, bill runs, meter reads and field orders behind it sit in other systems, and nobody joins them to the complaint.
So causes are argued in the monthly review from opinion — “if there's a delay it's the meters” — and never priced or put on an owner's desk. The same complaints come back, with the repeat calls, truck rolls and escalations they bring. Regulators now look at this capability itself: Ofgem has taken compliance action against an energy supplier over its lack of ability “to identify, understand and resolve the root causes of known issues to prevent further similar complaints”.
Estimated split for proving and owning one complaint cause, in working days, by hand and with the solution.
Seven agents join every complaint to the systems and follow it to the failing process step; the Head of CX accepts each cause.
Lumina Utilities supplies electricity and water. In four weeks, Mon 14 Sep – Sun 11 Oct 2026, it received 2,318 complaints. The largest challenge — reconnection delayed after payment — holds 412 of them, each closed on its own. Here is what happens next, screen by screen, in the working solution.
Leila owns the complaint return to the regulator. Her home screen is a flow: the 7 largest Voice of Customer challenges on the left, the 23,640 records they were joined to in the middle — payments, meter reads and bill runs, refunds, outage notices, contact-centre tasks, field work orders — and the 9 process steps that fail on the right, each with an owner and a status. Band width is complaints.
“The thickest band on the map is Reconnection delayed after payment: 412 complaints in 4 weeks. … a cause is ready for your decision.”
Grace cleared her arrears of $386.40 in the Lumina app at 09:12. The systems recorded the reconnection order at 02:03 the next morning and the meter relay closing at 02:07 — 16 h 55 m without power after paying. She called three times that day. Her complaint, CMP-26-118204, is one of the 412.
“I paid in the app at nine this morning. Why am I sitting in the dark at half past seven at night?”
Leila clicks Follow the thickest band. The band runs from the challenge through Payments → reconnection orders — 3,852 records from payments and billing — and thickens into one cause: Overnight batch 02:00, owned by Samuel Okafor in credit and recovery operations. 371 of the 412 complaints land there; 41 go to a separate cause, remote reconnects that fail.
“Proved with system data. 1,744 of 1,926 reconnection orders were created between 02:00 and 02:14, whatever time the customer paid.”
The proof plots all 1,926 reconnections after an arrears payment, not a sample. Whatever time people paid, the dots stack on one line at 2 am: 90.6 % of orders were created 02:00–02:14, and the median wait was 14 h 22 m — against a customer charter that promises the same day. At the meter, power came on 4 minutes after the order. The process tracer found the step: “Create reconnection order” runs only in job RCN-ORDERS-NIGHTLY, daily at 02:00.
Counter-check passed: “If the meters or the network were slow, the 182 ‘reconnect now’ orders would be slow too. They take 9 minutes. The wait is the job that creates the order.”
In 4 weeks the batch cost Lumina 942 repeat contacts at $9.54, 371 goodwill credits at $25, 38 truck rolls at $310 and 9 escalations to the Office of Utility Standards at $850 — $37,692, about $490K a year. Across all 9 causes the estimated cost of failure is $172,483 in 4 weeks, $2.24M a year. Every unit cost can be edited, and the figures recompute.
Every figure is labelled estimated, with its inputs shown.
The action writer has drafted the fix from the proof, with Samuel Okafor as owner of the step in Lumina's process catalogue. The change: create the reconnection order when the payment posts, not in the 02:00 job. The expected outcome, which the fix verifier will measure: 95 % of reconnections within 30 minutes of payment. Due Fri 30 Oct 2026.
A cause stays a candidate until the Head of CX accepts it; the action is assigned only when its owner accepts it.
The overnight batch moves to Action agreed, next to No final read at move-out (Rania Aziz, due Fri 13 Nov). In progress: auto-validating customer reads (60 % done, due Fri 6 Nov) and checking bank details at the refund request (Tomás Ferreira, due Fri 20 Nov). Four causes are still proving, with the agents joining access notes, permit approval times, meter firmware records and pressure logger readings.
“Priya Nair closes an action only after the fix verifier's before-and-after.”
The second proof explains “Estimated bill despite my reading”. Of 2,904 customer-submitted reads, 661 were validated after their bill run on day 26: the validation queue is staffed Mon–Fri, 08:00–16:00, with a median hold of 4.6 days. The counter-check: the 420 smart-meter reads in the same runs were never late. 298 complaints trace here; Rania's fix auto-validates reads within 15 % of the expected read, with Saturday cover for the rest.
“Smart-meter reads go through the same bill run and are never late; only reads that wait for a person are.”
When a call was wrapped as resolved, the promised callback never became a task. Mariam Chowdhury's fix went live on Wed 23 Sep. Complaints that a callback never came fell from 44 a week to 18, and 1,307 of 1,318 promises since then have a task. The 11 misses all come from the Old Quay walk-in desk, still on paper.
“Not closed yet. The fix verifier wants two more weeks under 15 a week and the Old Quay desk on the new form. Priya Nair closes the action after that.”
The Q3 complaint return to the Office of Utility Standards is drafted from the proven causes: 7,486 complaints, 93.8 % resolved within 8 weeks, 118 escalated to the regulator, 4 of 7 categories with a proven cause. Disconnection and reconnection now reads “Action agreed with Samuel Okafor · due 30 Oct”; the three categories still being traced are labelled under investigation.
Leila signs only after confirming that counts reconcile with the complaint log and each stated cause has a proof and an accepted owner.
The same complaints, seen by the four people who prove, own and close their causes — what their month looked like, and what it looks like now.
Seven agents join every complaint to the systems and follow it to the failing process step; the Head of CX accepts each cause.
Reads each complaint and its resolution notes and maps it to one Voice of Customer challenge and to the regulator's complaint category from the return guide.
Joins each complaint to the account's operational records inside a time window — payments, reconnection orders, bills, reads, refunds, outage notices, contact-centre tasks and work orders — in order of time.
Measures the time between each pair of steps, looks for cut-offs, batch times, weekday patterns and queues, and builds the proof chart, stating the pattern in one sentence with its count.
Matches the step where time is lost to Lumina's process catalogue and its owner, proposes the cause, then tests it against records that took a different path.
Counts the repeat contacts, goodwill credits, truck rolls and regulator escalations joined to each cause and prices them with Lumina's unit costs.
From an accepted cause, drafts one change that removes it, its owner, an outcome the fix verifier can measure and a due date.
Sets a 4-week baseline before a fix, then counts the complaints on the cause and the outcome measure each week, and reports the change and any records that still fail.
Accepts each cause; process owners accept the action; the CI lead closes it. The agents propose; a named person decides.
Ask about a cause, its proof, its cost or the regulator return. Every answer cites the record it came from.
Why do reconnections wait until 2 am?
Because the reconnection order is only created by a nightly job, RCN-ORDERS-NIGHTLY, at 02:00. Grace Liang's payment posted at 09:12; her order was created at 02:03 the next morning and power came on at 02:07. When an agent pressed “reconnect now”, power came on 9 minutes after the order, so the meters are not the cause. Lumina's charter promises the same day.
What does the reconnection delay cost us a year?
An estimated $37,692 in the last 4 weeks, about $490K a year: 942 contacts at $9.54, 371 goodwill credits at $25, 38 after-hours truck rolls and 9 escalations to the regulator. The $9.54 a contact is the industry average; your own figure can be set on the cost screen.
Did the callback fix work?
Yes, so far. Complaints that a promised callback never came fell from 44 a week to 18 since the fix went live on Wed 23 Sep, and 99.2 % of promises now have a task. The 11 misses all come from the Old Quay walk-in desk, which still uses paper. The verifier wants two more weeks under 15 before Priya Nair closes it.
Which causes go in the Q3 return to the regulator?
The return asks for the root causes you have identified and the actions against them, by category. Four categories have a proven cause: billing (reads held past the bill run, no final read at move-out), reconnection (overnight batch 02:00), customer service (callback task) and refunds (manual bank-detail check). The other three are listed as under investigation.
13 screens from the working solution, on its sample data. Pick one to see it large.
Which systems are joined, how records are matched, when a cause is proved, who decides and what a failure costs are all settings, chosen by the utility.
| Setting | Default | Choose from |
|---|---|---|
| Sources joined | 7 of 8 on | Complaints · customer challenges · payments and reconnection orders · meter reads and bill runs · field work orders · outage notices · contact-centre tasks · pressure loggers (requested) — each on or off |
| Join keys | Account + premise | Account number and premise number — never the customer's name |
| Payment → reconnection window | 72 hours | Any window — how far after a payment to look for the order; read → bill is set the same way (35 days) |
| Open a cause trace at | 25 complaints | Any count of complaints in a challenge in 4 weeks |
| Fewest records for a pattern | 50 | Any number — the timing analyst will not fit fewer |
| Counter-check required | On | On · Off — when on, no cause is proposed without one |
| Fix verified after | 2 weeks | Weeks under the target before an action can be closed |
| Approvals | Leila Haddad · Priya Nair | Accepts a cause and signs the return: Head of CX · accepts an action: the process owner · closes an action and approves setting changes: CI lead |
| Unit costs | $9.54 · $310 · $850 | Cost per contact · truck roll · regulator escalation; goodwill credits by type (reconnection $25 · estimated bill $40 · move-out $30) |
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 utility data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Energy Ombudsman, 2024 complaints data — 58 % of cases about billing; CCW, household complaint report 2024 — 222,956 complaints, up 10 %; E Source — $9.54 average cost per utility call). Regulatory context: Ofgem compliance action on a supplier's complaints handling. Cost of failure = Lumina's demo unit costs × the joined counts, shown on the cost screen. Lumina Utilities, the Office of Utility Standards, the people, customers, accounts and records named on this page are fictional — characters and sample data in the working solution.
It is the work of tracing a complaint trend to the process step that causes it, so the cause is fixed once instead of each complaint being resolved on its own. Complaint Root Cause Analysis does it with system data: it joins each complaint to the payments, orders, bills, reads and tasks behind it and proves where the process fails.
Seven agents read the complaints and Voice of Customer challenges, join them by account, premise and time to billing, payments, meter data, field service, outage notices and contact-centre tasks, and follow the thickest band on a flow map to the failing step. Each cause gets a proof chart, a counter-check, a cost of failure and a drafted action for a person to accept.
The timing analyst plots every record, never a sample, and will not fit a pattern to fewer than 50. The process tracer then runs a counter-check against records that took a different path. In the sample case, reconnections an agent pushed through took 9 minutes, which ruled out the meters and pointed to the 02:00 job.
Yes. The cost of failure estimator counts the repeat contacts, goodwill credits, truck rolls and regulator escalations joined to each cause and prices them with your unit costs. On the sample data that is an estimated $2.24M a year behind 9 causes, $490K of it from one batch job; every figure is labelled estimated with its working.
It drafts the return from the proven causes: each category with its counts, the root cause identified, its owner and the action and date against it. Causes still being traced are labelled under investigation, and the Head of CX reviews and signs the return before it is sent.
Yes. The map filters by electricity and water, and the sample causes include water ones such as valves not re-opened after works, where the agents join pressure logger readings from water network telemetry.
Yes. The agents read operational systems and never change a job, an order, a bill or an account. A cause stays a candidate until the Head of CX accepts it, an action is assigned only when its process owner accepts it, and only the continuous improvement lead closes it, after the before-and-after.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your complaint categories, process catalogue, customer charter, unit costs and access to the systems to join — and it goes live once every quality gate has passed. We will walk you through it on your own complaints 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 demoCustomer Satisfaction Index ReportingEvery move in the score explained by name, with the customers’ voices behind each line.
1-min demoWe'll run Complaint Root Cause Analysis on four weeks of your own complaints and the records behind them.