Voice of customer analytics for utilities, explained.

Voice of customer analytics in five minutes

The life of one complaint, in five steps

1. A customer speaksHelen calls about an estimated bill, then rates the call Unhappy in the app.
2. The words are capturedA recording, an email, a chat log, a survey comment: each lands in its own system.
3. Someone reads themAll of them, or a sample. This is where four similar complaints become one named problem, or stay four.
4. An owner fixes the causeBilling changes the rule that set the readings aside, with a date and a measure.
5. Or the regulator hears firstThe customer calls again, then complains to the commission; several states count those complaints against performance targets.

Step 3 is the subject of this guide. Skip it, and step 5 finds the problems for you. The people in this example are fictional.

Four things that surprise people

Four words you will see on this page

Twelve terms are defined in the one-page summary.

What is voice of customer analytics? Solicited and unsolicited feedback

In shortVoice of customer analytics reads what customers say, whether the utility asked or not, and turns it into named problems with evidence. Feedback the utility asks for, such as surveys, shows how satisfied customers are. Feedback customers give unasked, in calls, emails, chats and complaints, shows what went wrong.

Gartner describes the market for voice of the customer platforms around three kinds of feedback: direct, what customers intend to tell you, whether asked or volunteered; indirect, what they say in the course of service; and inferred, what their behavior and the operation's data show.10 Practitioners often use a plainer split, solicited and unsolicited, and vendors draw the line between the terms in slightly different places.

The channels a utility listens on

Why utilities are different

Running a voice of the customer program at a utility

A voice of the customer program is more than a survey or a dashboard. It has five parts, and most programs are strong on the first and weak on the last two.

1. Listen on every channel customers use, asked and unasked, including complaints to the commission.
2. Read what they say, as much of it as you can, in their own words.
3. Name each recurring problem once, in a shared vocabulary of services and failures, so counts add up across channels and months.
4. Own each problem: a named person in billing, metering, operations or digital who decides on a fix and a date.
5. Measure whether the fix worked, and close the loop with the customers who raised it.

Utility customer satisfaction: how J.D. Power and ACSI measure it

In shortTwo national studies score US utilities: J.D. Power on a 1,000-point scale and the American Customer Satisfaction Index (ACSI) from 0 to 100. Both are low in 2025 and 2026, and both point to bills. J.D. Power redesigned its 2025 electric business study and says those results are not comparable with earlier years. The residential release does not say, so quote 499 as released and do not set it against earlier scores.

ACSI 2026, by type of utility

Scores out of 100; change from the 2025 study. Cooperatives were the only segment to gain. Fieldwork January to December 2025, 33,759 surveys.6

Which parts of the service score lowest

ACSI also scores parts of the customer experience across the industry. In 2026, satisfaction with the call center was tied for the lowest of them.6

What drives the scores

Digital channels are now expected, not a bonus: ACSI says apps and websites "have become core expectations rather than differentiators",6 and J.D. Power's 2026 digital experience study found "28% of utilities in the study still do not offer an app".19

Utility complaints, the state commission, and performance-based regulation

In shortCustomers can complain to their state utility commission, and the utility must answer. Some commissions publish the counts, and some tie money to complaint rates, call answering and satisfaction. New York took $50.1 million from five utilities for missed 2025 customer-service targets.

How a complaint reaches the commission

A customer who cannot get a problem solved files an informal complaint with the commission's consumer office, by phone, mail, email or web form. The commission passes it to the utility, which must respond within a set time: in California, utilities must answer the Consumer Affairs Branch within 20 business days.11 A formal complaint is a separate legal proceeding before an administrative law judge. What each state counts, how it categorizes complaints and what it publishes all differ.

Where customer-service results carry money

Other states publish complaint statistics without tying money to them. Check your own commission's rules and your current rate plan.

Estimated bills: an enforcement topic

Satisfaction and the rate case

J.D. Power reports that "higher levels of customer satisfaction one year prior to a rate case are associated with higher ROE", the return on equity a regulator allows.25 It is a correlation, measured on J.D. Power's earlier index, not proof that satisfaction causes a higher return.26 What a commission sees directly is the complaint record and the service metrics in the rate plan.

What drives utility customer dissatisfaction, and how to get from a complaint theme to a root cause

In shortBills come first: high, rising and hard to predict. Then estimated or wrong bills, outages and how they are communicated, payment trouble and disconnection, and the contact center itself. For some drivers, such as moving in and out or smart-meter rollouts, there is no national figure.

How much of a contact center's volume is about bills? There is no national figure. A 2026 article by practitioners at one large utility and the AI vendor it works with (a different company from the publisher of this guide), reported by TD World, puts billing and usage questions at 35 to 50 percent of inbound calls at large utilities in peak seasons.34

From complaint themes to root causes

A theme says where customers are unhappy. A fix needs more: one named problem, the rule or process behind it, and an owner with a measure. Here is the example from the top of the page, step by step.

The example is fictional. In the US, most electric meters are read remotely, so the same failure often looks different there: a reading the utility already holds, from a customer, a failed smart-meter read or a system change, is not used before the bill run, and the bill goes out on an estimate. Water utilities, where manual and customer readings are more common, may see the example as written.

First-contact resolution, customer effort score, NPS and CSAT: what they measure, and what they miss

In shortThese scores tell you how much is going wrong, not why. Use them to size a problem and to prove that a fix worked. Use customers' own words to find the cause.

Try it: what a Net Promoter Score hides

Why scores alone do not find the fix

Customer data privacy, call recording consent and AI analysis of calls

In shortA voice of the customer program handles regulated customer data: recordings, account numbers and sometimes usage data. Federal law allows recording with one party's consent, but about a dozen states require everyone's, and California lawsuits are testing whether an AI vendor that can use call data is an eavesdropper. Rules differ by state and by type of utility, so check with your own counsel.

Vulnerable and low-income customers: disconnections, arrears and assistance

In shortMillions of US households fall behind on energy bills, and protection from disconnection depends on the state. Calls about payment plans, reconnection and medical needs are where a missed callback turns into a safety and regulatory issue, so a voice of the customer program should find them first.

The first federal count of disconnections

U.S. Energy Information Administration, 2024 Residential Utility Disconnections Report, released April 14, 2026.30 Not every state requires utilities to report disconnections.58

What it means for listening to customers

How utilities run voice of the customer programs today, and who is involved

In shortMost utilities piece the voice of the customer together from four to six systems, each owned by a different team. The questions that matter are how much gets read, whether each problem is named once across channels, and whether it reaches someone who can fix it.

Where it usually breaks

AI is arriving slowly. E Source says utilities' AI journey "continues to be slow, with early applications focused really on contact center operations, such as automated QA, automated note-taking, and some back office billing", and that its utility clients ranked AI their top investment area for customer operations.67 At least one billing system vendor has announced call summarization and tagging for its customer platform.68

A sampled read and a full read, side by side

Whichever way you work, people decide the fix, and the customer's own rating stays on the record.

Who is involved, and what each cares about

Follow the estimated-bill problem from fifty customer contacts to an accepted fix

In shortThis is the example from the top of the page, inside the product. Agents read every contact, name the problem, gather the evidence and draft a fix. Sara Mendes, the insights analyst, checks the evidence. Farah Siddiqui, who runs billing operations, decides.

1

Every contact read, none sampled

Between October 1 and 12, Lumina's customers got in touch 50 times: 12 inbound calls, 3 outbound feedback calls, 10 emails, 10 chats with the virtual assistant and 15 votes in the app and on the website. New contacts are read as they arrive, every 10 minutes; the overnight run on Tuesday, October 13 folds them all again in 4 minutes 12 seconds.

Seven are set aside, each with a reason: a silent call, an automatic reply, a delivery-failure notice, keyboard noise in a chat, and three votes with no real remark. The other 43 hold 41 problems, folded into 13 named challenges.

2

One cell, one named problem

The heatmap sets six services against seven service-quality lenses. The darkest cell is Electricity × Timeliness: 9 problems in 2 challenges, new connections not energized and reconnections waiting for the overnight batch.

Sara clicks Billing & Payments × Information accuracy: 4 problems, 1 challenge, 4 contacts by email, call, chat and app vote. The challenge has a plain name, Customer meter readings ignored, bills stay on estimates, and a three-word key: Mobile App · meter reading · not used.

3

The call that proves it

Helen Marsh's call of October 6 is 1 minute 52 seconds. At 0:44 the representative reads out her readings of July 28, August 28 and September 28: "They show as received, but not used." They arrived after the reading window, which closes five days before the bill run.

Her September bill was $412.80 on an estimate of 1,940 kWh; her reading puts it near 1,415. A re-bill is raised and a callback promised by Thursday, October 8. None is logged by October 12, so the same contact also joins the challenge Promised callbacks never made. Her app vote seven minutes after the call is joined to it as one voice.

4

A fix with an owner who decides

At 9:24 on October 6, after the challenge reached its second voice, the recommendation was written: accept app and portal readings up to 1 day before the bill run instead of 5; re-bill where a valid reading was set aside in the last 3 cycles; show in the app which bill a reading will be used on. Readings far from the estimate are also rejected outright today, so the last step replaces that outright rejection with a ±40% check against the same month last year.

It names Billing Ops, with Metering Ops, as the owner, a high risk, a timeline within 3 months, and a measure: 95% of valid customer readings used on the next bill, a target tracked monthly. Sara sends it with the four voices and the recording. Farah accepts it, due January 15, 2027.

The rest of the work: challenges, voices, ratings, recommendations, customers and settings

In shortOne problem is the story. These screens show the whole desk: every named challenge, every contact with the customer's rating beside the AI's reading, the good ratings that hide a failure, every recommendation by owner, the customers who need attention, and the utility's own settings.

Select any screen to open it full size. Lumina Utilities, its districts, staff and customers are fictional; figures in the screens are sample data.

Seven agents read, name and draft. People decide the fix.

In shortEach agent has one job and one thing it never does. None of them drops a voice silently, overwrites a customer's rating, assigns work or contacts a customer. The analyst confirms merges and corrections, the service owner accepts or declines each fix, and the head of customer experience signs off the monthly pack.

The seven agents

People decide; everything is cited

Every challenge cites its voices, and every voice keeps its transcript, recording and the words each agent relied on. Who accepted what, and when, and every agent step with its time, are kept on the audit trail.

The people in the screens

Customer contacts in, named problems with owners out

In shortIt reads the contacts customers already make, changes nothing in the systems they come from, and leaves behind named challenges with their evidence, recommendations with owners and measures, and the customers who need attention.

Systems to plan for

For each, check whether it has an API, a file feed or only screens, and who owns it. Start with the channels that carry the most volume and the most complaints.

AI customer feedback analysis that works alongside your billing system, contact center and survey platform

In shortKeep your billing system, your telephony and speech analytics, your survey platform, your outage system and your CRM. The new piece is reading across all of them as one: every contact, named once, with its evidence, and a fix that a named owner accepts.

Three principles

What unread feedback costs a utility, before you buy anything

In shortStart from your own costs. No regulator or study publishes what a utility spends on its voice of the customer program, or what a repeat call costs nationally, so the numbers that matter are yours. Once you trust them, you can judge any solution against them, ours included.

The money a regulator takes depends on your state and your rate plan; many states publish complaint counts without tying money to them (section 04). In New York, the penalties are set in basis points of the equity portion of rate base, and a repeated satisfaction miss can double them for some utilities.7

Find your own numbers

Costs no calculator counts

What automating voice of customer analytics could return for you

In shortPut in your own numbers. There is no published benchmark for any input here, so every default is marked illustrative: an assumption, not a benchmark or a promise. Replace each one with yours.

The value levers

The calculator

Are you ready to automate voice of customer analytics? Nine questions, five minutes

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

What your score means

Eighteen questions to ask before you buy voice of customer or contact-center analytics

In shortAsk every vendor the same questions, us included. Several vendors now say they analyze every interaction, so whether a tool reads everything is no longer the question. What matters is what it produces, how each finding cites its evidence, who decides, and what happens to your customers' data.

How to start: five steps to a decision you can defend

In shortMeasure first, then test on your own customers' contacts, then decide with numbers. None of these steps commits you to a vendor.

1. Count what you have: contacts a month by channel, what share is read today, and who owns each system.
2. Pull last year's numbers: repeat contacts, cost per contact, commission complaints, estimated bills and re-bills, and your rate-plan targets.
3. Gather one month of contacts from at least three channels, with privacy's approval to share them, and the problems your analysts found in the same month.
4. Run a pilot. Compare what is named with what your analysts found, check a sample of findings against the recordings, and have your service owners decide every fix.
5. Decide with three numbers: problems found that the sample missed, findings that did not hold up against the evidence, and days from a problem's second voice to an owner's decision.

Voice of customer analytics for utilities: frequently asked questions