Back to blog
A woman with long dark hair, wearing a headset with a microphone, sits at a desk beside a monitor and a glass of water and smiles at the camera, while colleagues in headsets work at the desks behind her
Survey designNovember 20, 202610 min read

CSAT and Customer Effort Score: what to ask after a customer interaction

A support ticket closes and a one-question survey goes out. Whether that question asks about satisfaction or effort, and how you word, time and read it, decides whether the score tells you anything you can fix.

By SurveyLane · The team building SurveyLane

A customer contacts support, the ticket closes, and a one-question survey lands in their inbox a few minutes later. Most teams set that up once and then read the result as a weekly percentage. But that single question hides a stack of decisions: satisfaction or effort, the exact wording, the trigger, who gets it. Get them wrong and the score moves for reasons nobody can name.

Two scores that answer different questions

CSAT, customer satisfaction, asks how satisfied someone was with one specific interaction. "How satisfied were you with the help you received today?" It measures a feeling about a moment.

The Customer Effort Score asks something narrower: how hard was it to get done what you came to do? A customer can like a friendly agent and still have spent forty minutes and two transfers getting there. Ask about effort and you hear about the forty minutes.

Both are transactional measures, sent right after a touchpoint such as a support contact or a checkout. Relationship measures like the Net Promoter Score ask about the company as a whole, on a schedule. Our piece on using NPS well explains why the two are separate measurements.

What the original effort research found

The effort score comes from a 2010 Harvard Business Review article, "Stop Trying to Delight Your Customers", by Matthew Dixon, Karen Freeman and Nicholas Toman of the Corporate Executive Board, the advisory firm Gartner later bought. Their team studied more than 75,000 people who had dealt with a contact centre or with self-service channels.

89 of the 100 customer service heads they surveyed said their main strategy was to exceed expectations, yet 84 percent of customers said their expectations had not been exceeded in their latest interaction. Eighty percent of service organisations used CSAT as their primary metric, and the authors found little link between satisfaction and loyalty: 20 percent of "satisfied" customers said they intended to leave, and 28 percent of "dissatisfied" ones intended to stay.

They then tested CSAT, NPS and their new effort question on how well each predicted loyalty, defined as intending to keep buying, to spend more and to speak well of the company. CSAT was a poor predictor. NPS did better. In customer service interactions, effort did best. Of customers reporting low effort, 94 percent said they intended to buy again, 88 percent said they would spend more, and only 1 percent said they would speak negatively about the company. Of those who had a hard time, 81 percent said they planned to spread negative word of mouth.

Read those numbers for what they are

Three limits apply. The outcomes are intentions. "I intend to repurchase" is a survey answer, and people are optimistic about their own behaviour, as our article on questions about behaviour explains. Effort and stated loyalty move together, but that says nothing about how many customers you keep per point of effort removed.

The claim is also scoped to customer service interactions, and the authors themselves call NPS a good company-level gauge. And it comes from a consultancy that sold advice built on the metric. So test it on your own data: link effort scores to what customers did in the months after.

The original effort question, and why it was revised

The 2010 version asked: "How much effort did you personally have to put forth to handle your request?" Scored from 1, very low effort, to 5, very high effort.

Two problems. The scale runs the opposite way from nearly every other score a team tracks, so a falling number is good news, and people misread the dashboard. And "put forth" is stiff English that translates badly once you survey in more than one language.

In their 2013 book The Effortless Experience, Dixon, Toman and Rick DeLisi introduced a revised version, usually called CES 2.0. It's an agreement statement, commonly cited as "The company made it easy for me to handle my issue", on a seven-point scale from strongly disagree to strongly agree. Higher is now better.

That fixes the direction but brings in the weak spot of every agree-disagree item: some respondents agree with almost any statement. Our guide to designing rating scales covers that acquiescence effect. If you don't need a benchmark, ask "How easy or difficult was it to get your issue resolved?", labelled from "very difficult" to "very easy". Then keep it, because new wording breaks your trend line.

Wording a satisfaction question that measures one thing

"How satisfied were you with our support?" leaves the customer to decide what "our support" means. One person rates the agent, another the outcome, a third the twenty-minute wait. You average them into one number.

Name the object. "How satisfied are you with how your question was handled?" is about the process. "How satisfied are you with the answer you received?" is about the outcome. Pick the one you can act on.

Ask about one thing at a time. Someone who got a fast, curt reply can't honestly answer "How satisfied were you with the speed and friendliness of our service?" Our article on writing better survey questions goes deeper into double-barrelled wording.

For the scale, a five-point scale from "very dissatisfied" to "very satisfied" is the usual choice. Label every point. A bare row of numbers lets each respondent invent their own meaning for 3 and 4.

Scoring: percentages, means and what each hides

CSAT is usually reported as the share of respondents who picked one of the top two answers: on a five-point scale, the 4s and 5s divided by everyone who answered. CES gets reported as a mean, or as the share who agree, for example 5, 6 or 7 on the seven-point version.

Both lose information. A top-two-box percentage doesn't budge when a customer moves from "very dissatisfied" to "neutral", which is just the improvement a process fix tends to produce. A mean hides whether you have mild answers or a split between happy and furious customers.

So show the distribution next to the headline. A CSAT of 82 percent with 10 percent "very dissatisfied" is a different problem from 82 percent with 2 percent. Write the calculation down once, and never convert a five-point score to seven points to line up old and new data.

When to send it

Send within minutes and you catch the experience while it's fresh. A week later you get a vaguer impression.

The bigger problem is firing on the wrong event. Most help desks send the survey when a ticket closes, and a closed ticket isn't a resolved issue. Ask "How easy was it to resolve your issue?" when it wasn't resolved and you annoy exactly the customer you most need to hear from.

Trigger on the moment the customer's goal was met: the refund arrived, the order shipped. In a product, that's task completion, not a page view. Use conditional logic to skip the question where it makes no sense, for instance when the contact was forwarded elsewhere.

Who answers, and who never does

Transactional surveys often get low response rates, and customers with a strong experience in either direction reply more often than the rest. A month's score can shift because a different mix of people bothered to answer. Our article on sample quality covers that nonresponse problem.

Two habits make it worse. If agents pick which tickets get a survey, the difficult ones quietly drop out. And if you survey after every contact, your most frequent contacts, often your unhappiest customers, learn to ignore you. Survey every eligible interaction or a random sample of them, and cap how often one person gets it, say once every 30 days. Our piece on survey fatigue and contact frequency helps you set those rules.

Keeping the score honest when people are judged on it

Put a satisfaction score in an agent's targets and it starts measuring something else. Agents learn to say "a 5 would really help me". The incentive bends behaviour, and the number climbs while customers notice nothing.

An agent can't fix a confusing returns policy, so a low score caused by that policy shouldn't count against them. Review scores per team or process, since ranking individuals on a handful of responses is mostly noise. And look twice at one team with near-perfect scores and an oddly low response rate.

The follow-up question that explains the score

To learn why a score moved, add one follow-up, shown only to people who gave a low score. "What made this harder than it should have been?" for effort. "What could we have done better?" for satisfaction. Keep it optional, because forcing an answer gets you worse ones.

The original research tells you where to look. Customers most resented contacting the company more than once, being transferred, repeating information and switching channels, such as calling after the website failed them. Well over half the customers in the study had run into this. Turn those into a short multiple-choice follow-up, with an "Other" field, because your effort drivers may differ. Our guide to analysing open-ended responses shows how to code what comes back.

Pair the score with what you already log

The effort score is a perception, and much of the effort behind it already sits in your own systems. Repeat contacts on the same issue within a week. Transfers between teams. A chat that ended in a phone call. The 2010 authors suggested combining the score with measures like these in an "effort audit".

That finds effort where nobody filled in the survey, and it checks the survey itself: if customers with three contacts on one issue report the same effort as customers with one, your question isn't measuring what you think. To see which parts of an interaction move the score, a key driver analysis is the next step.

Choosing between CSAT, CES and NPS

One question per touchpoint plus an optional follow-up is plenty. Choose effort when the interaction exists to solve a problem: support contacts, returns, password resets. Choose satisfaction when the quality of the experience is the point, like a consultation. Keep NPS for a periodic relationship survey. And before the first response comes in, write down who owns the score and what happens when it drops.

In SurveyLane you can build the survey with a conditional follow-up for low scores and analyse the results through the MCP server, for example by asking which channels produce the most high-effort answers. Give the assistant your scoring rule and check its first answer by hand. Without the rule, a model asked for "the CSAT" may just average the raw answers.

Frequently asked questions

Is the Customer Effort Score better than CSAT?

Not across the board. The 2010 research found that effort predicted stated loyalty better in customer service interactions, so for problem-solving touchpoints it's usually the more useful question. Where the quality of the experience matters most, satisfaction often fits better. Test either against your own retention data.

Should I use the five-point or the seven-point effort scale?

Use the seven-point CES 2.0 agreement version if you want to compare with published benchmarks, and an item-specific easy-to-difficult question if you don't. Skip the original five-point version, where a higher number means more effort: it runs the opposite way from your other scores and gets misread.

How soon after an interaction should the survey go out?

As soon as the customer's goal has actually been met, ideally within minutes or hours. Trigger on resolution, not on ticket closure, and leave out customers whose issue is still open, because asking them costs you trust and skews the score.

Can I compare my CSAT with another company's?

Only if the wording, scale, scoring rule, timing and type of customer all match, and they rarely do. Your own score over time, measured the same way each time, tells you far more than a benchmark built on a different question.

Further reading