How to measure CSAT, and how it differs from NPS and CES

The CSAT question and scale

A common wording is "How satisfied were you with your experience today?" SurveyMonkey describes the typical scale as 1 (very unsatisfied) to 5 (very satisfied), with 3 as "neither". The top two answers count as satisfied (SurveyMonkey). Some teams use 1 to 7, 1 to 10, emoji or a plain good/bad choice. Any of them works. What matters is that you pick one and keep it, because a score on one scale can't be compared with a score on another.

Ask about the contact that just happened, not the company in general. That's what makes CSAT useful for judging an agent or a team, and it's the main difference from NPS.

How to calculate CSAT

CSAT = satisfied responses ÷ all responses × 100. Neutral answers stay in the total, so a 3 pulls the score down. Here is a made-up month of 200 survey responses for one support team:

Scroll sideways to see the whole table.

Satisfied responses are 90 + 50 = 140. CSAT is 140 ÷ 200 × 100 = 70%.

Now the catch. Some tools work it out differently. Zendesk's agent satisfaction report divides tickets rated good by all rated tickets (Zendesk, agent scores). Its Surveys dashboard divides the sum of all responses by the maximum possible score (Zendesk, Surveys dashboard). Run the same 200 responses through the second method: (90×5 + 50×4 + 25×3 + 15×2 + 20×1) ÷ (200×5) = 775 ÷ 1,000 = 77.5%.

Same customers, same answers, 70% or 77.5%. Before you compare your score with anyone else's, find out which method produced it. Write the method next to the number in every report.

When to send the survey

Send it once per contact, soon after the ticket is solved or the call ends, while the customer still remembers it. On chat and messaging that can be in the same window, and on phone it's usually a text or an email afterwards. Long gaps mean customers rate something they've half forgotten.

Keep the scores for each channel apart. Phone, email and chat get different customers with different problems, and a blended number hides a weak channel behind a strong one. Tag each response with:

• The channel: phone, email, chat or messaging.
• The contact reason, from the tag the agent sets when closing the ticket.
• The agent and the team who handled it, so an outsourced team can be split out from your own.
• Whether the contact was a repeat about the same issue.

Why CSAT can mislead

Most customers don't answer. The ones who do tend to be either pleased or annoyed, so the middle is under-counted. That's fine for spotting trends in a team, and poor for ranking one agent against another on a handful of replies.

The customer also rates everything at once. A polite, correct agent who has to say no to a refund will often get a 1. If you read that as the agent's fault, you'll coach the wrong person. And if the survey only goes out on solved tickets, contacts that were never solved don't get a vote at all.

None of this makes CSAT useless. It means you report the response count and response rate with it, and you read the comments before you act on the number.

CSAT vs NPS vs CES

Fred Reichheld of Bain & Company created NPS in 2003 (Net Promoter System), introduced in his Harvard Business Review article "The One Number You Need to Grow" (HBR). Customers answer on 0 to 10. 9s and 10s are promoters, 7s and 8s are passives, and 0 to 6 are detractors. The score is the percentage of promoters minus the percentage of detractors (Net Promoter System, measuring).

CES (customer effort score) came from Matthew Dixon, Karen Freeman and Nicholas Toman in HBR in July 2010. They studied more than 75,000 people who contacted service reps or used self-service. Customers mainly wanted a quick, simple fix, and the authors argued effort predicts loyalty better than satisfaction or NPS (HBR).

Scroll sideways to see the whole table.

For a support team, CSAT is the one to report weekly. NPS moves with price and product as much as with service, so it's a poor measure of one team. CES is worth adding if repeat contacts and transfers are your problem.

What to do with the scores

Pull every 1 and 2 each week and score those tickets on your quality scorecard. The scorecard tells you whether the agent got it wrong or the customer was unhappy with something the agent couldn't change. Coach on the first. Take the second to whoever owns the policy or the product.

Then score a sample of the 4s and 5s too. Customers sometimes rate a wrong answer highly because it was friendly, and that wrong answer comes back later as a repeat contact. If you don't have a scorecard yet, our free QA rubric builder gives you a first draft.

Reporting CSAT for an outsourced team

If you outsource some of your support, you want to know whether the outsourced team is as good as yours. That only works if the two numbers are built the same way:

• The same survey, sent by the same tool, on the same trigger, for both teams.
• The same calculation. Pick top-two-box or the average and never mix them across teams.
• The same contact types. Compare email about orders with email about orders, not a whole queue with a pilot queue.
• The number of responses and the response rate next to every score, so a 95% from 20 replies is read as 20 replies.
• Low scores split by cause: the agent, the process or the policy.

The simplest way to get all of this is for the outsourced agents to work in your own help desk under named logins, which is how our teams work. Their tickets then trigger your survey and land in your reports, tagged by agent, with no separate provider report to reconcile. We go through the numbers with you in a weekly review, and how we work sets out the rest of the method. For the pilot and shared QA scoring that come before any of this, read how to outsource customer service without losing quality.

Questions

What is a good CSAT score?

There's no single answer that holds across scales, calculation methods and industries, and the example above shows one set of answers giving 70% or 77.5%. Compare your score with your own last quarter, and your outsourced team with your in-house team, on the same survey. That comparison means something.

Is CSAT or NPS better?

They answer different questions. CSAT tells you how a contact went. NPS tells you how customers feel about the company. Use CSAT to manage a support team and NPS, if at all, for the business as a whole.

How many responses do I need before I trust an agent's CSAT?

More than most teams have in a week. Look at agent scores over a month or more, always with the response count beside them, and use QA scoring for the week-to-week coaching.

Next step

If you're about to outsource part of your support and want CSAT you can compare from day one, book a call with me. Bring your current survey settings and a month of scores. We'll check how they're calculated and what a fair comparison would look like.