Customer success metrics you can act on

Sales has a clear end point. The deal is won or lost. Customer success does not. An account can look happy for two years and then cancel in a single call. Or it stays, never buys anything extra and uses half of what it pays for. On a revenue sheet you only notice when it is too late.

So you need two kinds of numbers.

Outcome metrics tell you what happened. How much revenue you kept, how many customers left, how much they expanded. These numbers are hard and leadership asks for them. But you only see them afterwards, usually per quarter or per year.

Signal metrics tell you what is likely to happen. How fast an account goes live, how often it opens a ticket, how satisfied it is with the handling, whether its team actually uses the product. These numbers are softer. But you can still do something about them.

A common mistake: a team only picks outcome metrics because leadership asks for them. Then every quarter you discuss accounts that are already gone. The opposite happens too. A team only tracks signals, such as satisfaction and usage, and loses touch with revenue. A good set has a few of each. Six to eight numbers in total is enough.

A second point: measure per account, not only per team. An average satisfaction of 8 says nothing about that one large account scoring 4. In B2B, revenue often depends on a handful of accounts. You want to know those by name, with their numbers next to them.

A third point: agree on definitions before you calculate. Does a customer moving from an annual to a monthly contract count as churn? Does a price increase count as expansion? Does time to value start at signature or at kickoff? Without an agreement you get the same discussion every quarter, and two teams with two different retention figures.

All numbers in the examples below are made up. They show the math, they are not benchmarks. What good retention or satisfaction looks like depends on your market, contract model and customer size. Compare yourself mostly with your own previous quarter.

Say a software company starts the quarter with 80 customers and $100,000 in monthly recurring revenue. Every number here is made up to show the math.

During the quarter:

• 4 customers cancel, together worth $6,000 a month.
• 3 customers downgrade to fewer users, together $2,000 a month less.
• 9 customers expand, together $11,000 a month more.
• Sales signs 7 new customers worth $9,000 a month.

GRR is starting revenue minus cancellations minus downgrades, divided by starting revenue. That is 100,000 minus 6,000 minus 2,000, which is 92,000, divided by 100,000. GRR is 92 percent.

NRR adds the expansion. 92,000 plus 11,000 is 103,000, divided by 100,000. NRR is 103 percent.

Logo churn is 4 divided by 80, so 5 percent. Revenue churn is 8,000 divided by 100,000, so 8 percent.

The new customers from sales do not count in any of these numbers. That is on purpose. You want to know how well you keep and grow the customers you already have. Count new revenue and a strong sales quarter hides a leak in customer success.

What does this tell you? An NRR above 100 percent looks fine. But a GRR of 92 percent says eight percent walks out every quarter, and expansion only just makes up for it. If expansion is weak for one quarter, your existing revenue shrinks. So the question for the next review is not how to sell more. It is why those four customers left and whether the three downgrades form a pattern.

Two things to agree upfront. Whether you calculate per month, per quarter or per year, and stick to it. And use a cohort: the customers who were already customers at the start of the period. Someone who joined halfway and left again does not belong in this number. That account tells you something about onboarding.

Outcome metrics tell the story afterwards. Signal metrics give you time. Four of them are the most useful in B2B.

Time to value and onboarding

The first weeks matter a lot. A customer who sees value quickly builds the product into its work. A customer still configuring after three months starts to doubt at the first invoice. So track, per account, the go-live date against the planned date and the number of completed milestones. A two-week delay is not a problem yet. A delay without a next step is.

Ticket pattern

Many tickets do not automatically mean an unhappy customer. A customer asking questions is using the product. Watch the change. An account that went from two tickets a month to eight has something going on. An account that used to ask questions and has gone quiet might too. Tickets that stay open long or breach their SLA are a stronger signal than the count. The guide on setting up a helpdesk shows how to make that handling measurable.

CSAT and NPS, and the difference

You ask CSAT shortly after a contact moment. It tells you whether that ticket went well. You ask NPS apart from any moment, for example twice a year. It says something about the relationship. Do not mix them up. And read the answers, not only the number. A 6 with the comment "your support is good but the product lacks reports" is an expansion opportunity or a churn risk, depending on what you do with it.

Be careful with NPS on small numbers. With twenty answers, one person moves the score by five points. Then do not compare quarters. Read the answers per account instead. And watch who answers. If only your main contact gets the question, you learn nothing about the users.

Adoption

If a customer pays for 25 users and 11 are active, chances are it downgrades at renewal. Or worse, the decision maker never sees the product work. Track adoption per account and per module. Also read how to increase adoption, which has a week by week plan.

A health score sums up the signals per account in one number from 0 to 100. The goal is not a perfect prediction. The goal is that every week you know which customers to call. Below is a template you can copy and adjust. The weights are a starting point, not the truth.

Signals and weights

• Adoption, 30 points. Active users divided by paid users. 80 percent or more gives 30 points, 50 to 80 percent gives 15, less gives 0.
• Onboarding or project status, 20 points. Live and on schedule gives 20. Behind schedule with a next step gives 10. Behind with no next step, or a blocker open longer than two weeks, gives 0.
• Tickets, 20 points. No SLA breaches in the last 90 days and a steady volume gives 20. A rise or one breach gives 10. Several breaches or an escalation gives 0.
• Satisfaction, 15 points. Average CSAT over the last 90 days positive and no detractor in the last NPS gives 15. Mixed gives 8. Negative gives 0.
• Relationship, 15 points. An active sponsor you spoke to last quarter gives 15. Contact only at working level gives 8. Sponsor gone or unknown gives 0.

Colours and thresholds

• Green, 70 to 100. Normal rhythm. Look for expansion.
• Amber, 40 to 69. A conversation with a concrete goal within two weeks.
• Red, 0 to 39. Action within a week, with the manager involved.

Cadence

Recalculate the score daily or weekly, but review it weekly. Every quarter, check whether the score was right. Take the customers who cancelled in the past year and look at their score three months before. If they were green, you are measuring the wrong things or one signal weighs too much.

Worked example

Say an account pays for 20 users and 13 are active, which is 65 percent: 15 points. The project is live and on schedule: 20 points. There was one SLA breach last month: 10 points. CSAT positive, no NPS answer: 15 points. The sponsor left three months ago: 0 points. Total 60, amber. Notice that satisfaction is not what pulls the score down. Adoption and the relationship are. So the conversation is about who the new sponsor is and why seven people do not log in. This example is made up.

Counting new revenue in retention. Once revenue from new customers ends up in NRR, the number says nothing about customer success anymore. Keep the cohorts strictly apart.

One NPS a year as the main metric. One measurement a year is too slow to steer on, and the score depends a lot on who happened to answer that week. Use NPS as a signal per account, not as a team target.

A health score nobody understands. A model with fifteen signals and decimal weights feels precise. But if a CSM cannot say in one sentence why an account is amber, nobody acts on it. Start with five signals.

A score without a playbook. Red without an agreed action means everyone looks at the same list and nobody calls. Tie each colour to an action with a deadline and an owner.

Averages across all customers. An NRR of 103 percent can be three large accounts that expanded a lot and twenty small ones that downgraded. Split your numbers by segment, for example by customer size or by product.

Reading ticket volume as unhappiness. An account without tickets can be very happy or no longer using the product. Watch the change and the handling, not only the count.

Definitions that shift per quarter. If a downgrade counts as churn this quarter and not last quarter, you compare apples with oranges. Write the definitions down and only change them with a note.