Customer Retention Cost: The Formula and What Goes In

Customer retention cost is what your company spends to keep the customers it already has, divided by how many customers that is. Every published version of the formula agrees on that. None of them agrees on which people go in the numerator, which customers go in the denominator, or what to do about the fact that most of the people you would count are also paid to grow the accounts they are retaining.

That last point is the one that decides the number, and it is missing from the field. We read 4,090 live post-sale job adverts in full, at 1,738 B2B software companies, and counted what each role is hired to do. In 1,761 customer success adverts read in full, 60.5% name both keeping existing revenue and growing it. Only 16.0% ask for retention without growth. The salary you are about to divide by your customer count is buying two different outcomes, and only one of them is retention.

So this page does the part the formula skips. It sets out the two denominators and what each one answers, the four arguments about who belongs in the numerator, the allocation step that almost nobody performs, a worked example with the arithmetic shown, and an honest answer on benchmarks, which is that there is not one you can use.

If you want a single sentence to take away: the formula is arithmetic, the scope is the analysis, and a retention cost published without its assumptions cannot be read by anyone, including you in six months.

What customer retention cost is, and the formula everyone publishes

Customer retention cost, often shortened to CRC, is the total amount spent over a period on keeping existing customers, expressed per customer. The standard formula is total retention spend divided by a customer count.

Written out:

Both halves hide a decision. The numerator asks you to decide which costs are retention costs, which is a question about how your post-sale team is organised. The denominator asks you to decide which customers to count, which is a question about what you want the number to tell you.

The metric exists because subscription businesses spend real money after the sale and rarely know how much. Sales and marketing spend is tracked to the dollar because it is budgeted as a growth investment. Post-sale spend sits inside salaries and tool subscriptions across two or three teams and is almost never pulled into one line. Retention cost is the line.

It is worth being precise about what it is not. It is not your churn cost, which is the revenue you lost rather than the money you spent, and you can size that separately with the cost of churn calculator. It is not cost to serve per account, which asks what one account consumes rather than what the whole retention effort costs. It is not customer acquisition cost run backwards, for reasons set out further down.

One more thing it is not. A discount given to save a renewal is not a retention cost. It is revenue you chose not to collect, and putting it in the numerator double counts the damage, because the lost revenue already shows up in your retention rate. Treat concessions as a separate line and decide them on their own terms.

The denominator the published formulas disagree on

This is the first real fork, and the published pages split on it without saying so.

Some divide by active customers, meaning everyone on the books during the period. Some divide by customers retained, meaning everyone who was there at the start and was still there at the end. On the same spend these produce different numbers, and the gap is exactly your churn rate.

In the worked example below, the same $229,788 of spend reads as $1,641 per active customer or $1,824 per retained customer. That is an 11.1% difference created by nothing except which customer count you chose, and in a year with heavier churn the gap widens.

Neither is wrong. They answer different questions. Dividing by active customers answers what it costs to run retention across the book, which is the budgeting question, and it is the one to use if you are deciding headcount or comparing quarters. Dividing by customers retained answers what each save cost you, which is the efficiency question, and it is the one to use if you are arguing that the retention effort pays for itself. If you use it that way, be aware that it flatters you in a bad year, because losing customers shrinks the denominator and makes each retained customer look more expensive, which is the opposite of the signal you want.

Pick one, write it beside the number, and keep it for at least four quarters. Switching denominators mid-year produces a trend line that means nothing, and it is the single easiest way to mislead a board without intending to. The same discipline applies to the retention rate you report alongside it, where retention rate and churn rate are routinely treated as mirror images when they are not.

Who belongs in the numerator

Four arguments come up every time, and the published guidance answers all four by assertion. Here is what the hiring data says instead.

Support. Include it. The argument against is that support is reactive and serves everyone, so it is an operating cost rather than a retention cost. The argument for is that support exists only because you have existing customers, which makes it retention spend by definition. The data settles it in a different way: support ticket work is named in about one customer success advert in ten, so at most companies support is a separate function sitting on its own payroll line. That makes it easy to include as its own entry rather than impossible to separate.

Account management. Include it if it owns renewals, and at most companies it does. Our side-by-side reading of the two functions found renewals named in a higher share of account management adverts than customer success ones, with the real dividing line sitting in commercial authority rather than in who touches the renewal, and that comparison is worked through in customer success versus account management. Leaving account management out because the title does not say customer success is the most common way a retention cost comes out too low.

Onboarding and implementation. Split it. First onboarding, the kind sold with the original contract, belongs with acquisition, because the customer is paying for it and it happens before there is anything to retain. Later onboarding, when an existing customer adopts a second module or a new team, is retention spend. If you cannot split it, say which way you went.

Customer marketing and programmes. Include the part aimed at existing customers, which is webinars, customer events, lifecycle email and the QBR material. Exclude demand generation. The test is whether a prospect could receive it.

The allocation nobody does, and why it decides the number

Here is the step that separates a retention cost you can defend from one you cannot.

Once you have decided that your two customer success managers belong in the numerator, the published formulas have you add their full loaded cost and move on. Look at Figure 1 again. Across 1,761 customer success adverts read in full, 60.5% name both keeping existing revenue and growing it, and of the adverts naming any retention duty at all, 79.1% also name a growth duty. The job as written by the people hiring for it is both, and only one of the two is retention.

Asking for growth work is not the same as making somebody accountable for a growth number. A quota is named in roughly one customer success advert in forty. That is exactly why this is an allocation problem rather than a title problem: the work is mixed even at companies where the accountability is not.

If you put the whole salary in the numerator, you are charging the cost of your expansion effort to retention. Your retention cost comes out high, your expansion looks free, and any comparison with acquisition cost is meaningless because expansion revenue was acquired by someone whose salary you filed under retention.

The fix is an allocation, and it does not need to be elaborate. Ask the people doing the work to split a normal fortnight across three buckets: keeping existing revenue, growing existing revenue, and onboarding. Two weeks of honest self-reporting beats a year of assumption. Use the result as a percentage and apply it to the loaded cost. Write the percentage next to the number.

Expect the split to differ by segment. A customer success manager on twenty enterprise accounts usually spends more time on growth than one holding a hundred and fifty, which is why the allocation should be done per coverage model rather than once for the whole team. If you have not set your coverage model yet, the accounts per CSM calculator gives you the band to start from.

In the worked example, moving this one assumption from 40% to 80% changes the final number by 65.7%. No other input in the model moves it that far, including the salaries themselves.

Scope moves the answer more than the formula does

The scope decision is not a rounding error. It is most of the answer.

Across 4,489 live post-sale adverts at 1,927 B2B software companies, customer success roles are 43.7% of post-sale hiring. Add onboarding and implementation and you are at 69.1%. Add account management and you are at 93.3%. Add technical account managers and customer success operations and you have the whole post-sale organisation.

So two companies with identical spend, identical customer counts and identical formulas can publish retention costs that differ by more than a factor of two, purely because one of them counts account management and the other does not. This is why there is no useful external benchmark, and it is why a retention cost is only ever comparable against your own previous number.

Use this table to make the call once and record it. The right-hand column is the part most teams skip.

The last row is the one that causes arguments. A concession is a reduction in what you collect, so it belongs in your retention rate rather than your retention cost, and the decision to discount to save a renewal deserves to be made on its own evidence rather than buried in a cost line.

What one person costs, and where to get that number

The largest line in the numerator is people, and it is the one most often taken from the wrong place.

Three mistakes recur. The first is using base salary rather than loaded cost, which leaves out employer taxes, benefits, equipment, software and allocated overhead. The second is using a national salary average instead of your own payroll, which imports somebody else's seniority mix and somebody else's city. The third is mixing advertised figures that are labelled as base with figures labelled as on-target earnings, which quietly inflates the base by a fifth or more.

We are not going to restate the loaded cost arithmetic here, because cost to serve per account already works it through on this site, with the advertised base figures, the 1.25 to 1.4 loading multiple and what each assumption does to the result. The worked example below uses its figure of about $142,000 for one loaded customer success manager, so the two pages agree rather than compete.

Use your own payroll if you have it. Use the published figure only as a sanity check, and only if you are willing to publish the multiple you applied. An organisation that cannot produce a loaded cost per head has a finance problem rather than a metrics problem, and the retention cost is not the place to solve it.

The second largest line is usually tools, and it is the easiest to get exactly right because most of it is invoiced. Pull the annual contract value of the customer success platform, the help desk, the survey tool and anything bought specifically for post-sale work. If you are sizing that line before you buy, customer success software under $10,000 a year sets out what the category actually costs at the small end.

A worked example, with the arithmetic shown

A B2B SaaS company with $2.4M ARR across 140 customers, so about $17,100 of ARR per customer. It kept 126 of them through the year. The post-sale team is two customer success managers, one onboarding specialist and half a support head. There is no separate account management function.

The numerator, base case. Two customer success managers at about $142,000 loaded is $284,000. The team's own time study put 60% of their fortnight on keeping existing revenue, so $170,400 goes in. Support is half a head at $90,000 loaded, counted in full, so $45,000. Tools are the customer success platform at $2,388 a year, the help desk at $4,800 and the survey tool at $1,200, so $8,388. Customer programmes, which is two webinars and the QBR material, are $6,000.

That totals $229,788.

Onboarding is excluded in the base case, because this company sells implementation with the first contract and its onboarding specialist works almost entirely on new logos.

The denominators. Across 140 active customers, retention cost is $1,641 per customer. Across the 126 retained, it is $1,824.

As a ratio. $229,788 against $2.4M of ARR is 9.57%, which is the number a CFO will ask for, because it is the only version that is comparable to anything.

What moves it. At a 40% allocation the ratio falls to 7.21%. At 80% it rises to 11.94%. Counting the onboarding specialist in full takes the base case to $349,788 and 14.57%, which is 52.2% higher than the base case from one scope decision, taken without touching a single salary.

The ratio finance reads: retention cost against ARR

Dollars per customer is the version the field publishes and the version finance cannot use.

The reason is that it moves when your average contract value moves, which means it moves when you change pricing, when you win a larger segment, and when a big customer leaves. A retention cost of $1,641 per customer means one thing at $17,100 of ARR per customer and something else entirely at $4,000. Two of your own quarters are not comparable if your mix shifted between them.

Retention cost as a share of ARR fixes that. It asks what proportion of the revenue you are protecting is spent on protecting it, and it survives a change in contract value. It is also the form in which the number can sit next to the other ratios a board already reads, such as sales and marketing as a percentage of revenue.

Track it quarterly, annualised. Read it against two things. The first is your own previous quarters, because the trend carries almost all of the signal. The second is your gross retention, because retention cost only means something in the company of the retention it bought. A ratio falling while gross retention holds is the result you want. A ratio falling while gross retention falls is a cut, not an efficiency. A ratio rising while net revenue retention rises may be money well spent, as long as you have allocated properly and are not reading your expansion investment as a retention cost.

Somebody has to own the calculation and run it the same way every quarter. In most teams that is customer success operations, and where there is no ops person the number tends to be rebuilt from scratch each time by whoever is asked, which is how the assumptions drift without anyone deciding to change them.

Retention cost against acquisition cost, and why the comparison misleads

Every page on this subject reaches for the same comparison: it costs far more to acquire a customer than to keep one. The comparison is fair in direction and misleading in practice, for three reasons.

The first is that the two metrics have different shapes. Acquisition cost is a one-off charge against a customer who has just arrived. Retention cost is a recurring charge against every customer every year. Suppose acquisition costs $14,000 and retention costs $1,641. The naive ratio reads as about eight and a half to one, and it is wrong, because the retention cost repeats for as long as the customer stays. Over five years it is $8,205 against $14,000, which is a different argument.

The second is the allocation problem again. If your retention cost includes the salaries of people who generated expansion revenue, you have charged growth to retention and the comparison is not measuring what it claims.

The third is that the most-cited number in this field comes with conditions nobody repeats. The claim that a 5% improvement in retention lifts profits by 25% to 85% comes from Reichheld and Sasser's 1990 Harvard Business Review article "Zero Defections: Quality Comes to Services". Its own worked cases were 85% for a bank branch system, 50% for an insurance brokerage and 30% for an auto-service chain, so the top of the range is one example rather than the typical result. A later restatement widened the range to 95%. The deeper objection, raised by critics since, is that the analysis treats the improvement in retention as free. It models the profit from lower defection without modelling what it costs to defect less, which is precisely the number this page exists to produce.

Use the comparison to make the directional case, which is sound. Do not use it as arithmetic in a budget.

Is there a benchmark? Not one you can use

No. There is no reliable published benchmark for customer retention cost, and the figures circulating are not consistent enough to be used.

The ones in circulation include keeping it under 25% to 30% of lifetime value, keeping it under 15% to 20% of monthly recurring revenue per account, and treating anything above 10% of total revenue as worth managing. One widely read page carries two of these in the same article, advising that under 25% to 30% of lifetime value is reasonable and, a few paragraphs later, that spending up to 70% of lifetime value is often sustainable. Those cannot both be guidance.

There are three structural reasons an external benchmark cannot work here, and they are the same three reasons the number is useful internally. Scope varies, as Figure 2 shows, by more than a factor of two. Allocation varies, and almost nobody states it. The denominator varies between active and retained customers. A benchmark requires all three to be standardised, and none of them is.

What you can do instead is build your own baseline and treat the first four quarters as the benchmark. Record the scope, the allocation and the denominator alongside each quarter's figure so that a successor can reproduce it. Then judge movement rather than level, and judge it against gross retention in the same period.

If somebody brings you an industry figure, ask three questions before you use it: who is counted, what share of their time is allocated to retention, and which customers are in the denominator. In our experience of reading the published pages, none of them answers all three, which tells you what the figure is worth.

Does retention cost fall over a customer's lifetime?

Usually, for a while, and then it flattens. The pattern is real but it is weaker than the field suggests, and it is worth knowing which part is genuine.

The decline is real in the first eighteen months. A customer in month three consumes onboarding attention, configuration help and a higher support volume. The same customer in month twenty is usually self-sufficient on the core workflow and the cost of keeping them falls accordingly.

After that it flattens, and three things push it back up. Renewal events recur, and the work around a renewal does not get cheaper with age. Champions leave, and a new buyer inside an existing account needs re-earning from close to zero. Larger accounts expand into new teams, which looks like growth and is paid for out of post-sale time.

The practical consequence is that lifetime retention cost, calculated by multiplying this year's figure by the expected number of years, will overstate the cost of a young customer and understate the cost of an old enterprise one. If you need a lifetime figure, build it from cohorts rather than from a multiplier. If you do not need it, do not build it, because the per-period ratio answers almost every question that a lifetime figure is reached for.

The cheapest saving is not doing retention work on accounts that were going to stay anyway.

How to bring it down without moving the cost somewhere else

Four moves work, and two popular ones do not.

Raise the coverage ratio where the data supports it. Moving low-ARR accounts from named coverage to a pooled or digital model is the only lever with real magnitude, because people are most of the numerator. Do it on evidence of which accounts actually need a human, not on ARR alone.

Cut the reactive work rather than the proactive work. Retention cost is inflated by rework: failed handoffs, unanswered risk, escalations that arrive late. Every hour spent on a crisis that early warning would have caught is a pure addition to the numerator with no corresponding retention.

Price the tools against the people they replace. A platform that costs a few thousand a year and removes a few hours a week from two customer success managers pays for itself inside the numerator. One that requires an administrator does not, because the administrator is also a salary in the same line.

Fix the allocation before you cut anything. Many teams discover their retention cost is high because they were charging expansion work to retention, which is a measurement fix rather than a spending one.

What does not work: cutting the tool line while keeping the headcount, because tools are usually under 5% of the numerator, as they are in the worked example at $8,388 of $229,788. And cutting support headcount, which moves cost into churn within two quarters and shows up as a worse retention rate against a better retention cost.

When a high retention cost is the right answer

Three situations where a high number is correct and cutting it would be a mistake.

You are in the first two years of a complex enterprise product, where retention cost per customer is high because every implementation is bespoke and the alternative to spending the money is losing the logo. The right response is a plan to bring it down as the product matures, not a cut now.

Your segment mix is weighted to accounts whose lifetime value justifies the spend. A $1,824 cost to retain a customer worth $17,100 a year is a reasonable trade. The ratio to watch is retention cost against the revenue it protects, not the absolute figure.

You are rebuilding after a bad year. Retention cost rises during a recovery because the work is remedial, and it should fall once the underlying cause is fixed. A board that understands this will fund it. A board that is shown only the cost line will not, which is an argument for reporting the ratio and the retention rate together rather than either alone.

The case where a high number is genuinely bad is the quiet one: retention cost rising, gross retention flat, and nobody able to say what the extra money bought. That is the state the measurement exists to catch.

The number is only worth what the assumptions beside it are worth

Most teams reach for customer retention cost expecting a formula problem and find a scope problem. The arithmetic takes ten minutes. Deciding who counts, what share of their time is retention and which customers go in the denominator takes a conversation with finance and a fortnight of honest time tracking, and that work is the entire reason the number is credible when you put it in front of a board.

Do it once properly, write the three decisions down next to the figure, and repeat them without changing anything for four quarters. What you will have then is the only version of this metric that is worth anything: a trend you can read, against a retention rate you can trust, with assumptions somebody else can reproduce.

How this was researched

The role and duty figures come from every live post-sale advert on the GainTrace customer success jobs board, pulled on 12 November 2026 from employers' own Greenhouse, Ashby and Lever boards: 4,489 roles at 1,927 B2B software companies, with full advert text read for 4,090 of them at 1,738 companies. The duty figures use the full-text subset; the scope figures use all 4,489. Benefits, perks and equal-opportunity sections were stripped before counting, because those blocks carry vocabulary that inflates every soft count.

Duty figures are mention rates. They say what an advert asks for, not how an hour is spent, and a role that does not name a duty may still perform it. Shares of hiring are a proxy for how post-sale work is divided, not a census of who is employed. The board is live, so the counts will move; the proportions in the prose are what to carry forward, and the dated figures are on the faces of the figures.

The worked example is illustrative. Every figure in it is stated in the text so that you can substitute your own, and the arithmetic reconciles end to end. The loaded cost per customer success manager is taken from our cost to serve page rather than recomputed, so the two agree.