Key Takeaways
The CS team's ROI slide looked strong. Covered accounts renewed at 94%, and uncovered ones renewed at 81%. The difference, times covered ARR, came to several times the team's cost. Then the CFO asked a fair question. Which accounts get a CSM? The answer was the largest ones. And the largest accounts renew at higher rates whether anyone calls them or not. The slide was measuring who got coverage, not what coverage did.
Most customer success ROI guides give a formula and a list of metrics. Our own guide to measuring the ROI of enablement and support covers the formula and the baseline work. This post is about the one term in that formula that's usually wrong: the benefit. It covers what CS costs, why the usual benefit estimate is inflated, and how to measure the increment instead.
What customer success costs
Start with the denominator, because it's the part you can benchmark. SaaS Capital's 2026 survey of more than 1,000 private B2B SaaS companies found that “the median percent of annual recurring revenue spent on customer support and customer success is 9%, up from 8% the previous year.”
That figure combines support and success. For CS alone, Gainsight and Benchmarkit's 2023 index of about 400 North American companies put CS expense at 4% of revenue at the 25th percentile, 8% at the median and 13% at the 75th. The same report found “98% of companies are maintaining or increasing their investment in customer success.”
At smaller companies, the cost is close to the cost of selling. In SaaS Capital's data, companies at $3 million to $5 million in ARR spent 10% on support and success, 12% on selling and 8% on marketing. A line that size gets the same scrutiny as sales, and it should.
Why the usual benefit estimate is inflated
The standard approach credits CS with the ARR retained or expanded in the accounts it covers. That number has three problems.
Coverage follows contract size
CSMs are usually assigned by account value. Retention also rises with account value. SaaS Capital explains why: higher-priced products involve “a longer sales cycle, in-depth scoping and implementation, and dedicated support and account management, all of which yield a stickier product.” Compare covered accounts with uncovered ones and you're mostly comparing big contracts with small ones. Our post on CS segmentation covers how coverage is usually assigned.
The aggregate trend points the other way
If more CS spend reliably raised retention, the industry numbers would show it. Bain's 2024 survey found NRR “decreased for 75% of software companies,” even as nearly 60% increased customer success spending. That doesn't prove CS is useless. It does mean spend alone isn't the lever. Bain's own question for CS leaders is still open: “Which activities most affect customers’ return on software investments and subsequent purchase decisions, and how can we measure this?”
The supporting data is correlation
Vendor studies often report that CS roles go with higher NRR. ChurnZero's 2025 study of 793 leaders says the presence of CSMs and other roles “correlates with higher NRR,” and notes that “all data is self-reported.” Correlation is a fair starting point. It isn't a return.
Few teams even try to separate the effect. Benchmarkit found that “less than 20% of companies are calculating the Expansion CAC Ratio”, the metric that allocates sales, marketing and CS cost to expansion ARR. That chart had only 21 responses, so treat it as a signal. TSIA says the same thing more broadly: most CS teams “still struggle to calculate the savings or efficiencies delivered by their technology investments.”
What the controlled evidence shows
The rigorous evidence on proactive retention comes from randomized field experiments. They're mostly in consumer and membership businesses, not B2B SaaS. Read them as a reason to test your own motion, not as B2B benchmarks.
Outreach can raise churn. One 2016 field study tested a campaign that recommended cheaper, better-fitting plans. “Whereas only 6% of customers in the control condition churned during the three months following the intervention, 10% did so in the treatment group.” The contact worked “by lowering customers’ inertia to switch plans” and by making past usage more visible. A before-and-after comparison wouldn't have caught that.
Risk isn't the same as responsiveness. Two field experiments found that “customers identified as having the highest risk of churning are not necessarily the best targets for proactive churn programs.” The settings were a wireless services provider and a membership organization. The recommendation was to target by lift: how much an account's outcome changes because you acted.
Targeting by lift pays. A profit-based targeting method from a 2020 study was “significantly more profitable” in two field experiments. The authors estimate one campaign could raise overall profit by at least 4%. A widely quoted 180% figure from the same work was a theoretical re-analysis of one firm's data, not a measured result. In a separate test at the Boston Globe, a learned targeting policy “had a net-positive revenue impact in the range of $4-5 million compared to the status quo” over three years.
B2B evidence is thinner. The one peer-reviewed study of CS management we found is qualitative. It reports that “not all CS implementations are initially successful.”
The incremental customer success ROI worksheet
Here's a four-line calculation that replaces the usual benefit estimate. The benchmarks come from the sources above. The method is our recommendation.
| Line | What to calculate | Reference point |
|---|
| 1. Cost | CS spend as a share of ARR | 9% median for support and success; 4%, 8% and 13% for CS alone |
| 2. Benefit, as usually reported | Total ARR retained or expanded in covered accounts | Inflated by contract size and by retention that would have happened anyway |
| 3. Benefit, incremental | (Retention in treated accounts − retention in holdout accounts) × treated ARR | Can be negative: 6% vs 10% churn in one field test |
| 4. Targeting | Rank accounts by expected lift, then re-test each cycle | Lift targeting beat risk targeting in consumer field tests |
Sources: SaaS Capital 2026; Gainsight and Benchmarkit 2023; a 2016 field study; a 2018 field study; a 2020 targeting study.
A worked example (illustrative numbers)
Take a team running a proactive adoption review for mid-market accounts. There are 200 eligible accounts at an average of $40,000 ARR. The team randomly holds back 40 of them for one renewal cycle.
- The 160 reviewed accounts renew at 90%. The 40 held-back accounts renew at 86%.
- The increment is 4 points. Across $6.4 million of treated ARR, that's $256,000 of retained ARR the reviews changed.
- The usual method would have credited the full $5.76 million renewed in treated accounts.
- If the review program costs $180,000 a year, the incremental return is positive but modest. The benefit usually reported would have been more than 20 times larger.
Forty accounts is a small holdout, so one cycle gives a rough read. Run it again next cycle before cutting or scaling the program.
How to run a holdout without risking key accounts
A holdout makes CS leaders nervous, and that's reasonable. Here's how to keep the risk small.
- Test a motion, not the whole relationship. Hold back one proactive touch, such as a review or an outreach sequence. Support and renewals continue as normal.
- Exclude your largest accounts. Run the test in a segment where you have enough accounts for a comparison.
- Randomize inside the segment. Don't let CSMs pick who gets held back, or you'll rebuild the selection problem.
- Set the outcome in advance. Renewal, expansion or a leading signal such as active users. Pick one before the test.
- Check for harm. Compare churn in both groups, not just renewal. The 2016 study shows a touch can backfire.
Then sort your motions. Keep the ones with clear lift. Rework the ones with none. Stop the ones with negative lift before adding headcount. Our guide to the customer success dashboard covers how to show treated and holdout results side by side.
Keeping the treated and holdout groups on one record
A holdout needs three things in one place: which accounts were treated, what the team did, and what happened next. In most stacks, the CS platform holds the plan, the helpdesk holds the requests, and the CRM holds the renewal.
MatrixFlows keeps them on one foundation. Accounts, success plans, customer requests and renewal dates are linked tables in Matrix. A field can mark each account as treated or held back. Automations can run the proactive touch only for treated accounts. Views can compare renewal, requests and adoption signals across the two groups without an export.
Back to the CFO's question. With a holdout in the same segment, the slide would show the difference coverage made, not the difference contract size made.
Create a Free Workspace → and set up your first CS holdout this quarter.
.svg)
.svg)
.svg)
