Gross retention, net retention, and the one that gets quoted
Gross retention cannot exceed 100% and net retention can, which is why NRR is the number said out loud — and why a UI change that reduces churn shows up in the two metrics very differently.
Retention is the route from the P&L lesson with the most money attached and the most contested causal chain. It is also two metrics wearing one word. Getting them the wrong way round in a room where somebody owns the number is the fastest way to be read as a person repeating vocabulary rather than using it.
The two formulas, side by side
SaaS Capital states net revenue retention as arithmetic rather than as a concept: “(Monthly Recurring Revenue in December of 2024 only from customers who were customers in December 2023) ÷ (Total MRR in December 2023)”. Gross retention is, in their words, “the same calculation… excluding the upsells, cross-sells, and price increases”.
Read the denominator. It is frozen: last December’s recurring revenue from a cohort fixed on that date. New customers won during the year appear in neither number. Both metrics are measuring what happened to a group of customers you already had.
Now read the numerator, twice. In gross retention, expansion is stripped out, so the numerator can only be smaller than or equal to the denominator — the same customers, paying at most what they paid before. Gross retention cannot exceed 100%. It has a ceiling built into the arithmetic, and a company reporting 103% gross retention has made an error or redefined the term.
In net revenue retention, expansion stays in the numerator while the denominator stays frozen. Existing customers who buy more push the top of the fraction above the bottom. Net retention can exceed 100%, and in the businesses that get talked about most, it routinely does.
| Metric | What is in the numerator | Can it exceed 100%? |
|---|---|---|
| Gross retention | Surviving revenue from the cohort, expansion removed | No |
| Net revenue retention | Surviving revenue plus upsell, cross-sell, price increases | Yes |
a16z gives the same distinction from the churn side, in monthly terms: gross churn is “MRR lost in a given month / MRR at the beginning of the month” while net churn is “(MRR lost minus MRR from upsells) in a given month / MRR at the beginning of the month”. Same structural difference, opposite direction of travel. Both firms are interested parties — SaaS Capital lends to these companies, a16z invests in them — and both are being unusually precise, which is why they are worth quoting rather than paraphrasing.
Which one gets said out loud
Bessemer publishes a good, better, best convention for net revenue retention: 100%, 110%, 120% and above. Note what that table implies. The floor of “good” sits at the exact point gross retention tops out. A benchmark built on the metric with a ceiling would look uniformly worse and would separate companies less, and Bessemer is a venture firm publishing the yardstick its own portfolio is measured by.
Here is this course’s own reading of why NRR is the number in the headline, offered as reasoning rather than as a finding: it is the bigger, less bounded number, it is the one that can be above 100%, and a metric that can clear a round number is a better sentence in a deck. No source in this module says that. It is an inference about incentives, and it is worth holding loosely.
The practical version is not loose at all. When a deck says “retention,” it usually means net. When somebody in the room gets worried about churn, they are usually thinking about gross. Two people can hold that conversation for ten minutes before noticing they are discussing different metrics.
Same metric name, different arithmetic
SaaS Capital’s formula is one method. It is not the method. Snowflake’s Form 10-K for the fiscal year ended 31 January 2025 reports a net revenue retention rate of 126%, computed a different way: over a trailing two-year measurement period rather than December to December, against a cohort defined in the first month of that period, with customers who stopped using the platform kept in the denominator and contributing zero to the numerator.
Both are honest. Both are filed or published. Neither can be compared with the other, and this course is deliberately not going to recompute Snowflake’s business under SaaS Capital’s method to show you by how much — that recomputation would be invented numbers wearing a filed company’s name, which is the exact move this course exists to argue against.
SaaS Capital adds the population warning themselves: their sample is private B2B software companies above $1M of ARR, and they argue that benchmarking retention against public SaaS companies is of limited usefulness. A portfolio company being compared against a public company’s filed NRR is being compared against a different population using different arithmetic over a different window.
Where people get burned
The failure here is not quoting a wrong number. It is quoting a right number from one method into a room using another, confidently, and being corrected by the person who owns the model. That costs more than not knowing, because it puts every other figure in your case under review at the same moment.
Where a UI change actually lands
Take the review gate at its most plausible: extraction errors that previously reached customers now get caught in flow, so a specific reason to leave gets weaker. Three things follow.
- The effect is a gross-retention effect. You are arguing that fewer customers leave, or that fewer downgrade. That is the numerator gross retention measures and the one it caps at 100%.
- It can be invisible in net retention. If expansion is running well, an improvement in the survival term is swamped by upsell in the same fraction. Net retention can rise in a year your churn got worse, and it can sit flat in a year your fix worked.
- It cannot appear faster than the renewal cycle. A customer who was going to leave leaves at a renewal date. If the contracts are annual, the earliest honest read on a change shipped this quarter is a year out, and the clock lesson already told you what a year costs inside a hold period.
That third point is where most retention claims quietly die. It is also the one that makes you credible when you raise it yourself, before anyone asks.
Check your recall
Answer from memory — no scrolling back.
Retrieval check
Write the retention sentence for the review gate that a data lead would not object to. One sentence.
Check your answer
Something close to: if the gate works, it should show up as an improvement in gross retention among accounts with high extraction volume, no earlier than their next renewal, and it will not be separable from expansion in the net number.
Count what that sentence gives away on purpose: the metric, the segment, the earliest date, and the admission that the headline figure cannot carry the claim. Every one of those concessions is a thing the room would otherwise have had to extract from you, and the person who says them first is the person whose remaining claims get believed.
Hands on
Pin a retention claim to a metric, a segment and a date
Done when: Every case in VALUE-CASES.md that mentions retention names gross or net explicitly, names the customer segment the effect would appear in, and states the earliest date the number could move.
- Find every retention word in your three cases — churn, renewal, stickiness, keeps customers. Replace each with gross retention or net revenue retention, chosen deliberately.
- Name the segment. “Accounts processing more than N documents a month” is a segment. “Our customers” is not, and a whole-book number is where a real effect goes to be diluted into nothing.
- State the earliest date the metric could move, and derive it from the renewal cycle rather than from when you would like to report. If you do not know the cycle, write that you need to ask.
- Add one line: definition to confirm — the question you would ask about how this company computes the metric before you commit to moving it.
- Bring the three claims into the chat. I will go straight at any case that names net revenue retention, because a design change almost never has a clean story for the expansion half of that fraction.
What this does not cover
This lesson stayed on customers you already have. It said nothing about what it costs to win one in the first place, which is the next lesson — CAC payback, and the choice of denominator hidden inside every payback number anyone quotes at you. After that comes the Rule of 40, the company-level score with the most folklore attached to it, and then the subtractive move that closes the module.
Read this next — primary source
What is a Good Retention Rate for a Private SaaS Company?SaaS Capital — free; a lender to SaaS companies, so an interested party in how healthy its borrowers look
This lesson takes the two formulas from it and nothing else. The rest of the post is worth your time for a different reason: it is one of the few places that states its sample out loud — private B2B software companies above $1M of ARR, surveyed annually — and then argues against its own readers benchmarking themselves on public-company numbers, which it calls of limited usefulness. Read how carefully a lender describes the population its figures come from, and then notice how rarely the retention number quoted at you comes with any population at all.
Stuck, curious, or think this lesson is wrong? Ask your teaching agent. The lessons are the scaffold; the conversation is where the learning gets unstuck.