Before you scale spend on a leaky bucket
SaaS NRR (Net Revenue Retention): A Retention Tripwire Before You Scale Spend
A board slide that prints “NRR is 118%” without naming the starting cohort, the window, or what you will do if it misses is not validation. Net revenue retention asks a smaller, honester question: after churn, downsell, and expansion, did the customers who were live at period start pay you more or less than they did then — against a bar you wrote first? Matching it is a decision signal — not a forecast that you will succeed, and not a substitute for naming the weakest retention or expansion assumption first.
Last updated: October 18, 2026
Direct answer
What is SaaS NRR — and why write the tripwire first?
NRR (net revenue retention / net dollar retention) measures how much recurring revenue you keep from a starting customer cohort after churn, downsell, and expansion/upsell. GRR (gross dollar retention) is the same idea without expansion — so GRR cannot exceed 100%. Matching or missing a pre-written tripwire is a decision signal — not a prediction that you will succeed.
Key takeaways
What to remember
- GRR / GDR (gross dollar retention) = (Beginning ARR − Churned ARR − Downsell ARR) / Beginning ARR. Expansion is excluded. Cap at 100%. NRR / NDR (net dollar retention) = (Beginning ARR + Expansion ARR − Churned ARR − Downsell ARR) / Beginning ARR. It can exceed 100% when expansion outruns churn plus downsell. Prefer a starting-base cohort (customers live at period start) over inventing a proprietary formula. Logo/customer churn % is not dollar retention.
- Bessemer (Scaling to $100 Million, 21 September 2021), private cloud portfolio 2010–1H21 — not a random market sample: retention preserves the unit economics of past CAC; every percentage taken out of retention is taken out of growth (“leaky bucket”). Strong retention in their lesson framing is about 85% gross, about 120% net. Gross retention sits relatively consistently around 85–90% across ARR scale; successful SMB cases can sit around 70–80% (HubSpot pre-IPO in their note) — then watch CLTV/CAC.
- Bessemer net retention ranges: about 105–145% at $1–10MM ARR; about 105–125% by $100MM+ ARR. Middle 50% still above 100% at every scale; only the bottom quartile sits below 100%. Averages they cite: about 140% NRR at $1–10MM ARR → about 120% for $10–100MM+. Medians by ARR scale: $1–10MM 125%; $10–25MM 117%; $25–50MM 120%; $50–100MM 120%; $100MM+ 115%. Segment examples: Mindbody IPO about 109% NRR on about $2K ACVs vs Okta about 123% on $50K+ ACVs.
- a16z (Casado, 10 February 2020): Net Dollar Retention is the best indicator of expansion motion, net of churn and downsell. Gross Dollar Retention reflects churn plus downsell and how mission-critical the product is. First Round Levels of PMF (8 April 2024), sales-led B2B: Level 2 NRR at least 100%; Level 3 >110% NRR; Level 4 >120% NRR. Optional attributed Looker Level-4 vignette: 141% NRR. Optional KeyBanc/Sapphire 2024 operator survey medians — labeled as survey, not law: NDR about 102% / 102% / 101% (2022 / 2023 / 2024E); GDR about 89% / 89% / 91%.
- Write the tripwire before you measure: signal (NRR and/or GRR) + threshold (e.g. NRR ≥100% / ≥110% — a Yibud-chosen tripwire, not First Round Level 3, which is strictly >110%) + definition (dollar vs logo; include/exclude expansion) + cohort window (typically 12 months) + named action (fix retention first / keep experimenting / scale spend / cut expansion bets). A Yibud score is not NRR and not a success prediction.
Why it matters
Why retention is a gate on scaling spend — not a trophy
Founders treat “we can grow this” as a feeling that arrives after a good week of new logos. Feelings do not name whether the customers you already paid to acquire are paying you more, or leaking. Two published sittings are useful here as instruments, not as personal odds: Bessemer on the leaky bucket, and a16z on expansion motion versus how mission-critical the product is.
Bessemer: every point of retention you lose is a point of growth you do not get
D’Onofrio, Ding & Atlas editors (Scaling to $100 Million, 21 September 2021) treat retention as what preserves the unit economics of historical customer acquisition. When you keep existing customers, you do not spend sales-and-marketing dollars refilling a leaky bucket — you can spend them on net new revenue. Every percentage taken out of retention is taken out of the growth rate. Their lesson framing for strong retention is about 85% gross, about 120% net. That is a spend gate on a starting base, not a proof you will succeed.
a16z: Net Dollar Retention is the expansion-motion indicator
Casado (11 Key GTM Metrics for B2B Startups, 10 February 2020) writes that B2B economics are lately more driven by expansion than by upfront ACV, and that Net Dollar Retention is the best indicator of that motion — remembered as net of churn and downsell. Use NRR when the decision is “does expansion on the existing base outrun what we lose?” Do not treat a high NRR as a forecast that new logos will keep arriving.
a16z: Gross Dollar Retention is the churn-plus-downsell / mission-critical signal
Casado’s Gross Dollar Retention sitting reflects churn and downsell, and is often an indicator of how mission-critical the product is. a16z’s 16 Startup Metrics (21 August 2015) makes the same split in churn language: gross churn estimates the actual loss; net revenue churn understates losses because it blends upsells with absolute churn. Report GRR next to NRR so expansion cannot hide a leak. A 118% NRR with an 82% GRR is a different company from a 118% NRR with a 96% GRR.
The five terms
GRR, NRR, starting-base cohort, logo vs dollar, tripwire
These five fields are the instrument. If any one is fuzzy, the number is a slide, not a tripwire. This page does not publish a grade table beyond Bessemer’s attributed ranges, First Round’s Level 2 ≥100 / Level 3 >110 / Level 4 >120, and the optional KeyBanc survey medians labeled as survey. Use those, then write your own bar.
GRR / GDR — gross dollar retention, expansion excluded, cap 100%
KeyBanc’s 2024 survey writes the same formula Bessemer describes in words: Gross Dollar Retention = (Beginning ARR − Churned ARR − Downsell ARR) / Beginning ARR. Expansion is out. The ratio cannot honestly exceed 100%. Bessemer: it tells you what percentage of revenue you maintained; it nets out customers who turned off or downgraded. Casado: it reflects churn and downsell, and often how mission-critical the product is. a16z 2015: gross churn estimates the actual loss to the business.
NRR / NDR — net dollar retention, expansion included, can exceed 100%
Net Dollar Retention = (Beginning ARR + Expansion ARR − Churned ARR − Downsell ARR) / Beginning ARR. Bessemer: net retention accounts for upsells and expansion over the period. Casado: it is the best indicator of expansion motion, net of churn and downsell. It can print above 100% when expansion > churn + downsell. That is not a license to skip GRR. a16z 2015: net revenue churn understates the losses because it blends upsells with absolute churn.
Starting-base cohort — customers live at period start
Prefer the customers who were already paying at the start of the window. New logos booked during the window are not this sitting — they belong in net-new ARR, Magic Number, or a CAC page. Bessemer’s gross/net retention language is about revenue maintained from the existing base. Inventing a proprietary Yibud formula that mixes this month’s new logos into “NRR” is how founders flatter a leak.
Logo churn vs dollar retention — different instruments
Logo / customer churn % counts accounts that left. Dollar retention weights by ARR. Ten $2K logos leaving can print a kinder logo-churn number than two $50K logos leaving. Bessemer’s Mindbody vs Okta sitting is dollar NRR on different ACVs, not a logo-churn table. If you report logos, label them as logos. Do not call a logo-churn percentage NRR.
The tripwire — written before you measure
Signal + threshold + definition (NRR vs GRR; dollar vs logo; include/exclude expansion) + cohort window (typically 12 months) + named action (fix retention first / keep experimenting / scale spend / cut expansion bets). Write it on a dated page before the first cohort report. A kinder bar after the results is a recap, not a tripwire. Matching the bar is a decision aid. It does not prove you will succeed.
The arithmetic
Cohort math — not a proprietary Yibud formula
There is no Yibud NRR formula. Use the starting-base cohort sketches the sources actually wrote, name which one you are running, and do not swap expansion in or out after the report. The fictional worked example later on this page is invented so you can see the crossing; it is not a benchmark.
Gross dollar retention (GRR / GDR)
GRR = (Beginning ARR − Churned ARR − Downsell ARR) / Beginning ARR. Cap at 100%. Expansion is excluded.
Beginning ARR is the recurring revenue of customers live at period start. Churned ARR left. Downsell ARR stayed and paid less. Do not add expansion here. If the ratio wants to print above 100%, you included expansion — that is NRR, not GRR. KeyBanc writes this formula explicitly; Bessemer describes the same sitting in words.
Net dollar retention (NRR / NDR)
NRR = (Beginning ARR + Expansion ARR − Churned ARR − Downsell ARR) / Beginning ARR. Can exceed 100% when expansion > churn + downsell.
Same beginning base as GRR. Expansion/upsell is the only extra term. Casado: remember it is net of churn and downsell. a16z 2015: do not let the net number understate the actual loss. Report GRR in the next column so a 118% NRR cannot hide an 82% GRR.
Same window, same starting base — worked crossing (fictional inputs)
Beginning ARR $100. Churned $8. Downsell $4. Expansion $22. GRR = (100 − 8 − 4) / 100 = 88%. NRR = (100 + 22 − 8 − 4) / 100 = 110%.
These dollars are invented so the arithmetic is visible. They are not Bessemer medians, not First Round Level 3, and not a Yibud report. Same starting $100, two answers. The 110% NRR would clear a ≥100% bar you wrote — First Round Level 2 (NRR at least 100%) — and would not clear Level 3, which is strictly >110%. The 88% GRR would sit inside Bessemer’s ~85–90% gross band, and would not be “saved” by the expansion. Keep both.
Do not annualize a 90-day window into “annual NRR” by multiplying. Do not mix logo counts into the dollar ratio. Do not swap a company-wide blend that includes this quarter’s new logos into a starting-base cohort after the close. The tripwire names the window and the base before the first report.
Published benchmarks
Bessemer, First Round, and an optional survey — labeled, not a grade table
The rows below are what the sources actually named. They are not a Yibud letter-grade chart and not a personal odds table. Bessemer’s numbers are a private cloud portfolio, 2010–1H21 — not a random market sample. First Round’s Levels are sales-led B2B. KeyBanc/Sapphire is an operator survey. Use them as vocabulary, then write your own tripwire for the cohort you can actually measure. This page does not invent First Round bands beyond Level 2 ≥100 / Level 3 >110 / Level 4 >120.
| Source | What they name | How to label it |
|---|---|---|
| Bessemer, Scaling to $100 Million — gross retention | Gross retention (dollar basis) relatively consistent around 85–90% across ARR scale. Successful SMB cases can sit around 70–80% (HubSpot in its 3 years pre-IPO in their note) — then watch CLTV/CAC. Lesson framing for strong retention: about 85% gross. | Bessemer portfolio sitting, 2010–1H21. Not a seed law. CLTV/CAC belongs on the LTV:CAC page. |
| Bessemer — net retention ranges | About 105–145% at $1–10MM ARR; decreasing slightly to about 105–125% by $100MM+ ARR. Middle 50% still exceeds 100% at every scale. Only the bottom quartile of cloud companies have <100% net retention. | Range, not a personal forecast. Bottom-quartile <100% is their portfolio cut, not “you will fail.” |
| Bessemer — averages | Average net retention about 140% between $1–10MM of ARR, decreasing to about 120% for $10–100MM+. | Averages they cite. Not a Yibud “excellent” grade, and not First Round Level 4. |
| Bessemer — median benchmarks by ARR scale | $1–10MM: 125% NRR. $10–25MM: 117%. $25–50MM: 120%. $50–100MM: 120%. $100MM+: 115%. | Medians by ARR band, attributed. Do not flatten them into one homemade grade table. |
| Bessemer — SMB vs enterprise examples | Mindbody IPO about 109% net retention on about $2K ACVs, versus Okta about 123% on $50K+ ACVs. SMB segments often see lower net retention than enterprise. | Attributed IPO examples, dollar NRR on different ACVs. Not a logo-churn table. |
| First Round Levels of PMF, 8 April 2024 | Sales-led B2B. Level 2 (Developing): NRR at least 100%. Level 3 (Strong): >110% NRR (with other Level 3 gates elsewhere — Magic Number, CAC payback, burn multiple). Level 4 (Extreme): >120% NRR. Looker Level-4 vignette, attributed: 141% NRR. | Labeled Level marks, not a law for every consumer app. This page does not invent extra First Round NRR bands. |
| KeyBanc Capital Markets & Sapphire Ventures, 2024 SaaS Survey | Operator survey medians: Net Dollar Retention about 102% / 102% / 101% (2022 / 2023 / 2024E). Gross Dollar Retention about 89% / 89% / 91%. Same formulas as above. | Survey medians, labeled as survey — not a law, and not Bessemer’s private-cloud averages. |
First Round puts NRR inside Efficiency, next to CAC, Magic Number, CAC payback, gross margin, and burn multiple. Those neighbor instruments live on their own Yibud pages. This page does not copy First Round’s Magic Number, payback, or burn-multiple tables, and does not invent Level 1 NRR bands.
Scope notes, still the sources: Bessemer’s sample is their private cloud portfolio through 1H21, not every SaaS company on earth. First Round’s Levels are written for sales-led B2B. KeyBanc/Sapphire 2024 is a private-company operator survey (100-plus respondents in the accompanying release). Do not flatten all three into one homemade “good NRR” letter grade.
How to run it
How to run an NRR tripwire without faking scale
This is Yibud’s suggested sequence, not a Bessemer checklist, not an a16z mandate, and not a First Round law. Honor the tripwire you wrote before the first cohort report. If you cannot name the decision the number would change, you are building a slide.
- 1
Write the tripwire before you look
Signal is NRR and/or GRR for a named starting-base cohort, plus its direction versus the prior vintage. Threshold is the miss / pass line you will honor — First Round’s Level 2 ≥100% and Level 3 >110% are classic gates, not a seed law, and not proof you will succeed. Name dollar vs logo, and whether expansion is in (NRR) or out (GRR). Window is typically 12 months. Action is fix retention first, keep experimenting, scale spend, or cut expansion bets. Do not rewrite the bar after the report.
- 2
Lock the starting base and the window
Customers live at period start. Beginning ARR is their recurring revenue that day. New logos during the window stay out of this sitting. If you only have 90 days, report a 90-day GRR/NRR and say so — do not multiply by four and call it annual. Bessemer’s published ranges are period retention of an existing base, not a slogan.
- 3
Compute GRR first, so expansion cannot hide churn
Subtract churned ARR and downsell ARR from beginning ARR, then divide. Cap at 100%. Casado: this is the churn-plus-downsell / mission-critical signal. a16z 2015: gross estimates the actual loss. If GRR is 82% and NRR later prints 118%, you learned something expansion had been covering. Keep the 82% on the dated page.
- 4
Compute NRR on the same base. Then look at neighbors as separate sittings.
Add expansion, subtract the same churn and downsell, divide by the same beginning ARR. Casado: expansion motion, net of losses. Do not average NRR with LTV:CAC, CAC payback months, Magic Number, burn multiple, or Rule of 40. Those are different decisions on different pages.
- 5
Honor the action. If you are pre-revenue, use a different instrument.
If the window closes and you missed the bar you wrote, do the thing you named — fix retention first, keep experimenting, or do not scale spend. Do not invent next year’s expansion so NRR clears 110%. Pre-revenue, no renewals yet, or no expansion motion: use willingness-to-pay, smoke, interviews, or the Ellis PMF survey first. A Yibud score is not NRR.
What you leave with
The NRR tripwire (copy this)
If the sitting produced a percentage and no dated action, you did not finish. Fill the brackets. Leave nothing as “retention is fine,” “soon,” or “we’ll know it when the round happens.”
The fill-in contract
Product: [name]. Cohort: [customers live as of start date — starting ARR $ ]. Hypothesis this number can falsify: [expansion on this base outruns churn + downsell, so we can scale spend]. Definition: dollar NRR and dollar GRR (not logo churn); expansion in NRR, out of GRR. Beginning ARR: [ ]. Churned ARR: [ ]. Downsell ARR: [ ]. Expansion ARR: [ ]. GRR: [ ]. NRR: [ ]. Signal: [NRR and GRR, and/or change vs last vintage]. Threshold: [e.g. honor a miss if 12-month starting-base NRR is below 100%, or at or below 110% (does not clear >110%) if we wrote First Round Level 3 as the gate — Yibud template using First Round L2 ≥100 / L3 >110, not Bessemer’s 140% early-stage average pasted onto a fourteen-person company, and not KeyBanc’s ~102% survey median treated as a law]. Window: [typically 12 months], books close [date], written before the first report. Action if miss: [fix retention first / keep experimenting / do not 3× spend / cut expansion bets]. Compare the next vintage on the same definitions. A Yibud score does not fire this line.
The “below 100% / at or below 110% (does not clear >110%)” line is Yibud’s suggested template from First Round’s Level 2 and Level 3 NRR marks — not Bessemer’s 140% average treated as a seed target, not KeyBanc’s survey median treated as a law, and not permission to skip GRR. Rewrite the threshold before you measure if you have a reason. Do not rewrite it after.
Example tripwire (Yibud’s suggested template, not a law)
If, for customers live on 1 January, 12-month dollar GRR is below 85% or 12-month dollar NRR is below 100% when we close that vintage on 31 December → we do not 3× paid acquisition in Q1; we fix retention first and name the weakest churn or expansion assumption. Do not annualize a Q1 window into “annual NRR.” Do not swap logo churn in after the close without saying you changed the definition. Report GRR next to NRR so expansion cannot hide the leak.
A 110% NRR on a 12-month window is still one vintage. A Yibud report score does not fire this tripwire. If you have no paying customers and no renewals yet, this template is the wrong first instrument — use willingness-to-pay, a smoke test, interviews, or the Ellis PMF survey instead.
The fork
When to use an NRR tripwire — and when not to
NRR is the wrong first instrument when there is no recurring revenue yet, no renewals, or no expansion motion you are willing to measure honestly. It is the right instrument when a starting customer base already pays, a window can close, and you are deciding whether expansion can outrun churn before you scale spend.
Use it when you have a starting base, a window, and a spend-or-fix decision
You can name beginning ARR for customers live at period start, and you can count churned, downsell, and expansion ARR on that same base. You are about to 3× a channel, tell yourself “NRR is fine,” or skip GRR because expansion looks kind. That is when a dated pair of NRR and GRR earns its keep. Prefer a 12-month window when you have it. Optional: run LTV:CAC, CAC payback, Magic Number, burn multiple, or Rule of 40 as separate sittings — do not average them into a homemade “retention score.”
Do not start here when you are pre-revenue, have no renewals yet, or still need first money, reach, or a who-filter
No paying customers yet? There is no honest beginning ARR. Do not invent a forecast numerator. No renewals, only first invoices? A 90-day “annual NRR” is a slogan. No expansion motion — seats, usage, or upsell — and you still want to print 120%? You are mixing new logos into the base. Still writing a deposit / pre-order / priced checkout? That is willingness to pay. Still testing whether anyone clicks? Smoke, fake-door, or landing-page tests. Still asking whether current users would be very disappointed? Ellis PMF survey. Still writing Stop / Pivot / Continue? Those are the decision contracts. This page supplies one retention instrument those contracts can later read. It does not replace them. A Yibud score is not NRR.
Worked example
One starting-base cohort you can copy (illustrative — fictional)
The cohort below is illustrative. Names, product, and dollars are invented so you can see the arithmetic. It is not a study, not an industry benchmark, and not a Yibud report. Copy the method, not the numbers.
The promise, written before the vintage closes
Illustrative names: founder Amira Chen at Northline, a scheduling tool for independent clinics. Recurring revenue exists. A starting base exists. The tempting trophy is a board slide that prints NRR = 124% from a 90-day window with this quarter’s new logos mixed in, and no GRR column. The tripwire is written first, on 2 January: honor a miss if the 1 January starting-base 12-month dollar NRR is below 100%, or if 12-month dollar GRR is below 85%, before we 3× paid acquisition — Yibud’s suggested template from First Round Level 2 (≥100% NRR) and Bessemer’s ~85% gross lesson framing, not Bessemer’s 140% early-stage average pasted onto a sixteen-person company, and not First Round Level 3 treated as a seed scorecard.
The inputs, kept honest
Customers live 1 January: beginning ARR $80,000. During the year: churned ARR $7,200; downsell ARR $4,000; expansion ARR $16,000. New logos booked in March ($12,000 ARR) are kept out of this sitting. GRR = (80,000 − 7,200 − 4,000) / 80,000 = 86%. NRR = (80,000 + 16,000 − 7,200 − 4,000) / 80,000 = 106%. A 90-day sketch that mixed new logos and annualized expansion could have printed something like 124%. Same company. Different definition. Different decision.
The tripwire, written 2 January
Signal: 1 January starting-base 12-month dollar NRR and dollar GRR. Threshold: honor a miss if NRR is below 100% or GRR is below 85% before we 3× paid acquisition. Window: vintage books close 31 December. Action if miss: do not 3× paid in Q1; fix retention first; name the weakest churn or expansion assumption. A Yibud score does not fire this line. The 100% / 85% pair is Yibud’s suggested template from First Round Level 2 and Bessemer’s gross lesson framing, not KeyBanc’s ~102% survey median treated as a target, and not permission to skip GRR.
What a miss looks like — and what must not happen
Illustrative close: NRR 106% clears the 100% bar they wrote. GRR 86% clears the 85% bar. That is a pass on the dated page — and still not a forecast they will succeed. Had GRR printed 78% while NRR printed 112% from a noisy expansion quarter, that would have been a miss on GRR: expansion hiding churn, the sitting a16z warned net numbers can understate. Do not swap the 124% annualized 90-day sketch in after the close and call the debate settled. Do not swap logo churn (2 of 40 clinics left = 5%) in and call it NRR. Do not treat First Round’s Level 4 >120% as permission to ignore a GRR miss. Do not treat 106% as proof the startup will succeed, either. Keep the next vintage on the same starting-base pair.
Write the tripwire before the first report. If the window closes and NRR or GRR misses the bar you wrote, honor the action. If both cleared the bar, continue to the next cheapest test — usually the next vintage on the same definitions, or a kill/pivot contract that now has a number to read. The method is the dated page, not the 124% slide. Neither 86% GRR nor 106% NRR predicts success.
Not the same page
NRR vs LTV:CAC, CAC payback, Magic Number, burn multiple, Rule of 40, vanity metrics, WTP, Ellis, NPS, failure rates, Yibud scores, and decision contracts
Several Yibud pages sit next to this one. Mixing them up turns a unit-economics ratio, months-to-recover-CAC, a sales-efficiency ratio, a company cash multiple, a growth-plus-profit sum, a waitlist, a recommend score, or a mid-band engine number into “NRR is 110%, so scale.”
LTV:CAC — lifetime value ÷ CAC, not period cohort retention
That page asks whether a cohort’s lifetime (or 12/24-month) contribution cleared a multiple of what it cost to acquire them. This page asks whether the starting ARR base, after churn, downsell, and expansion, cleared a retention bar you wrote. A 3.2 LTV:CAC does not tell you NRR. A 110% NRR does not tell you lifetime value is three times CAC. Keep the instruments on separate pages.
LTV:CAC for startups →CAC payback — months to recover CAC, not this percentage
That page asks how many months of contribution recover fully-loaded CAC for a named cohort or channel. This page is NRR% / GRR% on a starting base. A 7-month payback does not tell you NRR is 110%. A 110% NRR does not tell you LinkedIn ads pay back in months. Keep the instruments on separate pages.
CAC payback for startups →SaaS Magic Number — S&M → incremental revenue, not cohort retention
That page asks how efficiently prior-period sales and marketing spend turned into incremental subscription revenue (often with a one-quarter lag). This page is retention of customers who were already live. A 1.2 Magic Number does not tell you NRR. A 110% NRR does not tell you last quarter’s S&M paid for itself in new ARR. First Round lists both under Efficiency; they are still different sittings.
SaaS Magic Number →Burn multiple — company cash vs net new ARR, not starting-base retention
That page asks how much company cash you burned, in a period, per dollar of net new ARR (Net Burn ÷ Net New ARR). This page is NRR/GRR on a starting cohort. A 2× burn multiple does not tell you NRR. A 92% GRR does not tell you the company is burning too much cash per new ARR. Keep the instruments on separate pages.
Burn multiple for startups →Rule of 40 — company growth plus profit, not this cohort ratio
That page asks whether revenue growth rate plus profit margin cleared a ≥40 bar at the company level. This page is one starting base’s dollar retention. A 42 Rule of 40 does not tell you NRR is 110%. A 110% NRR does not tell you growth plus EBITDA is healthy. Keep the instruments on separate pages.
SaaS Rule of 40 →Vanity metrics — hygiene class, not this retention tripwire
That page is the sorting hat: hits, cumulative signups, and waitlist size feel like progress and do not tell you what to do next. This page is one specific actionable pair. A pretty NRR chart with new logos mixed in, no GRR, and no tripwire can still be vanity. A waitlist is not beginning ARR.
Vanity metrics for startups →Willingness to pay — first money on a deal, not cohort retention
That page is a deposit, pre-order, paid concierge, or priced checkout as proof that money moved once. NRR needs a starting base and a window with renewals (and, for the net number, an expansion motion). A paid first month with no renewal sitting should not be running this page yet.
Willingness-to-pay test →Sean Ellis PMF survey — disappointment among current users
That page’s signal is the percent of recent real users who would be very disappointed if the product disappeared. First Round’s >110% NRR gate is a dollar-retention sitting, not Ellis’s question. Do not treat “very disappointed” as NRR, and do not treat a 110% NRR as Ellis’s bar.
Product-market fit survey →NPS — one recommend-intent score
That page is stated recommendation intent among people who already used you. A high NPS does not raise NRR. A 110% NRR does not mean people would recommend you. Keep the instruments on separate pages.
NPS for early-stage startups →Startup failure rate — population frequencies, not the math
That page is what the failure data says at population level — BLS establishment survival, venture-backed “failure” as reported, and why a base rate is not your fate. This page teaches one retention instrument those tables do not compute. It does not invent a “percent of startups that die from sub-100% NRR.”
What percentage of startups fail →Yibud scores — not NRR, and not a success prediction
The validation-score and score-calculator pages explain Startup MRI numbers from a deterministic rule engine on a one-line idea plus five questions. Those scores point at the weakest assumption to test next. They are not NRR. They are not GRR, LTV:CAC, CAC payback, Magic Number, burn multiple, or Rule of 40. Do not read a mid-band monetization score as “net retention is 110%,” and do not read a 110% NRR as a Yibud GO.
Kill, pivot, critical assumption, decision — contracts that can read this number
Those pages write Stop, rewrite one dimension, name the one claim that would kill the idea if false, or decide after evidence. This page can supply the retention instrument those contracts later read: a dated NRR and GRR against a bar you wrote first. An NRR slide with no starting base and no GRR cannot fire a kill line honestly.
Where Yibud fits
Neither NRR nor a Yibud score predicts success
Yibud is a free, no-signup startup idea validator. Scores come from a deterministic rule engine, not from a language model guessing success. Optional AI text, when it is used, only polishes prose. It does not invent the number. A Startup MRI report can name a weak monetization or distribution assumption, then a Day-7 Continue / Refine / Re-test / Stop signal. The score is a flashlight on the weakest assumption to test next — not NRR, not GRR, not LTV:CAC, not CAC payback, not Magic Number, not burn multiple, not Rule of 40, not a forecast. NRR work sits next to that assumption: write the tripwire, do not hope a 124% slide means the bucket does not leak. If you already have a report, start with the named assumption, not the overall score. Then run the cohort yourself. The method on this page stands alone if you never open the analyzer.
Analyze my idea →Common mistakes
What usually wastes an NRR week
Reporting NRR without GRR, so expansion hides churn
a16z 2015: net revenue churn understates the losses because it blends upsells with absolute churn. Casado: Gross Dollar Retention is the churn-plus-downsell sitting. A 118% NRR with an 82% GRR is not the same company as a 118% NRR with a 96% GRR. Put GRR on the dated page. Do not let expansion cover a leak you already paid CAC to fill.
Calling logo churn NRR — or mixing new logos into the starting base
Logo/customer churn % counts accounts. Dollar retention weights by ARR. Bessemer’s Mindbody vs Okta examples are dollar NRR on different ACVs. New logos booked during the window are net-new, not this cohort. Mixing them in is how a leaky bucket prints 124%.
Annualizing a short window into “annual NRR”
A 90-day expansion spike, multiplied by four, is not a 12-month starting-base NRR. Bessemer’s published ranges and First Round’s Level marks assume a real retention period, not a slogan. Write the window. Keep it.
Treating ≥110% NRR as proof you will succeed
First Round’s Level 3 mark is >110% NRR for sales-led B2B, with other gates on other pages. Clearing it means that vintage cleared a bar you wrote — or cleared their Level 3 line if that is the bar you chose. It does not mean the startup will succeed. It is not Ellis’s very-disappointed question, not a deposit, and not a Yibud GO. Neither NRR nor a Yibud report score predicts success.
Scaling spend before the tripwire — or inventing First Round grade bands
Bessemer: retention preserves past CAC; a leaky bucket spends S&M refilling what you already bought. Write the bar before the first vintage. First Round names L2 ≥100, L3 >110, L4 >120; this page does not invent extra NRR bands. Do not 3× a channel because the slide looks like 124%. Do not paste Bessemer’s CAC or Magic Number tables onto this page — those belong next door.
Sources
Where these ideas come from
- Mary D’Onofrio, Ethan Ding & Atlas editors / Bessemer Venture Partners, “Scaling to $100 Million,” 21 September 2021 — Used for the leaky-bucket / growth-rate link; gross retention relatively consistent ~85–90% across ARR scale; successful SMB cases ~70–80% (HubSpot pre-IPO note) then watch CLTV/CAC; net retention ranges ~105–145% at $1–10MM ARR → ~105–125% at $100MM+; middle 50% >100% at every scale; only bottom quartile <100%; averages ~140% → ~120%; medians 125 / 117 / 120 / 120 / 115 by ARR band; Mindbody ~109% vs Okta ~123%; lesson framing ~85% gross / ~120% net. Private cloud portfolio 2010–1H21 — not a random market sample. Bessemer CAC and Magic Number tables are not copied here.
- Martin Casado / a16z, “11 Key GTM Metrics for B2B Startups,” 10 February 2020 — Used for Net Dollar Retention as the best indicator of expansion motion, net of churn and downsell; Gross Dollar Retention as the churn-plus-downsell sitting and often an indicator of how mission-critical the product is. This page does not pull Casado’s CAC-payback month bands — those belong on the payback page.
- Jeff Jordan, Anu Hariharan, Frank Chen, and Preethi Kasireddy / a16z, “16 Startup Metrics,” 21 August 2015 — Used only for gross vs net churn wording: gross churn estimates the actual loss to the business; net revenue churn understates the losses because it blends upsells with absolute churn. This page does not invent extra NRR benchmarks from that essay.
- First Round Capital, “Levels of PMF,” 8 April 2024 — Used for sales-led B2B Level 2 NRR at least 100%; Level 3 >110% NRR; Level 4 >120% NRR; and the attributed Looker Level-4 vignette of 141% NRR. This page does not invent extra First Round NRR bands, and does not copy their Magic Number, CAC payback, or burn-multiple tables.
- KeyBanc Capital Markets & Sapphire Ventures, 2024 SaaS Survey PDF — Used for operator-survey medians only, labeled as survey not law: Net Dollar Retention ~102% / 102% / 101% (2022 / 2023 / 2024E); Gross Dollar Retention ~89% / 89% / 91%; and the formula definitions Gross Dollar Retention = (Beginning ARR − Churned ARR − Downsell ARR) / Beginning ARR and Net Dollar Retention = (Beginning ARR + Expansion ARR − Churned ARR − Downsell ARR) / Beginning ARR.
In one paragraph
Summary you can quote
NRR (net revenue retention / net dollar retention) measures how much recurring revenue you keep from a starting customer cohort after churn, downsell, and expansion. GRR (gross dollar retention) is the same sitting without expansion, and cannot exceed 100%. Bessemer (21 September 2021), private cloud portfolio 2010–1H21: leaky bucket — every point of retention lost is a point of growth lost; GRR relatively consistent ~85–90% (SMB cases can sit ~70–80%); NRR ranges ~105–145% at $1–10MM ARR → ~105–125% at $100MM+; middle 50% >100%; only bottom quartile <100%; averages ~140% → ~120%; medians 125 / 117 / 120 / 120 / 115 by ARR band; Mindbody ~109% vs Okta ~123%. a16z Casado (10 February 2020): NDR = expansion motion net of churn/downsell; GDR = churn+downsell / mission-critical. First Round (8 April 2024): L2 NRR ≥100%; L3 >110%; L4 >120%; Looker L4 vignette 141% NRR. Optional KeyBanc/Sapphire 2024 survey medians: NDR ~102/102/101; GDR ~89/89/91 — labeled as survey. Write signal + threshold + definition + cohort window + action before you measure; report GRR next to NRR. Matching the gate is a decision signal. Neither NRR nor a Yibud score predicts success.
FAQ
Questions founders actually ask
What is SaaS NRR (net revenue retention)?
How much recurring revenue you keep from customers who were live at period start, after churn, downsell, and expansion. KeyBanc writes NRR / NDR as (Beginning ARR + Expansion ARR − Churned ARR − Downsell ARR) / Beginning ARR. GRR / GDR leaves expansion out and cannot exceed 100%. It is a retention tripwire, not a success score. Write the bar before you measure.
What is a good NRR for a startup?
This page does not publish a universal “good” grade table beyond the sources. Bessemer (portfolio, not a random sample): ranges ~105–145% at $1–10MM ARR and ~105–125% at $100MM+; averages ~140% then ~120%; medians 125 / 117 / 120 / 120 / 115 by ARR band; middle 50% still >100%; only bottom quartile <100%. First Round, sales-led B2B: Level 2 ≥100%; Level 3 >110%; Level 4 >120%. KeyBanc/Sapphire 2024 operator survey medians ~102/102/101 — labeled as survey, not law. Write your own tripwire. Do not invent extra First Round bands.
How is NRR different from GRR and from logo churn?
GRR excludes expansion and is capped at 100%; NRR includes expansion and can exceed 100%. a16z 2015: the net number can understate actual losses because it blends upsells with churn. Logo/customer churn % counts accounts, not dollars. Bessemer’s Mindbody vs Okta examples are dollar NRR on different ACVs. Report GRR next to NRR, and label logos as logos.
How is NRR different from LTV:CAC, CAC payback, Magic Number, burn multiple, and Rule of 40?
LTV:CAC is lifetime (or 12/24-month) value ÷ CAC (a ratio). CAC payback is months of contribution to recover CAC (time). Magic Number is how efficiently prior-period S&M turned into incremental subscription revenue. Burn multiple is company-level Net Burn ÷ Net New ARR in a period. Rule of 40 is growth % + profit %. NRR/GRR is starting-base dollar retention over a window. They answer different decisions. Use this page for the retention pair.
Is a Yibud score the same as NRR? Does it predict success?
No and no. Yibud scores come from a deterministic rule engine on your idea inputs. They point at the weakest assumption to test next. They are not NRR, not GRR, not LTV:CAC, not CAC payback, not Magic Number, not burn multiple, not Rule of 40, and not a forecast. Neither an NRR pass nor a Yibud report score predicts success. You do not need an account to use this page.
Write the NRR tripwire before you scale spend
Yibud scores weak assumptions with a deterministic engine. You write signal, threshold, definition, window, and action while you are still honest — then a starting-base vintage is a tripwire, not a 124% slide.
Generate a Yibud report →