How do I calculate true gross retention vs net retention in 2027?

What gross versus net retention actually measure

The two metrics are built from the same four inputs — starting ARR, contraction, churn, and expansion — and they differ in exactly one term. That single structural difference is the whole point of running both.

Gross revenue retention = (starting ARR − contraction − churn) / starting ARR. It measures downside only: how much of the revenue you started the period with is still there, before you credit yourself a single dollar of upsell. Because there is no expansion term in the numerator, GRR is mathematically incapable of exceeding 100%. If your spreadsheet ever prints 101% gross retention, you have a formula bug, not a great quarter — almost always expansion leaking into the GRR numerator at the account level.

Net revenue retention = (starting ARR + expansion − contraction − churn) / starting ARR. It adds expansion back, which is why it can and routinely does clear 100%. A business with net negative churn — expansion exceeding all losses — grows its existing base without signing a single new customer.

Put numbers on it. A cohort starts the fiscal year at $10.0M ARR. Over twelve months, $600K contracts through downgrades and seat reductions, $900K churns as logos lost entirely, and $1.8M expands through upsell, cross-sell, and seat adds.

• GRR = (10.0 − 0.6 − 0.9) / 10.0 = 8.5 / 10.0 = 85%
• NRR = (10.0 + 1.8 − 0.6 − 0.9) / 10.0 = 10.3 / 10.0 = 103%

Same cohort, same period, same underlying events, an 18-point spread. The gross number says you lose 15% of your revenue base every year to churn and contraction. The net number says you grow that base 3% without acquiring anyone. Both are true. Reporting only the flattering one is how companies mislead a board without technically lying.

The reason gross retention is the truer read on business health is that it cannot be dressed up. Consider a 200-customer cohort where 40 customers churn entirely — a 20% logo loss — while three large enterprise accounts triple their spend. The expansion from those three can be large enough to offset all 40 lost logos plus every dollar of contraction, pushing NRR to 110-115%. The board sees 112% and concludes the product is sticky. Meanwhile a fifth of the customer base walked out, and the entire story depends on three accounts continuing to expand. Strip expansion out and that same cohort posts something like 78-82% gross retention — the number that screams what the net number buried.

This asymmetry is why experienced diligence teams ask for gross retention first and treat a company that volunteers only NRR with suspicion. Gross retention is the floor: what you would keep if your expansion motion went to zero tomorrow and every account simply renewed flat. Compound it and the difference is stark. At 92% GRR, roughly 78% of the original cohort revenue survives three years before any expansion. At 75% GRR, roughly 42% survives — a fundamentally different company running up a down escalator, re-acquiring a quarter of its revenue annually just to stand still.

The step-by-step calculation

The formulas are trivial. The discipline that makes them meaningful is cohort construction, and that is where most retention numbers quietly go wrong.

Step one: freeze the cohort. Identify every customer active and contributing ARR at T0. That list is your cohort and it never changes for the life of the measurement. A customer who becomes a customer at T0 + 3 months is not in it and will not be in it — they belong to next period's starting cohort.

Step two: sum the T0 ARR of that cohort. This is your denominator, and it is frozen alongside the customer list. It should be true recurring ARR — not total contract value, not one-time implementation fees, not professional services.

Step three: measure the same customers at T0+12. Only those customers. A churned account contributes $0. A downgraded account contributes its reduced amount. An expanded account contributes its increased amount. The numerator floats; the cohort and denominator do not.

Step four: build the ARR walk. Decompose the change into expansion, contraction, and churn, netted at the logo level. An account that dropped 20 seats but added a module gets netted to a single positive or negative figure — never counted as both expansion and contraction, which double-counts movement and distorts both metrics.

Step five: compute both metrics with the cap enforced. For GRR, cap each customer's numerator contribution at that customer's own T0 amount, so no account can contribute more than it started with. For NRR, leave the per-customer contribution uncapped so expanders pull the aggregate above 100%.

Step six: compute logo retention alongside dollar retention. Logo GRR = (logos at start − logos churned) / logos at start, weighting a $2K account and a $2M account identically.

The trap that destroys this whole construction — and it is astonishingly common — is measuring against "all current customers" instead of a fixed historical cohort. Divide today's total ARR by last year's total ARR and you are not measuring retention at all; you are measuring net growth, because today's ARR includes every new logo signed during the year. A company that grew 60% in new logos can post a "retention" number well over 130% while genuinely losing a third of its base. The math is not lying; the definition is.

The second trap is the rolling or fuzzy cohort, where customers drift in and out of the measured set based on current status. That produces a number nobody can audit or reproduce — ask two analysts to compute it and you get two answers, because the population changes every time the query runs.

A third and subtler trap is deciding what "active at T0" means for customers mid-implementation, in a trial converting to paid, or in a contractual grace period. Pick a bright line — signed contract with positive recurring ARR is the usual one — write it down, and hold it.

Getting the input definitions precise matters as much as the formula. Contraction is any negative ARR movement inside a retained logo: seat reductions, tier downgrades, scope reductions where a module or add-on gets dropped, usage decline in consumption models, and — the one operators most love to pretend is not contraction — discounts given at renewal. If a customer renews at $90K having previously paid $100K because you cut 10% to save the account, that $10K is contraction. It does not matter that seat count was identical or that sales calls it a pricing concession. The ARR went down, and that is what retention measures. Exempting renewal discounts inflates gross retention period after period and trains the sales organization to discount more freely every cycle.

Churn is total logo loss — the account contributes $0 at the measurement date. Non-renewal counts as churn even if the relationship ended amicably and the customer might return; a later win-back is a new-logo or win-back event, never a retroactive un-churn. Back-dating win-backs to erase churn is one of the quieter ways retention gets gamed. Split churn voluntary versus involuntary even though both hit GRR identically: a company with 6% annual churn that is 4% involuntary has a fixable billing-operations problem and can plausibly claw back two or three points with a dunning project, card-updater services, and pre-expiry payment nudges. A company with 6% churn that is entirely voluntary has a product, value, or competitive problem no billing fix will touch. Identical gross retention, completely different fire to fight.

Expansion is any positive ARR movement inside a retained logo: seat adds, tier upgrades, cross-sell into new products or modules, and usage growth. The genuinely contested input is price increases. The purist position says a list-price increase pushed to the existing base is not expansion, because expansion should reflect customers choosing to consume more value, and a price hike is charging more for the same thing. The pragmatic position says ARR genuinely went up and the customer chose to stay rather than churn at the higher price, which is real pricing power. Both are defensible. What is not defensible is silence. Most sophisticated companies count price increases but disclose them on a separate line — "NRR was 116%, of which 4 points was a list-price increase" is honest; folding a 7% across-the-board increase into expansion and celebrating the jump as product stickiness is not. The same separate-disclosure logic applies to FX-driven ARR increases and to contractual auto-escalators baked into multi-year deals: real ARR, but not fresh demand signals.

Benchmarks, windows, and typical ranges

Retention benchmarks only mean something segmented by customer size, because SMB and enterprise books differ structurally, not incrementally.

Gross revenue retention by segment (2026):

Best-in-class enterprise clears 95-97% and the very best sustain above 97%. The pattern is monotonic: the larger the customer, the higher the gross retention, because larger customers are harder to win and correspondingly harder to lose. An SMB SaaS at 88% gross retention is doing very well; sustained below 80% the unit economics are usually broken, because CAC cannot be recovered against a base that thin.

Net revenue retention by segment (2026):

Best-in-class enterprise posts 125-135%, and a few category leaders have sustained above 140% for years.

The practical use of these bands is diagnostic. They tell you whether your number is a structural feature of the segment you sell to or an actual fixable problem. An 85% gross retention is a five-alarm crisis for an enterprise company and a perfectly good result for an SMB company — same number, opposite conclusions. Benchmark within segment, never against a blended industry average, and if you sell across segments, report by segment so a strong enterprise number does not hide an SMB book bleeding out underneath it.

The measurement window is its own decision with real consequences. Annual cohort measurement takes the customers active at the start of a twelve-month period and measures them at the end. It is clean, it maps to how boards and investors think, and it is the standard for external reporting. Its weakness is lag: you do not know the cohort's annual retention until the year closes, by which point the churn has happened and the renewals are lost.

Monthly cohort measurement takes each month's starting cohort, measures it one month later, and chains or aggregates those figures — often as a trailing-twelve-month number. Monthly cohorts give a far earlier read on a deteriorating trend; you can see month three of a bad cohort long before the annual number lands, which is the difference between a fixable problem and a fait accompli. The roll-up also smooths noise: a single large renewal slipping a few days, a lumpy expansion, or a holiday billing lull averages out across twelve data points.

The trade-off is that compounding twelve monthly rates assumes a smoothness that lumpy B2B contract books do not have. A company with 70% of contracts renewing in Q4 gets a misleadingly smooth monthly-compounded figure that averages away the very renewal cliff that matters most. Run both — annual for the scoreboard, monthly for the steering wheel — and reconcile them, because a divergence of more than a couple of points between the monthly-compounded TTM and the true annual cohort is itself a diagnostic signal about renewal concentration or seasonality.

There is a parallel choice about how you bucket the cohort. Calendar-period cohorts group everyone active on January 1 and re-measure December 31 — simple, aligned to the fiscal year, and what external reporting uses. The weakness is that within that window customers sit at wildly different points in their own contract lifecycles: someone who renewed in November is barely at risk inside the calendar year, while someone whose contract ends in March faces a genuine renew-or-leave decision. Calendar retention therefore blends customers who faced a renewal event with customers who did not.

Renewal-date cohorts instead group customers by when their contract actually comes up — the Q2 renewal cohort is every customer whose contract expires in Q2, and you measure what fraction of that specific ARR renewed, contracted, or churned. By construction every account in the cohort faced a real decision, which makes it a much purer read on renewal performance and a fair scoreboard for the renewals team. Its weakness is that it does not map to a fiscal quarter and it structurally ignores mid-term contraction and expansion that happen off the renewal date. Mature companies run both: renewal-date cohorts to manage, forecast, and compensate the renewals motion; calendar cohorts for the board-facing annual numbers. The binding requirement is being explicit about which lens any given number came from, because the two will differ for the same company over the same year, sometimes by several points.

Alongside both metrics, put the SaaS Quick Ratio on the same page: (new ARR + expansion ARR) / (churned ARR + contracted ARR). A Quick Ratio of 4 means every dollar lost was matched by four dollars added. A ratio of 1 means you are running hard to stay in place. Below 1, the bucket is winning. It complements rather than duplicates retention because it pulls new-logo acquisition into the same frame, revealing whether growth comes from a healthy base or papers over a sick one. Above 4 is excellent for an early-stage company, 2-4 is solid, and below 2 warrants real scrutiny — though the bar drifts down as companies scale and large numbers make high ratios arithmetically harder to sustain.

Where teams get the calculation wrong

The formulas are simple enough that people assume they cannot get them wrong, and then they get them wrong the same handful of ways, quarter after quarter.

New logos in the denominator or numerator. The most frequent and most damaging error. It converts a retention metric into a net-growth metric and makes churning businesses look fine, sometimes spectacular. The denominator is only the T0 cohort's T0 ARR; the numerator is only that same cohort, later. If a customer was not in the room at T0, they are in neither number.

Mixing monthly and annual windows. Comparing a monthly figure to an annual one, computing a denominator on an annual basis against a numerator on monthly run-rate, or annualizing a monthly rate by simple multiplication instead of compounding all produce nonsense. Pick one window and apply it identically to both halves of the ratio.

Counting reactivations as expansion. When a previously churned customer returns, that ARR is new or win-back ARR, not expansion of a retained account — the account was not retained, it left and had to be re-acquired. Folding win-backs into expansion inflates NRR and hides a real cost: acquisition effort spent twice on the same logo.

Not capping gross retention at 100%. Any GRR above 100% is a formula bug, typically a customer who contracted on seats and expanded on a module, netted positive, and was then allowed to contribute more than their T0 amount. Enforce the cap per customer and treat any breach as a data-quality incident.

Inconsistent contraction classification. Usually the renewal-discount exemption described earlier, but also treating a "true-up" as neutral or letting a scope reduction pass unrecorded because the renewal closed on time.

Honorable mentions that sink real companies: annualizing partial-period and ramped contracts incorrectly; double-counting an account that both contracted on one product and expanded on another instead of netting at the logo level; counting one-time fees and professional services as ARR; and letting the cohort definition drift between periods so the numbers stop being comparable to your own history.

Reporting only one metric is its own error. There are two completely different things you can retain — logos and dollars — and the gap between them is one of the most informative signals available. Take a SaaS with 100 customers that loses 20 of them, all small accounts averaging $5K. Logo GRR is 80%, which sounds alarming; dollar GRR might be 95%, because the churned accounts were tiny. That 80/95 split is not a contradiction — it is a precise description of a business with a small-account churn problem (bad-fit segment? weak low-end onboarding? a self-serve tier that overpromises?) whose revenue sits in durable large accounts. The inverse is far more dangerous: 95% logo retention with 80% dollar retention means you are keeping the small accounts and losing the whales, a concentration time bomb where the "good" logo number provides false comfort. Report four numbers — logo and dollar, gross and net — not one.

Boundary decisions get made implicitly instead of explicitly. Currency effects: a customer whose local-currency ARR is perfectly flat shows different USD ARR when rates move. Normalize at T0 rates so retention reflects customer behavior rather than currency markets, or report as-is and disclose the FX impact separately — either is defensible if disclosed, neither is defensible if silent. M&A'd customers: when two accounts in your cohort merge, decide whether the survivor is one logo or two and how to treat the combined ARR, which very often contracts as the merged company rationalizes duplicate spend; the common convention treats the consolidated entity as one continuing logo and nets the ARR change. Re-segmentation, entity splits, and contract consolidation raise the same class of question. The meta-rule: the specific choice matters far less than consistency and disclosure. A defensible-but-imperfect rule applied identically for eight quarters beats a theoretically perfect rule applied differently each quarter, because the imperfect-consistent rule still yields a trustworthy trend, and the trend is usually the actionable signal.

Usage-based pricing breaks the clean framework in a way teams routinely mishandle. Revenue moves every period without any customer making a decision: a perfectly happy account spends $40K one quarter and $34K the next because their own business had a slow quarter, a batch job ran less often, seasonal traffic ebbed, or an engineering team optimized their consumption. Under a strict point-in-time definition that $6K drop is contraction, but it is variance, not a churn signal — and a framework treating every dip as contraction makes gross retention look volatile when nothing is wrong, while making a random spike look like expansion you did not earn. Four approaches consumption businesses actually use: measure on a trailing-twelve-month basis so swings average out; use the committed-spend or minimum-commitment floor as the retention base where contracts have one; define an explicit materiality threshold so only sustained declines (persisting two or more consecutive quarters, or falling outside a defined noise band) count as contraction; or publish a purpose-built, clearly labeled usage-retention metric kept separate from any contract-based figures. Whichever you pick, disclose it plainly — a TTM-smoothed 95%, a commitment-based 95%, and a raw point-in-time 95% are not comparable to each other, let alone to a seat-based peer.

The data model underneath is where the numbers actually break. Retention is only as good as the object model it is computed from, and in most companies that model lives in the CRM. Both Salesforce and HubSpot need the same foundation: a contract or subscription object holding source-of-truth ARR with accurate effective dates, and renewal opportunities — a distinct opportunity or deal created automatically for every contract's renewal, close date equal to the renewal date, amount equal to the renewing ARR. The renewal opportunity is where contraction, churn, and flat renewal all get recorded: closed won at the new amount, or closed lost at zero. Expansion is captured either as a positive line on the renewal opportunity or, more auditably, as separate upsell and cross-sell opportunities with their own amounts, close dates, and reason codes.

With that in place, the calculation becomes a deterministic query. The hygiene it demands is the genuinely hard part: every contract needs accurate start and end dates; every renewal must generate a renewal opportunity so nothing slips through a silent auto-renewal that bypasses pipeline entirely; amounts must be true recurring ARR; contraction and expansion need reason codes so the *why* is queryable; and the customer-to-contract relationship must be deduplicated so you never double-count ARR across two records for the same logo or orphan ARR on a contract whose account got merged. In HubSpot the pattern runs on native deal-based recurring-revenue tooling or a subscriptions integration; in Salesforce, CPQ or a billing integration usually owns the contract object and renewal automation. Platform differs, principle does not: no clean contract object and no disciplined renewal process means no trustworthy retention number, just a query returning a confident-looking lie.

And the CRM number will not match the billing number. Stripe, Chargebee, Maxio, NetSuite, Recurly, and Zuora record what customers were actually invoiced and collected; the CRM records what sales believes the contract says. They diverge for legitimate reasons — a contract signed in late December but billed in January, proration on mid-cycle changes, mid-term amendments that updated billing but never got written back to the CRM, credits and refunds, dunning outcomes the CRM cannot see, multi-entity and multi-currency complications. The discipline is to tie the CRM's ARR walk line by line to the billing system's ARR walk every month and chase every variance until it is explained or corrected. The goal is not zero variance — some timing difference is structurally unavoidable — but a known, bounded, documented variance captured in a written bridge anyone can follow. Designate one system as the source of truth for the externally reported number, most commonly billing because it reflects cash reality and survives an audit, and treat the CRM number as the faster operational indicator that should converge to it over the close. When a board asks whether they are looking at the billing number or the CRM number, you need a one-sentence answer and a bridge sitting behind it. A retention number finance has never signed off on is a number that will eventually embarrass whoever presented it.

Reading the pair: a decision framework

The real diagnostic power comes from reading gross and net retention as a pair, because each combination points to a different underlying problem with a different fix. Treat it as a 2x2 where your quadrant is a diagnosis.

High NRR, low GRR — say 118% net on 82% gross — means expansion is masking churn. A subset of accounts expands hard enough to drag the net number well above 100% while the broader base leaks underneath. The danger is twofold: concentration risk, because the story depends on a few whales continuing to expand, and false institutional security, because the board sees 118% and stops asking hard questions. The fix is to attack churn directly and urgently — onboarding, customer-success coverage ratios, the product gaps showing up in churn reason codes, the involuntary-churn billing leaks — and in parallel to model what NRR collapses to if the top ten expanding accounts merely go flat. Run that stress test before the board asks for it.

Low NRR, high GRR — say 99% net on 94% gross — means retention is excellent but the expansion engine is broken. You keep customers extremely well; you simply do not grow them. The base is durable and genuinely valuable, but the problem is monetization and account development. The fix lives in upsell and cross-sell motion, in packaging and pricing that deliberately create expansion headroom, in usage-based or tiered components that let good customers naturally grow spend, and in a customer-success organization compensated and equipped to drive expansion rather than only prevent churn.

Both low — say 98% net on 80% gross — is a crisis. You are churning badly and failing to expand, and the net number hovers near 100% only by luck or because new-logo ARR is leaking into a sloppy cohort definition. This combination almost always signals a product-market-fit or value-delivery problem, and it is existential: every other initiative is secondary until gross retention recovers, because you cannot build a durable company on a base that evaporates.

Both high — say 95% gross with 125% net — is a category winner: a watertight base with a powerful expansion engine compounding on top. The job is to protect it and watch concentration.

The visualization that makes the diagnosis legible is the cohort triangle, also called the cohort retention table or, stacked as an area chart, the layer-cake. Each row is an acquisition cohort — everyone who first became a paying customer in a given quarter. Each column is that cohort's *age*: quarter 0, quarter 1, quarter 2, and so on. The cell at row "Q1 cohort," column "age 4" shows the retained ARR of those customers four quarters later. Because newer cohorts have not aged as far, the populated cells form a triangle.

Read it two ways. Scan *down a column* to compare different cohorts at the same age: if age-4 retention climbs cohort over cohort, retention is structurally improving; if it degrades down the column, the customers you sign more recently are worse, which is a sales-quality, ICP, or onboarding problem wearing a retention costume. Scan *across a row* to watch one cohort's curve unfold: does it decay for a few periods and then flatten, meaning you have a sticky core that stops eroding, or does it decay relentlessly with no floor, meaning there is no retention equilibrium and the cohort eventually goes to near zero? The layer-cake plots the same data as a stacked area chart over calendar time, each cohort its own band; in a net-negative-churn business the bands actually thicken after acquisition as expansion outpaces churn, and that rising stack is the unmistakable visual signature of NRR comfortably above 100%. A blended 90% gross retention can sit on top of a stable 90% equilibrium or on a number that is 96% for old cohorts and 78% for new ones and merely averages to 90% on its way down. The triangle exposes which one you are.

What to put in front of a board. Both metrics, never NRR alone. Segmented by tier so different dynamics are visible rather than blended into mush. Logo and dollar, so concentration is not hidden. A trend line of at least eight quarters so they react to slope rather than to one possibly-noisy point. And ideally the cohort triangle. The rigor standard is GAAP-adjacent — not because retention is a GAAP metric (it explicitly is not) but because a serious board expects the same discipline: a written definition, consistent application, reconciliation to the audited financials, and footnoted treatment of every edge case.

The credibility cost of a shifting definition is severe and asymmetric. Show 120% NRR in Q1 and 108% in Q2, and if the drop turns out to be a definition change rather than a business change, the board will not discount that one quarter — they will retroactively discount every retention number you have ever shown them and demand independent verification thereafter. Diligence teams probe specifically for definition stability because they have been burned, and a company caught quietly redefining its metric mid-stream takes a real haircut on its multiple. The reverse holds too: walk in with a modest-but-honest 91% gross retention, a written definition, a clean reconciliation bridge to billing, and eight unbroken quarters of consistently measured history, and you earn trust that is worth more in both fundraising and exit than a flattering number that cannot survive scrutiny.

Public filings prove how much definitions vary even among elite companies, because there is no mandated definition of net revenue retention — each company discloses its own construction. Snowflake's famously high net retention is calculated on a trailing-twelve-month basis and is driven by consumption growth rather than seat expansion, which the company states explicitly. Datadog has sustained net retention well above 100% for many consecutive quarters, driven both by cross-sell across a broad product portfolio and by underlying usage growth. HubSpot's net retention sits materially closer to 100-110% than infrastructure peers, precisely because it serves a large SMB and mid-market base carrying the structural churn drag the benchmark section described — a useful example of an honest mid-market number rather than a flattering enterprise one. Monday.com discloses net dollar retention *segmented*, showing a meaningfully higher figure for its larger customer bands than for its blended base, a textbook public-market illustration of segment monotonicity. Identical metric name, materially different constructions. Never compare two companies' net retention without reading both footnotes.

Operationalizing the number in RevOps

A retention dashboard is not a chart. It is a governed, owned, monthly-cadenced system, and treating it as a chart is how an organization ends up with three conflicting retention numbers circulating simultaneously.

Required fields: starting ARR for the cohort at T0; ending ARR for that frozen cohort at T0+N; the full ARR walk (starting + new + expansion − contraction − churn = ending); gross and net retention with the cap and no-cap correctly enforced; logo versions of both alongside the dollar versions; all of it segmented by tier and ideally sliceable by product line, region, and acquisition channel; the Quick Ratio; the voluntary-versus-involuntary churn split with reason codes; and the documented variance bridge to billing.

Required cohort views: the headline annual-cohort number, the monthly-cohort trend line, the full cohort triangle, and the renewal-date cohort view the renewals team manages against.

Cadence: refresh monthly on a fixed, published close calendar, and only after billing reconciliation completes. The official number is never published before finance has tied out the ARR walk. A provisional mid-month view can exist for day-to-day operational use, provided everyone understands that the board-grade number is the monthly reconciled one.

Owner: retention reporting needs a single named owner, almost always in RevOps or Finance, who owns the definition document, runs the monthly reconciliation, publishes the dashboard, and can answer "where exactly did this number come from" without hedging. Shared ownership means, in practice, no ownership — and no ownership is exactly what produces divergent numbers. That owner also owns the definition document itself: a written, signed, version-controlled artifact specifying cohort construction, measurement window, explicit treatment of every edge case, the designated source of truth, and the reconciliation process. Any change to it is a versioned, dated, announced event, never a quiet edit.

Instrument the leading indicators, because both metrics are lagging by construction. By the time the annual cohort number lands, the churn happened and the renewal is lost. Product usage is the strongest predictor: depth of features adopted, breadth of active users, and above all trend — an account whose usage has slid three straight months is at real risk regardless of how warm the last business review sounded, because behavior leads sentiment. Health scores blending usage, support history, engagement, and contract data are useful only if the inputs are real and the composite has been validated against your actual churn outcomes; an unvalidated health score is worse than none, because it gets trusted while being wrong. Support trends — ticket volume, severity, time-to-resolution, and especially spikes in repeated or escalated tickets — predict churn early. Champion turnover is among the most reliable and most chronically underweighted signals: when the person who bought you and spent political capital on the decision leaves, renewal risk jumps sharply, because their replacement inherited a vendor they did not choose. Login recency matters too — an admin who has not logged in for thirty days, a sliding daily-active count. Also well-validated: chronically late invoice payment, business-review attendance, NPS and CSAT *trend* rather than absolute level, onboarding milestone completion for newer accounts, and the contract-value-to-usage ratio, where an account paying for far more than it consumes is a near-certain contraction at next renewal. Validate each signal individually against your own churn history so you know which ones are genuinely predictive for your business and segment rather than borrowed folklore.

Standardize one definition before the number ever reaches a board. A sloppy or shifting construction costs more credibility than a mediocre-but-honest figure ever will, and the correction is far more expensive than the discipline.

Related questions

What is a good gross retention rate?

It depends entirely on segment. Roughly 80-88% for SMB, 88-92% for mid-market, and 92-95%+ for enterprise, with best-in-class enterprise clearing 95-97%. An 85% figure is a crisis for an enterprise business and a solid result for an SMB one.

Can net revenue retention exceed 100%?

Yes, routinely. NRR includes expansion in the numerator, so when upsell, cross-sell, seat adds, and usage growth exceed contraction plus churn, the result clears 100% — that is net negative churn. Gross retention cannot exceed 100%; if yours does, it is a formula bug.

Should renewal discounts count as contraction?

Yes. If a customer renews at $90K having paid $100K, that $10K is contraction regardless of identical seat count or sales calling it a pricing concession. Exempting discounts inflates gross retention every period and teaches the sales organization to discount more freely.

How do I handle usage-based pricing in retention?

Pick one of four approaches and disclose it: trailing-twelve-month measurement, committed-spend floors as the base, a materiality threshold counting only sustained declines, or a separately labeled usage-retention metric. Raw point-in-time measurement turns ordinary consumption variance into phantom contraction.

Why do my CRM and billing retention numbers differ?

Timing on late-signed contracts, proration, mid-term amendments never written back to the CRM, credits and refunds, dunning outcomes the CRM cannot see, and multi-currency effects. Tie both ARR walks monthly, designate billing as the source of truth, and document the bridge.

FAQ

Do new customers signed during the period belong in the retention calculation?

No — never, in either the numerator or the denominator. Retention measures a frozen cohort of customers who existed at T0. A customer signed at T0 + 3 months belongs to the *next* period's starting cohort. Including new logos is the single most common way retention numbers get inflated, and it silently converts a retention metric into a net-growth metric that can read above 130% while the base genuinely shrinks by a third.

What is the difference between logo retention and dollar retention?

Logo retention counts customers, weighting a $2K account and a $2M account identically. Dollar retention weights by ARR. An 80% logo / 95% dollar split means you lost a fifth of your customers but they were all small — a small-account fit problem sitting on durable large-account revenue. The inverse, 95% logo / 80% dollar, is far more dangerous: you are keeping the small accounts and losing the whales. Report both for gross and net — four numbers, not one.

Does a win-back count as expansion?

No. A previously churned customer who returns generates new or win-back ARR, not expansion of a retained account — the account was not retained, it left and had to be re-acquired at cost. Folding win-backs into expansion inflates net retention and hides that acquisition spend went twice into the same logo. Equally, do not back-date a win-back to erase the original churn; the churn happened at the measurement boundary and stands.

Should price increases count as expansion?

Both positions are defensible; silence is not. Purists exclude them because expansion should reflect customers choosing more value, not you charging more for the same thing. Pragmatists include them because ARR genuinely rose and the customer chose to stay at the higher price, which is real pricing power. Most sophisticated companies include them but disclose separately: "NRR was 116%, of which 4 points was a list-price increase." Apply the same separate-disclosure treatment to FX gains and contractual escalators.

How often should retention be recalculated?

Refresh monthly on a fixed, published close calendar — but only after the billing reconciliation is complete, so the official figure never gets published before finance has tied out the ARR walk. Keep a provisional mid-month view for operational use if the team understands it is not board-grade. Annual cohorts are the external headline; monthly cohorts are the leading indicator; a divergence of more than a couple of points between the two is itself a signal about renewal concentration.

Which metric do investors care about more?

They want both, and a company volunteering only net retention draws suspicion. Diligence teams typically pull gross retention first because it cannot be dressed up by a handful of expanding accounts, then look at what expansion added on top. What earns the most trust is neither number in isolation but a written definition, consistent application across at least eight quarters, and a clean reconciliation bridge to billing.

Sources

• SaaS Capital — Retention benchmarks and private SaaS survey research
• Bessemer Venture Partners — State of the Cloud and cloud metrics benchmarks
• OpenView Partners — SaaS benchmarks resources
• KeyBanc Capital Markets SaaS Survey coverage
• ChartMogul — Net revenue retention and cohort analysis guides
• Stripe — Revenue recognition and subscription metrics documentation
• Maxio — SaaS metrics and retention reporting resources
• a16z — 16 startup metrics
• Salesforce — CPQ and contract management documentation
• HubSpot — Recurring revenue and subscriptions documentation

Related on PULSE

• How do I build a renewal forecast that finance will actually trust?
• What belongs in a customer health score, and how do I validate it?
• How should I structure the ARR walk for a monthly board packet?
• How do I reduce involuntary churn with dunning and payment recovery?
• What is the right customer-success coverage ratio by segment?
• How do I design pricing and packaging that creates expansion headroom?