SaaS Churn: Cohort vs Aggregate vs Net Dollar Retention Math

Churn is a single word that hides at least three different calculations, each of which can be defended as correct in isolation and all of which produce different numbers from the same dataset. Reporting one without naming which is the most common source of pitch-deck overstatement in B2B SaaS metrics.

The piece below covers the three calculations in order of investor utility, the worked example that shows where they diverge, and the five mistakes that show up repeatedly when bootstrapped founders graduate to investor conversations.

Three numbers, three different stories

Aggregate churn looks at the whole customer base in a single window. Cohort churn slices the base by signup month and tracks each slice independently. Net dollar retention asks the dollar question: of the revenue you had a year ago from a fixed set of customers, how much do you still have today, including upgrades, downgrades, and cancellations.

The three numbers can move in opposite directions inside the same business. NDR can rise to 115% while logo churn deteriorates from 4% to 6% monthly, because expansion from the surviving cohort outpaces logo loss. A board that sees only NDR misses the customer-acquisition problem; a board that sees only logo churn misses the expansion engine. Both views matter, and they answer different questions.

1. Aggregate churn (and why it lies)

The standard aggregate formula:

Aggregate monthly churn = Customers lost in month / Customers at start of month

The structural problem is the denominator. A SaaS at 200 customers that adds 40 in month one and loses 8 by month-end reports 8 / 200 = 4% churn. The 8 lost customers, however, were not drawn evenly from the 200. They were drawn disproportionately from the most recent signups, which always exhibit higher month-one drop-off than mature cohorts.

Aggregate churn averages a high-churn early-cohort population with a low-churn mature-cohort population and produces a single number that does not describe either. Founders reporting flat aggregate churn during periods of fast new-customer growth are usually masking deteriorating early-cohort retention with the dilution of fresh signups in the denominator.

The Fader and Hardie probability-of-retention work shows that cohort retention curves follow predictable shapes well-fit by Beta-Geometric distributions, where month-one drop-off is consistently the steepest segment^[4]. Aggregating across vintages flattens the curve and obscures the most actionable signal in the data.

When aggregate churn is acceptable: stable customer base, low growth rate, no cohort-quality changes between periods. The conditions under which aggregate churn is acceptable are also the conditions under which the metric is uninformative.

2. Cohort churn

Cohort churn pins each signup month to its own retention curve and reports the curve shape, not a single point estimate.

Calculation procedure:

• Group customers by signup month (the cohort).
• For each cohort, count how many remain active in month 1, month 2, month 3, and so on.
• Express each as a percentage of the cohort starting size.
• Plot the curves on the same axis to compare cohort quality across time.

What the curve shape reveals: a healthy SMB SaaS cohort retains 70 to 80% of signups through month one, 60 to 70% through month three, and 40 to 50% at month twelve, with the decay rate flattening after month six. Mid-market cohorts (ACV above $25k) retain 90%+ through month one and 75%+ at month twelve per ChartMogul 2024 data^[3]. Sharp month-one drop is an onboarding problem; a steep month-three drop is a value-realization problem; a steady month-twelve decline is a price-vs-alternative problem.

Cohort churn rates feed the right LTV calculation. The CLV Calculator uses a single churn rate by design, so the input has to be the cohort-stabilized churn (typically the month-six-to-twelve average), not aggregate churn. Plugging aggregate churn into an LTV formula produces a number 30 to 60% too high in most SaaS data, which is the single biggest source of LTV overstatement.

The minimum honest cohort report has three vintages: the most recent fully-completed cohort, a six-month-old cohort, and a twelve-month-old cohort. Three points define a curve well enough to spot improvement or deterioration trends.

3. Net dollar retention

NDR is the dollar version of cohort retention with expansion added back. The formula:

NDR = (Starting MRR + Expansion − Contraction − Churn) / Starting MRR

The starting MRR is the recurring revenue from a fixed cohort of customers, measured at a starting date. Twelve months later, the same cohort produces some new MRR figure: some customers churned (their MRR is gone), some downgraded (contraction), some upgraded or added seats (expansion). NDR is the ratio of ending to starting MRR for that fixed cohort.

NDR above 100% means the cohort is generating more revenue today than it was a year ago, even after losing some customers entirely. Bessemer 2024 Cloud Index reports public-cloud SaaS median NDR of 100%, top-quartile 110%, with the best-in-class data-infrastructure firms (Snowflake, MongoDB, Datadog) holding 120 to 130%^[1]. OpenView 2024 SMB-segment medians come in lower, around 95 to 102%, reflecting weaker expansion mechanics in self-serve products^[2].

Why NDR is the investor-facing number: it captures the compounding revenue engine. A SaaS at 110% NDR doubles its starting cohort revenue every 7.3 years with zero new customer acquisition. That is the specific dynamic that turns a SaaS from a high-CAC growth-at-all-costs business into a self-funding compounding asset, and the only metric that exposes it is NDR.

The expansion side of NDR is where the CAC math reverses sign: dollars spent on customer success and onboarding flow into expansion revenue with effective payback periods often below 6 months, an order of magnitude better than new-acquisition CAC payback.

Logo churn vs revenue churn

Logo churn counts customers. Revenue churn counts dollars. The two diverge whenever customer ARPU varies, which is most of the time.

A SaaS with 100 customers averaging $50 ARPU loses 5 customers in a month: 5% logo churn. If the 5 lost customers were on the $50 plan, revenue churn is also 5%. If three of them were on a $200 enterprise plan and two were on the $20 starter plan, revenue churn is ($600 + $40) / $5,000 = 12.8%. Same logo churn, very different revenue impact.

Investors evaluating mid-market and enterprise SaaS focus on revenue (gross dollar) churn because customer count is a less faithful indicator of business health when ARPU is bimodal. Investors evaluating SMB self-serve focus on logo churn because ARPU is tighter and customer count is the better proxy for product-market fit.

Honest reporting always shows both: logo churn for product health, revenue churn for financial health, NDR for the full revenue cohort picture.

One additional dimension worth tracking explicitly: gross dollar retention (GDR), which is NDR with expansion stripped out. The formula:

GDR = (Starting MRR − Contraction − Churn) / Starting MRR

GDR has a hard ceiling at 100% (you cannot retain more than what you started with) and represents the floor of revenue durability. NDR can mask weak GDR if expansion is strong: a SaaS at 110% NDR with 78% GDR has a 32-point expansion engine offsetting a structurally leaky base. The same SaaS at 110% NDR with 95% GDR has a healthy base with moderate expansion. Same headline, very different operational reality. Investors increasingly ask for GDR alongside NDR for exactly this reason.

Worked example: same data, three results

One product, twelve months of data, three churn calculations. A B2B productivity SaaS at $50k starting MRR, 250 customers averaging $200 ARPU.

Starting state (Jan 1):
  Customers      250
  MRR            $50,000
  Avg ARPU       $200

Activity over 12 months:
  New customers gained        +180  (avg ARPU $180)
  Customers churned (logo)    -55
  Customers downgraded         15   (lost $9,000 MRR)
  Customers expanded           60   (gained $14,400 MRR)

Ending state (Dec 31):
  Customers      375
  MRR            $87,400

CALCULATION 1: Aggregate monthly churn
  Total months: 12
  Avg customers in window: ~310
  Logo churn rate: 55 / 310 / 12 = 1.48% monthly
  Annualized: 17.7%

CALCULATION 2: Cohort churn (Jan 1 cohort only)
  Starting cohort: 250 customers
  Still active Dec 31: 195 customers
  12-month logo retention: 195 / 250 = 78%
  Equivalent monthly churn: 1 - 0.78^(1/12) = 2.04%
  Annualized: 22%

CALCULATION 3: Net Dollar Retention (Jan 1 cohort)
  Starting MRR (Jan 1 cohort): $50,000
  Lost to churn:                -$11,000
  Lost to contraction:           -$9,000
  Gained from expansion:        +$14,400
  Ending MRR (Jan 1 cohort):    $44,400
  NDR: $44,400 / $50,000 = 88.8%

Three numbers from one dataset: 1.48% aggregate, 2.04% cohort-true, 88.8% NDR. The aggregate rate looks great. The cohort rate shows the underlying retention is 38% worse than the aggregate suggests. The NDR shows the cohort lost real revenue net of expansion, which means this business is acquiring its way out of a leaky-bucket problem.

An LTV calculation built on the aggregate 1.48% reports an LTV of $200 / 0.0148 = $13,514. The same calculation built on the cohort-true 2.04% reports $9,804. That is a 38% LTV inflation built entirely on choice of churn metric, before any other adjustment.

Which number an investor actually looks at

Investor metric priorities by stage and segment, drawn from Bessemer and OpenView 2024 benchmarks^[1][2]:

• Pre-seed and seed (any segment): cohort retention curves matter more than NDR because expansion mechanics have not had time to compound. A 60%+ month-six retention is the bar.
• Series A SMB SaaS: NDR above 95% and gross dollar retention above 85% are the investor-facing thresholds. Below those, the ARR growth story collapses regardless of new-acquisition pace.
• Series A mid-market and enterprise: NDR above 110% is the bar that separates fundable from unfundable in the public-comp world. 110%+ NDR + 130%+ growth = top-decile valuation multiple per Bessemer Cloud Index.
• Late-stage and public: NDR is the headline metric in every quarterly report. Snowflake, Datadog, MongoDB consistently report 120%+ NDR; Salesforce and HubSpot report 100 to 105%; the multiple difference is partly a function of that gap.

The implicit ranking: NDR is the headline, gross dollar retention is the floor, cohort logo retention is the leading indicator. Aggregate churn rarely appears in investor materials because the metric is too easy to mask growth-driven dilution behind.

Annualizing churn correctly

The most common arithmetic mistake in churn reporting is converting between monthly and annual rates with linear math. Linear conversion overstates annual churn at every monthly rate above zero, with the error growing as the rate grows.

The correct conversion uses geometric compounding:

Annual churn = 1 − (1 − monthly churn)^12

The opposite direction:

Monthly churn = 1 − (1 − annual churn)^(1/12)

Numerical examples showing the error magnitude:

• Monthly churn 1.0%: linear annual = 12.0%, geometric annual = 11.4%. Error: 0.6 points.
• Monthly churn 2.0%: linear annual = 24.0%, geometric annual = 21.5%. Error: 2.5 points.
• Monthly churn 3.0%: linear annual = 36.0%, geometric annual = 30.6%. Error: 5.4 points.
• Monthly churn 5.0%: linear annual = 60.0%, geometric annual = 46.0%. Error: 14.0 points.

At low rates, the error is small enough to ignore. At rates above 3% monthly, the linear conversion is wrong by amounts that change LTV calculations meaningfully. The fix is one cell in the spreadsheet, and consistency across reporting periods.

The same compounding logic applies inside cohort retention: a "12-month retention of 78%" is equivalent to a 2.04% monthly geometric churn rate, which is the rate to plug into the LTV formula. Working from cohort retention curves to monthly churn is straightforward; working from a single point estimate of "monthly churn" to a 12-month retention number requires the geometric conversion that linear math gets wrong.

Common reporting mistakes

• Reporting one number without naming which. "Our churn is 3%" is unparseable until clarified as monthly or annual, logo or revenue, aggregate or cohort. The honest report names all four dimensions.
• Using aggregate churn in LTV calculations. The single biggest source of overstated LTV in pitch decks. The fix is one cohort table away.
• Conflating gross retention with NDR. Gross dollar retention strips out expansion and reports only churn plus contraction. A business at 95% gross retention and 110% NDR has 15 percentage points of expansion compounding the cohort. Reporting them interchangeably collapses the most important diagnostic in the data.
• Excluding free-to-paid conversions from cohort definitions. A cohort defined as "first paid month" excludes the trial-period attrition that explains most month-one drop-off. Some founders report cohort retention from first paid month and a separate trial-conversion rate; others bundle them. Pick a convention and disclose it; the comparability collapses without it.

Picking the right churn calculation is not a stylistic choice. Each of the three metrics answers a different question, and the gap between them is where most LTV inflation hides. The cleanest reporting practice is showing all three side by side in any document that touches an investor, an acquirer, or a board, and naming the calculation explicitly every time the word "churn" appears.