NDR above 120% is almost never NDR.

The single most-claimed metric in the inbox — and the one most often mis-defined. A short field test for whether the number actually means what the deck implies.

01The most-quoted, least-defined metric

Net dollar retention is the most-quoted single metric in the seed-to-Series-A inbox. It is also the least-defined. In our corpus of 1,142 dossiers, 184 founders claimed an NDR figure above 120%. We reconstructed each claim against the underlying revenue data. Of the 184, 52 — roughly 28% — reconciled to the figure the founder presented. The remaining 132 used a definition of NDR that, when applied across the cohort, produced a smaller number than the headline.

This is not a story about founder dishonesty. The structural problem is that NDR has six commonly-used definitions, no industry-wide convention has emerged for which one is canonical, and every definition produces a different number from the same underlying data. The founder picks the definition that produces the most favourable number, often without realising the alternative exists. The IC, having no way to know which definition was used, defaults to the most generous interpretation and prices the round accordingly.

The reason 120% is the threshold this note flags is that 120% is the floor below which the metric is generally not worth highlighting. A growth-stage company quoting 105% NDR has a metric worth defending; a company quoting 102% probably has revenue retention slightly below the cohort median and is using the metric because there is no stronger one available. The threshold matters because founders escalate the claim past 120% precisely when they need a stronger metric than the data supports. The escalation is where the mis-definitions appear.

02The four common mis-definitions

Four definitions account for almost all the variance in our reconstruction. Each appears regularly. None is technically incorrect — they are all defensible interpretations of what NDR could mean. The diligence is which one the founder actually used.

• Mis-definition one — including new logos in the numerator. The most common error. NDR is supposed to measure the retention of an existing cohort — start-of-period customers only — without adding new logos to the numerator. Including new logos turns NDR into something closer to total revenue growth, which is a different metric. A company adding new customers fast enough produces a 130% NDR claim that disappears entirely when the new logos are removed from the numerator.
• Mis-definition two — contracted, not realised. The founder quotes NDR using booked ARR at the contract anniversary, including expansion clauses the customer has signed but has not yet been billed for. The customer's contracted upgrade in month eighteen, scheduled to take effect in month twenty-four, counts in the month-eighteen NDR figure. Realised NDR — what the customer actually paid — comes in lower because contracted expansions slip, get renegotiated, or hit the end of the prior contract before the upgrade activates.
• Mis-definition three — survivor-biased cohort. The founder reports NDR only for the customers who renewed, excluding those who churned. The 130% claim is true of the renewing cohort. It is not the NDR of the original cohort. Including churned customers — at $0 of retained revenue — brings the number down materially. This is the easiest mis-definition to make without realising; the cohort the analytics tool returns is often already churn-filtered.
• Mis-definition four — services revenue mixed with subscription. A company that bills $500K of subscription and $200K of professional services to a customer in year one, and $600K of subscription only in year two, has 86% NDR on subscription and 86% on total. The founder occasionally calculates 120% by treating year-two subscription against year-one subscription as the comparison and ignoring the services revenue that did not recur. Defensible if disclosed; misleading if not.

The pattern across the four: each mis-definition is internally consistent and produces a real number. The number it produces is not the number the IC needs to make a comparable judgement against the rest of the cohort.

03The field test

We can disambiguate the four mis-definitions in roughly thirty seconds of conversation with the founder. The field test is three questions in sequence.

Question one. “In the cohort you used for the NDR calculation — were new logos in the period included or excluded?” Founders using mis-definition one answer with hesitation; the question is unfamiliar. Founders using a clean definition answer immediately. If the founder says “included,” the metric is not NDR; it is revenue growth. The number has to be reconstructed before the IC can use it.

Question two. “Is the dollar number contract-recognised or realised?” Founders using mis-definition two will defend the contracted view; some will acknowledge the gap to realised. The diligence is whether the founder has the realised number available — if they do, the conversation can move on. If they do not, the NDR claim is conditional on contracted expansions activating as scheduled, which is a different kind of claim.

Question three. “Does the denominator include customers who churned during the period?” Founders using mis-definition three almost always answer “the renewing cohort” or “customers who were still active.” The right answer is that the denominator should include all customers active at the start of the period, with churned customers entering the numerator at $0. If the founder cannot recompute the metric on the full cohort, the headline number is overstated.

A fourth question is sometimes worth adding for companies with services revenue: “Is the calculation subscription-only or all-revenue?” The answer should match the metric label. A founder reporting “NDR” using mixed revenue is implicitly claiming the services revenue was recurring, which it usually is not.

Most founders pass three out of four. The pattern is durable across our corpus.

04Reconciling to the right number

After the field test, the diligence is to reconstruct the metric using the canonical definition and present it back to the founder. The canonical definition we use, and which we recommend the IC use, is the one most often referenced in operator memos at growth-stage:

NDR = (start-of-period revenue + expansion within cohort + reactivation within cohort − churn within cohort − contraction within cohort) ÷ start-of-period revenue.

The cohort is the set of customers active at the start of the period. New logos are excluded. The numerator is calculated against realised revenue, not contracted. Services revenue is excluded if the company reports subscription-only metrics; included if the company reports blended.

When the canonical reconstruction returns a materially smaller number than the headline, we present both numbers in the dossier — headline alongside reconstructed, with a one-paragraph note on which definition difference produced the gap. The founder is given a chance to comment in the dossier before delivery.

Two patterns appear consistently in the reconciliation. The first is that companies in vertical SaaS, healthcare, and regulated finance often have legitimate NDR above 120% on the canonical definition because their customers have low churn and structural expansion paths — more seats, more workflows, more sites. The second is that companies in horizontal SaaS, developer tools, and consumer-adjacent B2B almost never have canonical NDR above 110%, and when they claim it, the gap to reconstructed is usually largest.

The diligence-positive signal is not the NDR number. It is the founder's willingness to walk through the reconstruction with the IC. Founders who can do this in real time, on the call, almost always have a defensible metric — they may be using a non-canonical definition, but they understand the math and can produce the canonical version on request.

05Why founders aren't lying — and why it still matters

The 132 cases where the headline NDR did not reconcile to the canonical definition are not, in most cases, intentional misrepresentations. Three things produce the pattern.

First, the founder learned the definition from the same source as the next founder, and that source — usually an investor memo, a SaaStr post, an analytics tool's default calculation — used a non-canonical formula. The founder is reporting honestly against the definition they were taught.

Second, the analytics tool the company uses returns NDR using its own internal definition, often the wrong one. ChartMogul, Stripe Sigma, Mixpanel, Amplitude, and the major SaaS metrics dashboards each have their own NDR formula and the founder takes the number the tool returns.

Third, the comparison set the founder is benchmarking against is also using non-canonical definitions. The founder hears that the cohort median NDR for Series A SaaS is 120%, picks a definition that produces 120%, and reports it. The benchmark is wrong, the founder's number is wrong, but the founder's number is consistent with the benchmark — which feels like validation.

The reason the IC has to reconstruct anyway is that the reconstructed number is the one that survives contact with the operating data. A round priced against 130% NDR that turns out to be 95% canonical NDR is priced on a different business. The founder did not lie. The price is still wrong. The work of the diligence is to surface this before the term sheet, not after the first year of operating data invalidates the model.

06How we surface the gap

Every dossier with an NDR claim above 110% receives a reconstructed number. The reconstruction lives in section four — Financial Forensics — alongside the burn reconciliation and the concentration ratios. The format is intentional.

We publish both numbers, headline and reconstructed, with the formula we used for each. The founder is given the dossier in draft form for comment before delivery. Most founders accept the reconciliation; a small fraction push back on the formula and propose an alternative. We accept the alternative if it is defensible and re-run the reconstruction. The dossier reflects whichever formula the founder is willing to commit to in writing.

Where the gap is material — more than 15 points between headline and reconstructed — the dossier escalates the NDR claim to a kill flag in the closing section. See Note 05, flag ten. The flag is recoverable: the founder restates the metric using the canonical definition and the IC continues the diligence. The flag escalates only if the founder defends the original claim past the reconstruction, which happens rarely but does happen.

The point of the exercise is not to embarrass the founder. The point is to make sure the IC and the founder are using the same number when they negotiate the round. A misaligned definition produces a price the founder and the IC will later disagree about, which is the worst possible outcome — worse than a lower price both parties commit to up front.