MRR and the Arithmetic of Recurring Revenue

A run rate, not revenue earned

MRR answers one question: if nothing at all changed from today, how much would the subscriptions currently on the books bill in a month? It is a snapshot of a rate, in the way a speedometer is a snapshot of a rate. It does not tell you how far you travelled, and it is not the revenue figure your accounts will report. Under IFRS 15 and ASC 606 revenue is recognised as the performance obligation is satisfied, which for a subscription means spread evenly over the service period regardless of when the customer paid. MRR ignores cash timing and ignores the accounting calendar; it asks only what the contracts are worth per month right now.

That distinction is not pedantry, because the two numbers diverge violently whenever the business is moving. A company at $100,000 of MRR growing 5% a month quotes an ARR of $1,200,000 today. Over the next twelve months it will actually bill $1,671,298 — 39.3% more than the run rate it is quoting — and it will exit the year at an ARR of $2,155,028. Neither of those three figures is wrong; they answer different questions. Trouble starts when a board pack uses one of them and a cash-flow forecast uses another.

Normalising: where the number actually goes wrong

Every contract has to be converted to a monthly equivalent before it can join the total, and there are exactly three places this goes wrong. First, annual and multi-year contracts. A customer who pays $12,000 up front for a year adds $1,000 to MRR, not $12,000 — the cash arrived in one month, the obligation spans twelve. Booking it as $12,000 makes the month look extraordinary and then makes the next eleven look like a collapse. Second, one-off fees: implementation, migration, training, professional services, hardware. None of them recur, so none of them belong in a recurring-revenue figure, however genuinely they were earned.

The third is discounts and trials. A customer on a 50% first-year discount is worth what they actually pay, not the list price, and a free trial is worth zero until it converts. Adding list prices to MRR because "they will renew at full rate" turns a forecast into a reported metric.

The reason to be strict is that ARR is almost always quoted as MRR × 12, so every normalisation error is multiplied by twelve on its way to the headline. Leave a single $9,000 implementation fee in a $100,000 MRR month and you report $109,000 of MRR and $1,308,000 of ARR. The overstatement is not $9,000 — it is $108,000, twelve times the mistake, and it will still be sitting in the comparison base a year later.

The movement analysis: five components, two net figures

A single MRR figure is inert. What explains it is the movement between two month-ends, split into components that cannot overlap. New MRR comes from customers who were not there last month. Expansion comes from existing customers paying more — upgrades, seats added, usage tiers crossed. Contraction comes from existing customers paying less while staying. Churn comes from customers who left entirely. Reactivation, if you track it, is a fifth bucket for customers who left and came back; most companies fold it into new, which is defensible as long as they say so.

Two net figures come out of those components, and they are not the same thing. Net new MRR is new + expansion − contraction − churn: in our month, 12,000 + 6,000 − 4,000 − 6,000 = $8,000, which takes MRR from $100,000 to $108,000. Net revenue retention deliberately drops the "new" term, because it is a question about the customers you already had: NRR = (opening + expansion − contraction − churn) ÷ opening = (100,000 + 6,000 − 4,000 − 6,000) ÷ 100,000 = 96%. Gross revenue retention drops expansion too, so it can never exceed 100%: (100,000 − 4,000 − 6,000) ÷ 100,000 = 90%.

Growing 8% while the installed base shrinks 4%

Read the two net figures side by side and the month tells a different story from the headline. MRR is up 8%, which any dashboard will render in green. The installed base — the $100,000 that was already there — ended the month at $96,000. Those customers, collectively, cancelled or downgraded $10,000 and expanded by $6,000. They went backwards by 4% while the company as a whole went forwards by 8%, and the entire difference is the $12,000 of new business bought that month.

Compound that and the shape of the business appears. A monthly NRR of 96% is not "nearly 100" — it is 0.96 raised to the twelfth power, which is 61.27%. A base of $100,000 with no new sales at all would be worth $61,271 in a year and $23,002 in three. And because the leak is proportional while the new business is a roughly fixed monthly amount, MRR does not grow forever: it converges. With $12,000 of new MRR a month and a 4% monthly net loss, MRR settles at $12,000 ÷ 0.04 = $300,000 and stops. The trajectory is $177,458 after a year, $224,917 after two, $253,996 after three, $282,730 after five — visibly flattening against a ceiling nobody put in the plan.

Above 100%: growth without a single new customer

Net revenue retention above 100% means expansion from existing customers exceeds everything those customers cancelled or downgraded. The installed base grows on its own. Freeze acquisition entirely and revenue still rises — which is why NRR is the single most consequential number in a subscription business, and why compounding comes from retention rather than acquisition. Over three years on a $1,200,000 base, with not one new customer: 90% annual NRR leaves $874,800; 100% leaves $1,200,000; 105% gives $1,389,150; 110% gives $1,597,200; 120% gives $2,073,600; 130% gives $2,636,400.

Put as doubling times, it is starker still. On retention alone, a 110% NRR doubles the installed base in 7.27 years, 115% in 4.96 years, 120% in 3.80 years and 130% in 2.64 years — with the sales team doing nothing. The same arithmetic run below the line is what makes low-NRR businesses exhausting: at 96% monthly the acquisition machine spends its entire output refilling a bucket. This is why a CAC payback figure has to be read alongside NRR rather than instead of it; we work the payback arithmetic itself in a separate article.

Why ARR is not MRR × 12

MRR × 12 is an annualised run rate: what the next twelve months would produce if the month you just closed repeated identically twelve times. That is a legitimate quantity, and in a flat month it is a fair proxy for annual revenue. In any month where the mix changed, it is not. A deal signed on the 28th contributes almost nothing to the month's cash and its full monthly value to the closing MRR, so multiplying by twelve projects it forward as though it had always been there. A churn event on the 2nd does the same in reverse. Neither is dishonest; both mean the annualised figure describes the exit rate, not the year.

There is a second, sharper reason in businesses selling annual contracts. Committed ARR — the sum of contracted annual values, with renewal dates attached — is a different object from ending MRR × 12, because renewals cluster. If a quarter of your book renews in January, January's outcome moves the annual figure far more than any other month's, and an annualised run rate taken in December quietly assumes that renewal has already happened. When the two measures disagree, the honest move is to publish both and say which one the forecast used.

Frequently asked questions

Should an annual contract be counted as MRR at all?

Yes, at one twelfth of its annual value. A $12,000 annual contract is $1,000 of MRR for each of the twelve months it covers, whatever the payment schedule. The cash arrives in one month and belongs in the cash-flow statement; the recurring value spreads across twelve and belongs in MRR. Companies that sell only annual contracts often track committed ARR directly instead, which avoids the conversion but makes renewal timing the thing to watch.

What is the difference between net revenue retention and net new MRR?

Net new MRR includes new customers; net revenue retention deliberately excludes them. Net new MRR is new + expansion − contraction − churn and reconciles opening MRR to closing MRR. NRR is (opening + expansion − contraction − churn) ÷ opening and answers a narrower question: what happened to the customers you already had? In our worked month net new MRR was +$8,000, an 8% rise, while NRR was 96% — a 4% fall. Both are true; only the second one tells you whether the product is holding.

Can net revenue retention be above 100% while customers are leaving?

Yes, and it frequently does. NRR is a revenue measure, not a headcount measure: a handful of large customers expanding can more than offset a long tail of small ones cancelling. That is why gross revenue retention should be read next to it. GRR strips out expansion and can never exceed 100%, so the gap between the two tells you how much of your retention story is upsell covering a leak. In our month, NRR 96% against GRR 90% says expansion was recovering six of the ten points lost.

Do usage-based charges belong in MRR?

Only the part that genuinely repeats. A committed monthly minimum is recurring and belongs in MRR at its committed value. Variable overage on top is not contracted and swings with the customer's activity, so most companies either exclude it or include a trailing average and say which. The one thing not to do is include a spike month at face value: it will be annualised by twelve and then treated as the base the following month, producing a contraction that never happened.

Is a high NRR enough on its own?

No, for two reasons. A high NRR built entirely on price increases rather than usage growth is borrowing from next year's renewals, and it will show up as contraction later. And NRR says nothing about how much you paid to acquire the base in the first place — a business can retain beautifully and still never earn back its acquisition cost. Read NRR alongside the payback period on customer acquisition cost, and treat the two as one question rather than two.