Growth Metrics: What to Track, Formulas, and Benchmarks

A.Five categories, one underlying shape

Unlike vanity metrics, which count activity without connecting it to an outcome, a growth metric puts a business quantity over a business denominator: revenue over a month, cost over a customer, activated users over signups, revenue retained over revenue started with. Five denominators cover the field.

Revenue growth. For subscription businesses, MRR is the base: predictable revenue from active subscriptions, with ARR as twelve times it for investor conversations. The headline is close to useless undecomposed. Net new MRR equals new plus expansion minus churned minus contraction, and the same net figure comes out of very different businesses. Growth rate sits on top of it, month over month for early companies where growth compounds fast, year over year for mature ones because it smooths seasonality. Roughly 20 to 30% year on year is the band that reads as healthy to investors, with top-quartile performers above 50%.

Acquisition. New customers per period, and what each one costs. CAC is marketing plus sales plus overhead over customers won, and the blended version can look healthy while one channel quietly burns money, so it is only meaningful segmented by channel. The viral coefficient sits alongside it: how many new users each existing user brings, where very few products clear 0.5 and even 0.2 meaningfully reduces effective CAC.

Activation. Activation rate is the share of new signups who complete the actions that indicate they reached core value, and time to value is how long that took. The definition of the activation event is yours; the requirement is that it correlates with long-term retention in your own data. Activation gates every downstream number. A user who never reaches the value moment cannot retain, expand, or refer, so a low activation rate caps retention and LTV no matter what you do later.

Retention. The share of customers still active over a period, plus net revenue retention, which compares a cohort’s revenue at the end of a period against its start including expansion, contraction, and churn. The compounding is the part people miss: 95% monthly retention sounds excellent and means losing 46% of customers in a year.

Efficiency. Whether the economics support the rate you are pursuing. LTV:CAC asks whether a customer returns more than they cost to win. CAC payback asks how many months of gross margin it takes to recover that cost. Burn multiple, for venture-backed companies, is net burn over net new ARR.

B.They resolve at different speeds, and the ordering is inverted

Line the five up by when you learn them. CAC is known the day the invoice clears. Activation rate resolves within a week or two of a signup cohort. Retention needs 90 days to show a shape. Net revenue retention and CAC payback need a full billing cycle and usually a year. Burn multiple needs four quarters to be anything other than noise.

Now line them up by how much they tell you about the business rather than about your own choices. CAC is largely an input you control: bid less and it falls, bid more and it rises, and neither movement says anything about whether the product is working. Growth rate is partly a function of how much you spent. Retention and NRR are outputs you discover, and you cannot buy either of them this quarter. The two orderings run against each other, which is why the metrics that arrive first are the ones least able to tell you whether to keep going.

Even the day-one numbers are shakier than they look. A growing share of first touches now happens inside ChatGPT, Claude, Gemini, and Perplexity rather than on a search results page, and those visits land as direct or referral traffic, so the channel is invisible in a standard acquisition report while it sends buyers. The LLM Acquisition report isolates the real humans arriving from an assistant and follows them to signup and revenue, which is the only way that traffic gets judged on the same terms as paid search. Until a channel is separable, its CAC is being paid by whichever channel it hides inside.

A.Retention is the one measurement you cannot buy

Aggregate retention blends loyal old cohorts with unproven new ones and describes neither, which is the entire reason cohort curves exist. Separate them and the two questions come apart: how good is the base you already have, and is the product getting better at keeping the people arriving now.

The flattening is the signal, not the height. A curve that settles at 40% says a defined group has found permanent use for the product. A curve that keeps declining at a steady slope says nobody has, and it is a product-market fit problem wearing a retention problem’s clothes. NRR carries the same information in revenue: past 100%, the existing base grows without a single new signup.

None of that is purchasable inside the measurement period. You can halve CAC by narrowing targeting and double the growth rate by opening the budget, and both show up next month. A cohort curve refuses to move until the product changes and a new cohort has aged, which is precisely why it is worth organising around. Our cohort analysis guide covers how to cut them.

B.But the numbers you trust most are estimates, and the benchmarks can be gamed

The counterweight is that retention and efficiency metrics are the ones most easily improved by doing less. Stop spending on the marginal channel and CAC payback shortens, LTV:CAC rises, and NRR climbs because the low-intent cohorts that were diluting it have stopped arriving. Every one of these reference points improves for a quarter in a company that has simply slowed down.

Two of the three are also forecasts wearing the clothes of measurements. Lifetime value is a projection of revenue that has not happened, so LTV:CAC inherits every assumption in the churn model underneath it. Payback is the more honest of the pair because it is built from realised gross margin, which is why it tends to be the one that survives contact with a finance team.

Activation is softer still, and in a specific way worth naming: the rate is a function of a definition you chose. Move the activation event one step earlier and the rate jumps with nothing having changed. The discipline is to fix the definition empirically by comparing 30-day and 90-day retention for activated against non-activated cohorts in your own data, then leave it alone. A definition that gets revised whenever the number disappoints is not a metric.

A.Stage picks the trade, not the metric list

Before product-market fit, the only question is whether anyone wants this, so activation and retention are the whole dashboard. Optimising CAC here is optimising the cost of attracting people to a product they will not keep, and the north star metric should be whatever most directly captures users reaching value. The trade you accept is that you have no reliable efficiency read at all, and you should stop pretending otherwise in board decks.

While scaling, the question becomes whether growth can be bought efficiently, so acquisition rate, CAC by channel, growth rate, and LTV:CAC move to the front. This is the stage where the inversion does the most damage, because every fast metric is now available and the slow ones are still maturing. The trade is that you are steering on numbers you can move directly, so the cohort curve has to stay on the same screen as a check.

At maturity the question is whether the growth is worth what it costs: NRR, expansion revenue, CAC payback, burn multiple. A company at 130% NRR grows 30% a year without acquiring anyone, which makes acquisition spend optional rather than existential. The trade here is speed, because every number that matters at this stage is a year old by the time it is trustworthy, so each needs a leading indicator beside it and activation is usually the best available.

B.The dashboard is a decision device, not a display

Five to seven metrics, chosen for the current stage, with everything else pushed into drill-down reports. Our guide to picking the right KPIs covers the selection. Each one carries at least twelve weeks of history, because a number alone says nothing and the same number with a direction and a rate of change is a decision. Each has an owner and a threshold that names an action, not a colour. The purpose of a growth dashboard is not to generate reports. It is to trigger decisions, and a dashboard reviewed weekly and never acted on is decoration.

This used to fail because building each metric was a data-team ticket, so dashboards ossified at whatever shipped a year ago and the metric set stopped tracking the stage. In KISSmetrics you describe the metric in the chat (“weekly net new MRR split by plan”, “90-day retention for signups from paid search”) and it builds the metric and the report behind it, with autocapture and an autoconfiguration pass handling the tracking rather than a hand-instrumented event list.

Which introduces the opposite failure. When adding a metric is free, the dashboard becomes twenty metrics unless somebody owns deletion. The discipline that mattered when each chart cost a sprint still matters now that it costs a sentence: if a metric has not changed a decision in a quarter, it goes into the drill-down and something with a shorter lag takes its place.