PLG metrics: Why benchmarks fail, and what to track

Take activation rate. It is the share of signups who reach a meaningful milestone, and both halves of that are yours to define.

You decide what counts as a signup: an email address, a verified account, a completed profile. You decide what counts as activated. Require email verification and your activation rate rises, because the denominator lost everybody who never clicked. Loosen the signup form and it falls, while the business improves.

So two companies with identical products can report activation rates thirty points apart, and a table comparing them across an industry cannot mean anything. The same is true of conversion rate, feature adoption, and every engagement ratio in circulation.

This is why almost every benchmark in this field is unusable, and it is not a sourcing complaint. Even a perfectly sampled, transparently-published survey would produce numbers you cannot act on, because the metrics are not measuring the same objects.

Which PLG metrics are actually comparable?

The only PLG metrics comparable across companies are the ones denominated in money: CAC payback, net revenue retention and gross margin. Everything else is a ratio whose denominator each company defines for itself. It is a short list, and that is the whole reason it is short.

Even these need their definitions written down, because the variants disagree. But they are denominated in currency, not in events you define, which makes them the only figures worth quoting against anyone else.

Filings are the other place comparable numbers live, with a caveat that proves this page’s point: companies do not all use the same name for it. Snowflake reported a net revenue retention rate of 158% in its 2020 S-1. Twilio reports a dollar-based net expansion rate; Datadog reports a dollar-based net retention rate. Both have run well above 130% in recent years.¹ Three labels, three definitions, one broad idea, and if you are comparing yourself to any of them you need to read their definition first.

One that genuinely lands in the other direction: median B2B software companies were reported in 2025 to be spending around two dollars to acquire a dollar of new annual recurring revenue.² That is roughly comparable across companies, and it is the clearest single statement available of why product-led motions are being combined with sales, not replacing them, and why growing existing accounts has become the priority over new logos.

The metrics worth keeping, and what each is for

Not a stack of ten to report on a dashboard. Five questions, each with the number that answers it.

Are people reaching value? Activation rate, against a milestone you have validated by intervention rather than correlation. Watch it over time against a fixed definition; do not compare it to anybody.³

How fast? Time to value, at the median and the ninetieth percentile, with the share who never arrive reported alongside. The average alone hides every user you could have saved.

Are they coming back? Retention by cohort, and the shape of the curve, not a single number. A curve that flattens means you have a product; one that keeps declining means you have a trial nobody canceled.

Is the account growing? Net revenue retention, which is worth writing out because half the disagreements about it are definitional:

Then decomposed into seats, usage, tier and price. A single percentage conceals whether growth came from customers succeeding or from a price increase, and only one of those compounds.

Is any of it paying? CAC payback, with the free tier’s cost included. It is easy to exclude, which makes acquisition look cheaper than it is.

This is the metric where an external comparison comes closest to meaning something, and we are still not supplying a number. Published medians move year to year and the ones in circulation come from the same unciteable vendor surveys as everything else on this page. What the direction of travel supports points one way, and the acquisition-cost figure above is consistent with it, but that is a ratio of spend to new revenue rather than a payback period. One vendor-adjacent point estimate cannot establish a trend, and this page is the last place that should pretend otherwise. Get a dated median from a source you can name, or compare against your own trend and your own segments.

Those five have definitions worth writing down once, because most disagreements about them are arithmetic, not judgment:

LTV:CAC deserves a warning that applies to nothing else on the list. LTV is a forecast dressed up as a measurement: it is built from a churn rate you chose and a margin you estimated, so a flattering LTV:CAC ratio is often just an optimistic churn assumption. Read it alongside payback, which uses only money that has actually moved. A ratio that looks unusually good is worth interrogating, not celebrating, and one plausible explanation is that you are spending too little to grow.

DAU over MAU is a consumer metric

This one gets imported into B2B unexamined.

Daily over monthly active users measures what fraction of your monthly users show up on a given day. High ratios describe products people open reflexively, which is a consumer-social property. Most business software is used weekly by design: a scheduling tool, an expense system, a reporting product. Judging those on a daily ratio produces the conclusion that your engagement is terrible when the truth is your cadence is different.

It is a ratio of distinct-user counts, not a per-person visit frequency, which is a distinction worth keeping: a 20% ratio does not mean the average user shows up six days a month, only that a fifth of your monthly users appear on a given day. The B2B version is weekly over monthly, and the useful question underneath it is not the ratio at all but whether the rhythm you are seeing matches the job’s rhythm. A product used exactly as often as the work requires is fully engaged at whatever ratio that produces.

Retention comes first, and the arithmetic says so

The conventional funnel runs acquisition, activation, retention, revenue, referral, usually called AARRR or pirate metrics, after Dave McClure’s 2007 “Startup Metrics for Pirates”. Reversing it into retention-first order (sometimes written RARRA) is the more useful arrangement for anything past the earliest stage, and it is an arithmetic point, not a philosophical one.

Improving acquisition by a quarter gets you a quarter more users, once. Improving retention changes the base that every subsequent period compounds on.

The arithmetic is worth doing, not asserting. Take a thousand users:

• At 90% monthly retention, 282 of them remain after twelve months.
• At 95%, 540 remain.

Five percentage points of monthly retention produces 91% more surviving users after a year, from the same acquisition. Nothing you can do to the top of the funnel compounds like that, and retention also improves every stage downstream, because the users who stay are the ones who form habits, expand and tell other people.

The same compounding runs in the other direction, which is why preventing churn outranks nearly everything upstream of it: that chapter does the arithmetic, and the result is worse than intuition suggests.

How to know whether you are winning

Without cross-company benchmarks, three comparisons do the work:

Against your own past. Same definition, dated, quarter over quarter. This is the primary tool and it is available to everyone immediately.

Between your own segments. Activation by acquisition channel, retention by plan, expansion by company size. Internal differences are genuinely comparable because the definitions are shared, and they are usually more actionable than any external figure.

Against a competitor’s observable behavior. Not their metrics, which you cannot see, but the things you can time and read: how long their signup takes, what their free tier includes, what their pricing page gates. That comparison is real, and this handbook’s chapters are largely about how to run it.

What to do this quarter

1. Write down the definition of each metric you report, with today’s date. Expect to find that two people mean different things by activation.
2. Delete the benchmark column from your internal dashboards. It invites comparisons the numbers cannot support and it is where false comfort lives.
3. Add the ninetieth percentile next to every median you report, and the never-arrived count next to every rate.
4. Write out the five formulas above and circulate them. Most metric arguments in a company are two people using one word for different sums.
5. Decompose net revenue retention into seats, usage, tier and price. This usually changes the roadmap.
6. Put the free tier’s cost into CAC. It is the omission that flatters a dashboard most.
7. Pick the one number that reflects your constraint and put it somewhere everyone sees. The flywheel chapter is about finding which one that is.

Metrics tell you where the problem is. They do not tell you what to do about it, and the gap between diagnosis and action is where most of this handbook lives. If you would rather work from a list of concrete moves than read thirty chapters in order, the checklist sequences them by what to do first.

Updated August 2026