Product Metrics

"How's the product doing?" never had a single correct answer. Sales looks at this month's closed deals and says it's going well. Engineering looks at the incident count and says it's about average. The CEO looks at the bank balance and says she's worried. Nobody is lying. Each of them is reading the number in front of them, and no one has written down how the three connect.

Without that connection, meetings turn into a contest of who sounds most certain. Early Pagedeck ran that way. A slide went up reading 120,000 total signups and the room applauded, and nobody asked the one question that mattered: how many of those arrived this month. Counted afterward, it was 1,900, and 400 of them came back in their second week. A cumulative signup count can only go up, so it keeps rising while a product falls apart. That was the week the team learned that a number which looks good and a number you can act on are different objects. Since then, choosing what goes on the slide starts with one test: if this gets worse, does the slide show it getting worse.

Product metrics are the set of numbers a team fixes in advance to cut down that confusion. The point isn't collecting more of them. It's pairing each number with the decision it answers before anyone needs it. Whether to rework the signup screen is answered by signup completion rate; where engineering time goes this quarter is answered by the share of new accounts still active four months later. With the pairs fixed, a meeting has one number to reach for, and when that number refuses to move, the list of possible causes is already narrow.

Inside the product field, this term names the group that measures whether what shipped is actually working. That group also holds usage metrics for how much the product gets used, activation for the moment a new user first feels the product pay off, retention for how reliably people come back after that, and satisfaction metrics for the parts that only a survey can reach. Those four divide by what gets measured. The five terms under product metrics divide by something else — the job a number is hired to do.

Five terms sit under this one, and all five describe a metric's character rather than its subject. They split by where a number is used, not by what it counts, which means the same figure can be a headline metric on one team and a vanity metric on another.

1. North star metric — the single number a whole team steers by. You pick whatever comes closest to the value the product actually delivers. Pagedeck uses weekly collaborative documents: the count of documents that two or more people edited together in a week. Signups lost out to that number because the reason anyone opens the product is to work on one document with other people.
2. Leading indicator — a number that moves before the result does and hints at what's coming. The share of new teams that create three or more documents in their first week is one. When that share drops this month, paid conversion drops three months later. Moving early means you can still act on it, and the cost is that the link to the result is sometimes a guess worth checking.
3. Lagging indicator — a number that confirms what already happened. Monthly revenue and quarterly churn belong here. They're reliable and late: by the time you read March churn in April, those customers are gone. That's why the two get paired. You steer on leading indicators and check yourself against lagging ones.
4. Vanity metric — a number that looks good and changes no decision. Cumulative signups of 120,000 is the classic. Whether it reads 130,000 or 140,000 next quarter, tomorrow's build list is identical. The test is short: ask what you'd do if this number were cut in half, and if there's no answer, it's a vanity metric.
5. Counter metric — a number that watches what might break while you push a target up. Send three onboarding emails a day and signup completion rises along with the unsubscribe rate; the unsubscribe rate is the counter metric. It gets set in advance alongside the target, and if it crosses its agreed line, the change comes out even though the target improved.

Defining a metric means writing down the counting rule as a sentence, not picking a name. For weekly active users there are four things to write: what action qualifies as active, how the window is cut, how a person who returns several times is counted, and who is excluded outright. The Pagedeck rule reads "accounts that opened at least one document in the last seven days, counted once each, excluding employee and automation accounts." Without that sentence, a number with one name produces different values depending on who pulled it. Whether people who logged in and left are included is enough on its own to move the figure 20 percent.

Ratios need both halves written down before they mean anything. Retention of 30 percent says something entirely different when the denominator is this month's signups than when it's the cohort that joined four months ago. Grouping the denominator by when people joined is called cohort analysis — put the four-month survival of the January group next to the June group and read them side by side. A single blended average can't tell you whether the product improved, because heavy new signups drag the average down on their own.

A good level doesn't get imported from outside. The most common mistake is reading that industry-average retention is 35 percent, seeing 30 on your own chart, and declaring a problem. You compared against a number without knowing who sat in its denominator or over how many months it was measured. Build the standard from three places instead: your own figure last quarter, what the same metric did the last time you moved it deliberately, and the value the business arithmetic requires. If the plan needs $30,000 a month and an account pays $300, that's 100 paying accounts; at 5 percent monthly churn, five new accounts a month only holds the line.

Put the five together and you get one instrument panel. The north star sits in the middle, two or three leading indicators to its left, two or three lagging indicators to its right, and a counter metric underneath holding the line nobody should cross. Vanity metrics are what you take off the board. The shape matters because it forces a question order. You read the north star to learn whether the product is delivering more value than it was, the leading indicators to learn what next quarter looks like, the lagging ones to learn whether last quarter's call was right, and the counter metric to learn what the last change cost. Rebuilding a metric set in practice usually ends not with finding more numbers but with removing half the ones already on screen. Four of the five also depend on each other: leading and lagging only mean something as a pair, a counter metric needs a target to guard, and vanity only becomes visible once the other four are chosen. The north star can stand alone, and left alone it hides whatever breaks while you raise it.

"That's a vanity metric. It's cumulative, so it can't go down. Let's look at this month's new accounts."

The number on screen can't reveal a product getting worse, so the basis should change.

"The target metric for this test is signup completion and the counter metric is first-week churn."

Stating in advance that more signups who don't stick around won't count as a win.

"That definition gives a different value on every team. Pull it again with employee accounts excluded."

The counting rule was never agreed in writing, so one name is producing several numbers.

Pagedeck holds a metrics meeting on the first Tuesday of each month. Seven numbers go on the screen and no more: one north star, three leading, two lagging, one counter metric. The first version had twenty-three and the meeting ran past two hours without producing a single decision, because explaining why each number moved consumed all of it. Mia asked everyone to write one line next to each number saying what they would do if it changed, and sixteen of those lines came back blank. The blank ones came off the screen. Several of them were the numbers people had been quoting most often, which is part of why the list had grown in the first place.

The meeting now finishes in thirty minutes. Grace reads the north star across four months first, then checks whether any of the three leading indicators bent. Last month the share of teams creating three documents in their first week had fallen from 41 to 33 percent while revenue held flat. Read only from lagging indicators, nothing happened that month. Ravi traced back what had changed that week and found that one ad channel was delivering teams of a different shape, and the decision to cut that channel's budget was made in the room.

The most common mistake is believing that watching more metrics produces a more accurate picture. Analytics tools display dozens at once, which makes it look that way, and proposing to remove some can sound like proposing to hide something. The effect runs the other way. As the count grows, every month guarantees that a few numbers rose and a few fell, which makes it possible to explain using only the ones that rose. With twenty-three on screen, any decision can find support somewhere. Left alone, the metrics meeting stops being where decisions get made and becomes where decisions already made get justified. So the rule is to write the sentence first — if this moves one way, we will do that — and to leave off any number that can't carry such a sentence. Taking a number off the screen doesn't mean you stop recording it. It keeps accumulating, and the split is between numbers you review together every month and numbers you pull when a specific question comes up.

The second mistake is expecting the numbers to make the decision. They were built as evidence, so evidence reads like an answer. A metric reports what happened and stays silent about why. That retention fell from 30 to 24 percent is one question; why it fell is another, and the second one gets answered by talking to users or following the records, not by looking harder at the chart. Miss that split and you start treating the number directly: more onboarding emails, more banners in the interface, all aimed at a figure whose actual cause was last month's pricing change. The figure stays where it was and the users get pestered. A metric is a map telling you where to dig, not the hole. Teams that keep this straight tend to pair every metric review with one qualitative source — a handful of exit surveys, a session recording, a support thread — so the question of why has somewhere to start.