Input Metric

When goals were set only as results, a quarter could start without anyone knowing what to do on Monday. A team told to raise retention by 5 percent has no screen to open. Retention is hundreds of actions accumulating over months, and nothing inside that number is something you can touch this morning.

Quarterly reviews settled into a shape because of it. Month one, everyone acted on guesswork. Month two, the number hadn't moved, so everyone tried something else. Month three, the result arrived with no time left to change anything. With one number being tracked and that number only moving at the end of the quarter, there was no way to tell from inside whether things were going well.

Input metrics exist to fill that gap. They count the actions believed to produce the result, so that a team can see whether the plan is running before the result exists. It isn't a substitute for the result — it's what you watch while waiting for it.

Count what the team actually did. Onboarding emails sent this week, help articles rewritten, average time from a ticket arriving to the first reply — tally the team's own actions and the tally is the number. Nothing has to be collected from outside, which makes the arithmetic the easy part.

The hard part is choosing what to count. Run each candidate through three questions.

1. Can the team move it alone? If another team's approval or a customer's decision sits in the middle, it isn't an input metric. The team decides whether the onboarding email goes out; the customer decides whether it gets opened.
2. Does it change within a week? What the team did this week has to land in this week's number, or there's no chance to adjust next week. Anything that takes two months to reflect belongs on the result side.
3. Is there reason to believe it connects to the result? The reason is usually historical data. If teams that opened help articles at least twice in their first two weeks retained better at three months than teams that didn't, rewriting help articles is plausibly connected to retention.

Keep one or two that survive all three. Post five and none of them gets reviewed weekly.

Aggregate by week. Daily figures swing with the day of the week, and monthly ones give you two chances to correct course per quarter. Count fresh each week rather than cumulatively, and put the paired output metric on the same screen so both movements are read together.

There is no absolute value worth calling good here. Whether 200 onboarding emails is a lot doesn't resolve on its own. Three things get checked instead.

First, whether the planned volume actually happened. Plan ten help articles for the week, rewrite three, and the week didn't run to plan — and that shows up on Friday rather than at the end of the quarter. This is the first thing an input metric buys you.

Second, whether the paired result follows. If onboarding emails doubled and the count of active teams held flat for two quarters, that input isn't connected to the result, so the answer is to change the metric rather than work harder at it. Reading inputs weekly and results quarterly, and checking that link, is how the metric gets used.

Third, whether the way to raise the number is good for the product. The fastest route to halving response time is sending an acknowledgment first, which lowers the value and changes nothing the customer experiences. The failure described by Goodhart's law hits input metrics especially hard, because counting actions directly makes number-filling behavior easy to find. So response time gets a companion, like repeat-contact rate, that closes the shortcut.

Candidates usually sit on the lower branches of a metric tree. Split a large result through multiplication and addition and keep going down, and at some level the values become things a team can touch.

"Keep retention as the quarterly goal and watch two input metrics weekly."

Read results by quarter, but keep separate action metrics to check every week.

"That isn't an input metric. The customer decides that value."

The team can't move it alone, so swap it for something controllable.

"Inputs went to plan and the output didn't move. I think the link is wrong."

The assumption between action and result needs revisiting.

Pagedeck once set paid-team retention as a quarterly goal. Grace asked for three-month retention to go from 78 percent to 85 percent, and for two weeks nothing happened. The target was clear, and there was no screen to open on Monday that pointed at it.

Mia dug through past data and found the fork. Teams that created three or more documents within the first two weeks of their contract retained noticeably better, and teams that didn't usually failed to reach three months. So the weekly metrics changed to two: the share of new teams that got an onboarding call inside two weeks, and the number of those calls where three documents were created together. Both were things Sofia could move that week. Six weeks in, retention hadn't budged yet, but call coverage had gone from 40 percent to 72 percent, and the plan was confirmed as running every single week.

The most common belief is that hitting the inputs brings the result along. The pair was designed as a chain, so the belief comes naturally. But that chain is an assumption the team made, and assumptions can be wrong. Left unexamined, it produces a quarter spent doubling onboarding emails with nothing improved, while the team reports having worked exactly to plan. So each quarter the inputs and the result get laid side by side and the link gets a verdict.

The second mistake is treating these as safe to set as targets because they only measure activity. They feel fairer than results, since the team controls them. That control is precisely what makes the number easy to fill. Set rewritten help articles as a target and fixing one typo files as one article; three months later the count has tripled while the articles customers search for are untouched.