Leading vs. Lagging Indicators: Metrics You Can Move
The short version: a lagging indicator tells you what already happened and can't be changed anymore (last quarter's churn, this cohort's 90-day retention). A leading indicator is an earlier signal that reliably moves before the lagging one does, so you can still act on it. Churn is lagging. "Did this user hit the core action twice in week one" is leading. The whole game is finding the earliest leading indicator that's both predictive of your lagging metric and something your team can actually influence.
Most product teams already know the definitions. Where they get stuck is the derivation: given a lagging metric you care about, how do you find its leading indicator instead of guessing? That's the part I want to walk through, with actual numbers.
Why lagging metrics feel useless (and aren't)
Lagging indicators get a bad reputation because you can't do anything with them in the moment. By the time 90-day retention prints, the cohort is 90 days old and gone. Pendo's 2025 benchmark data puts average software retention around 39% after one month and roughly 30% after three. If yours is 22%, great, now you know. The quarter is already over.
But you still need lagging metrics, because they're the thing you're actually trying to improve. A leading indicator is only useful if it points at a lagging one that matters to the business. Nobody's board asks about "week-1 feature depth." They ask about revenue retention. So the lagging metric defines the target, and the leading indicator is how you steer toward it before the result is locked in.
Think of it as the difference between the speedometer and the fuel gauge. Retention is your arrival time, known only when you arrive. Leading indicators are the gauges that tell you, mid-trip, whether you're on pace.
A four-step method for deriving a leading indicator
Here's the procedure I use. It's deliberately mechanical, because the failure mode is people picking a leading indicator that feels right instead of one that's demonstrated.
- Name the lagging metric precisely, with a window. Not "retention" but "percent of new signups still active on day 30."
- List candidate early behaviors that happen inside the first few days: features touched, actions completed, invites sent, integrations connected.
- For each candidate, split users into those who did it early and those who didn't, then compare their day-30 retention. The gap is your signal.
- Among the candidates with a big gap, pick the one that's earliest and most movable. Earliest so you have time to react; movable so onboarding or product changes can actually push it.
That's the correlation step, and it's exactly the analysis any product-analytics tool is built to run, whether you're in Mixpanel, Amplitude, or a chat-first platform like Kixo where you ask the retention-split question in plain language. The tool matters less than doing the split honestly.
Facebook's famous version of this was "7 friends in 10 days." As Mode's write-up of the story recounts, the team found that users who added seven friends within their first ten days were far more likely to stick, and they made that single number the company's sole focus. It wasn't handed down from theory. It fell out of splitting retained users from churned ones and noticing where the curves diverged.
A worked example with 12 users
Say you run a small analytics-tooling product and you want to move day-21 retention. You take one week's cohort. It's tiny on purpose, 12 signups, because the logic is easier to see than in a chart of 40,000.
You track two early behaviors in each user's first three days: whether they connected a data source, and whether they created a dashboard. Then you check who's still active on day 21.
| User | Connected source (day 1-3) | Created dashboard (day 1-3) | Active day 21 |
|---|---|---|---|
| 1 | yes | yes | yes |
| 2 | yes | no | no |
| 3 | yes | yes | yes |
| 4 | no | no | no |
| 5 | yes | yes | yes |
| 6 | yes | no | no |
| 7 | no | no | no |
| 8 | yes | yes | yes |
| 9 | yes | no | yes |
| 10 | no | yes | no |
| 11 | yes | yes | yes |
| 12 | yes | no | no |
Count it up. Of the 5 users who created a dashboard in their first three days, 4 retained (80%). Of the 7 who didn't, 1 retained (14%). Connecting a source barely separates anyone: plenty of connectors churned, because connecting is passive setup, not value.
So "created a dashboard in first 3 days" is your leading indicator. It's early (day 3, not day 21), it's movable (onboarding can nudge it), and the retention gap is large and clean. You'd now go rewrite onboarding to get more people to that first dashboard, and you'd watch that day-3 rate weekly instead of waiting three weeks for retention to confirm what you already suspect.
Twelve users won't survive statistical scrutiny, obviously. In production you want the split across a few hundred at minimum, and you sanity-check it holds across a couple of cohorts. But the shape of the reasoning is identical.
Where this goes wrong
The trap is treating a predictive leading indicator as a causal lever. They're not the same thing, and conflating them burns real roadmap time.
Back to Facebook's seven friends. The seductive misreading is: "friends cause retention, so we'll auto-add friends and retention will climb." Maybe. Or maybe adding seven friends is just what engaged people naturally do, and engagement causes both. If it's the second story, force-feeding friend suggestions to disengaged users won't move retention at all. You'll have optimized the proxy while the real thing sits still.
Mixpanel made this point sharply in a piece bluntly titled "Magic numbers are an illusion." Their argument is that no matter how you slice the data, there's rarely one magical tipping point where a person becomes a lifelong user. The number correlates with users getting value; it doesn't flip a switch that manufactures value. I think that's the correct read, and it's worth saying plainly because a lot of growth writing still treats these thresholds as causal buttons.
So how do you tell correlation from a real lever? You test it. Run an experiment that moves only the leading indicator for a random subset and check whether the lagging metric follows. If you push more users to that first dashboard and their day-21 retention rises versus a holdout, you've got a lever. If the dashboard rate goes up and retention doesn't budge, you found a symptom of engagement, not a cause of it. That distinction is the entire difference between a leading indicator that steers and one that just decorates a slide.
There's a second, quieter failure: picking a leading indicator that's predictive but immovable. "Users on enterprise plans retain better" might be dead accurate and completely useless as a lever, because you can't turn a free user into an enterprise buyer by wishing it. A good leading indicator has to be something a product or lifecycle team can nudge this week.
A short reference table
Some common lagging metrics and the leading indicators teams actually derive from them. Treat the right column as candidates to validate against your own data, not gospel.
| Lagging metric (confirms) | Candidate leading indicator (predicts) |
|---|---|
| 30-day user retention | Reached the core action twice in week 1 |
| Revenue / net dollar retention | Number of active seats or integrations in month 1 |
| Annual churn | Product usage decline over the trailing 3 weeks |
| Expansion revenue | Adoption of a second high-value feature |
| Support-driven churn | First-week ticket volume or time-to-first-response |
| Day-30 ROAS | Early cohort behavior depth, not day-1 install counts |
That last row connects to a related trap in growth marketing, where early return-on-ad-spend numbers flatter you before the truth arrives. If you live on the acquisition side, the case for reading day-30 ROAS over day-1 is the same lagging-beats-early-vanity logic applied to spend.
How to actually put this to work
Pick one lagging metric your team is judged on. Just one. Run the four-step split this week and find its earliest movable predictor. Then do the thing most teams skip: put that leading indicator on the wall next to the lagging one, and check it on a cadence short enough to matter. If retention is your target, a leading indicator you review quarterly is barely leading at all.
Two guardrails before you commit the org to a number. First, don't chase a benchmark that isn't yours. Userpilot's 2024 data pegs average B2B SaaS activation around 37.5% with a healthy band of 30 to 50%, which is a fine gut-check, but your activation event and your value moment are specific to your product. Borrowed thresholds import someone else's definition of "activated." Second, keep the lagging metric in view. The leading indicator exists to serve it, and the day you start optimizing the proxy for its own sake is the day the two quietly decouple.
If you want the fuller picture of how a leading indicator fits under the metric you're ultimately accountable for, the north-star metric tree is the structure that hangs these together: one output metric on top, the movable inputs beneath it. Leading indicators are those inputs. Lagging ones are the output. Everything else is just knowing which is which, and refusing to wait three weeks to learn something you could have seen on day three.