The Real Driver Behind the Metric

Revenue is a confession, not a plan. By the time MRR moves, the decision that moved it happened weeks ago — somewhere upstream, in a step you probably are not measuring. The dashboard shows you the last domino to fall, then asks you to push it directly. You cannot. Nobody can push revenue. You can only push the thing that makes the thing that makes revenue.

So I treat every top-line metric as a lagging output of a causal chain. For most products the chain runs something like this: visits become activated users, activated users become power users, power users become payers, payers become revenue. Five links. Teams stare at the fifth and wonder why it will not budge. The honest answer is usually that the bottleneck is at link two, and they have spent the quarter polishing link five.

The work is unglamorous: reverse-engineer the chain to the earliest link you can actually control, then move that. Not the number your CEO quotes in the board deck — the number three steps before it that quietly determines it. If activation is where people fall off, no amount of pricing cleverness at the paying step will save you. You are tuning a faucet while the pipe upstream is cracked.

A useful test: for each link, ask what single thing, if it doubled, would do the most for the link after it. Then ask whether you can even influence that thing this quarter. The leading driver is the earliest link that is both high-leverage and controllable. It is almost never the one on the slide. It is usually boring — a step in onboarding, a habit-forming action, a moment of first value — and it is where the effort should go.

Here is the part people skip. Before you trust a relationship between two numbers, prove the data underneath it is even real. A gorgeous correlation is often an artifact of how the data was assembled — a join that double-counts, a cohort that silently excludes the people who left, or a definition that quietly contains its own answer. If your active-user metric is defined as someone who did the paid action, do not be surprised when active users predict revenue. You built that loop with your own hands.

So I try to break my own metric before I believe it. I pull the raw rows and count them by hand for a single day. I check whether the same person shows up twice under two IDs. I ask what would have to be true for the correlation to be a coincidence, then go looking for exactly that. More striking patterns die this way than survive — and the habit that kills them is not optional.

Pick your loudest top-line metric — the one in every standup. Draw its causal chain on one line, left to right, four or five links. Mark the link you currently spend the most effort on. Then mark the earliest link you could genuinely move this quarter. If those two marks are not the same link, you have found your problem, and probably your fastest win.

Then, before you build anything on top of that chain, spend an afternoon trying to falsify its strongest correlation. Recount one day by hand. Hunt for the double-count. Read the metric’s definition out loud and check that it does not already contain the outcome. Move the real driver, distrust the pretty number, and you will stop optimising the wrong thing with great precision.