Funnel metrics and conversion decomposition

Your overall conversion rate is a product of the steps — and the product hides where the money actually leaks.

The idea

People don’t go from stranger to paying customer in one leap. They pass through stages: visit → sign-up → activate → subscribe → paid. Your overall conversion is every stage rate multiplied together.

Because it’s a product, the single overall number can’t tell you where you’re losing people — and two very different funnels can share the exact same overall rate. To find a leak you compare stage rates, not the total. And often an early stage predicts trouble that only shows up in revenue much later.

How it works

Overall conversion is not a stage — it’s the chain of stages multiplied. That has three consequences worth internalising:

1 · it is a product, not a sum
  overall = r1 × r2 × r3 × r4 × r5
          = 25% × 60% × 40% × 70% × 80% = 3.36%
  a 40% relative drop in ONE stage (activate→subscribe 40%→24%)
          = 25% × 60% × 24% × 70% × 80% = 2.02%

2 · the same total can hide opposite stories
  funnel A: 30% × 50% = 15%        funnel B: 50% × 30% = 15%
  identical overall — but A leaks at step 2, B leaks at step 1

3 · early stages predict late trouble
  bad top-of-funnel traffic converts to sign-ups fine, then
  activates poorly — the damage shows up in PAID weeks later

So you never diagnose from the overall number. You lay this month’s stage rates next to last month’s (or one segment next to another) and look for the rate that moved.

When to use it

Decompose the funnel when…	The trade-off / limit
Overall conversion moved and you need to know which stage caused it.	You need clean stage definitions and enough volume per stage to trust the rate.
You’re comparing segments (paid vs organic, mobile vs desktop).	Thin segments give noisy rates — a “collapse” can be small-sample wobble.
You want an early-warning metric before revenue is affected.	Early proxies can mislead if the downstream link isn’t actually causal.

Watch out for

• Reading the overall rate as a diagnosis. “Conversion is down 1 point” tells you nothing actionable. Only the stage comparison does.
• Mix shift masquerading as a stage change. If cheap, low-converting traffic grows, the blended rate falls even though every segment held steady. Always ask whether the mix moved before blaming a stage.
• Chasing a tiny-volume stage. A rate computed on 30 users swings wildly. Confirm the stage has the sample to be believed before you act on it.
• Optimising an early stage that hurts a later one. Loosening sign-up (say, dropping email verification) can lift sign-ups while tanking activation and paid. Watch the whole chain, not one bar.
• Redefining a stage mid-analysis. If “activated” changes meaning between periods, the rate change is an artifact, not a leak.

Worked example

An interviewer says: “Paid conversion dropped 15% this quarter. Where do you look?” A weak answer guesses a cause. A strong one decomposes: “Overall paid conversion is visit×sign-up×activate×subscribe×paid. I’d pull each stage rate this quarter against last and find the one that moved — that localises it. If sign-up held but activation fell, the problem is onboarding, not acquisition. Then I’d split by source, because a blended drop is often a mix shift: if we scaled a cheap paid channel that converts poorly, the blended number falls while organic is untouched — the fix is the channel, not the product.” Naming the decomposition, then the segment check, is the whole move.