In a signup funnel where email verification has the worst step conversion, where do you invest?

## The funnel and its two arithmetics A signup funnel with four ordered steps, over one week: ``` landing page 100,000 signup 30,000 step rate 30% email verified 12,000 step rate 40% first purchase 7,200 step rate 60% overall 7.2% ``` Two numbers describe every step and they answer different questions: - **Step conversion** — entrants to that step who advance. It measures how well that one step works, independent of everything upstream. - **Cumulative conversion** — share of the original 100,000 who reached that step: 30%, 12%, 7.2%. It measures the funnel's compounded loss and moves when any earlier step moves. The overall rate is the product of the step rates: `0.30 * 0.40 * 0.60 = 0.072`. A team that reports only cumulative numbers cannot tell whether a decline came from that step or from a change further up. ## Worst rate is not the same as biggest loss Verification has the lowest step rate at 40%. But the absolute headcount lost is 70,000 at the landing step versus 18,000 at verification. Optimising by lowest rate and optimising by largest headcount give different answers, and **both are wrong as a decision rule** because neither accounts for how tractable the step is or how much a rescued user is worth. ## The quantity that actually ranks steps For each step, estimate: ``` expected added conversions = users lost at the step * plausible recoverable share of them * conversion rate through the remaining steps ``` Apply it: - **Verification, 40% to 55%.** 30,000 signups times 0.55 is 16,500 verified, up 4,500. At the 60% downstream purchase rate that is roughly **+2,700 purchases**, taking 7,200 to about 9,900 — a 37% improvement in the end-to-end number. - **Landing, 30% to 33%.** 3,000 extra signups, times 0.40 verification, times 0.60 purchase, is about **+720 purchases**, roughly 10%. Verification wins here, but notice that it won on the arithmetic, not on having the worst rate. Change the downstream rates and the answer flips: a bad step immediately before a near-zero final step is worth almost nothing to fix. ## Sanity checks before you spend anything **Is the step real or is it measurement?** Email verification is a notorious tracking break: the click happens in a mail client, often on a different device or browser, so the event may be attributed to a different identity or lost entirely. A step that looks catastrophic may be partly an accounting artefact. Compare the count of verification emails delivered, links clicked, and accounts marked verified in the backing system before believing the funnel view. **Is the step even required?** Steps that exist for compliance or deliverability reasons may have a hard ceiling. Removing verification entirely would move all 18,000 forward on paper, but if it exists to suppress fraudulent or mistyped addresses, the removed friction reappears as bad addresses and support cost. **Are the rescued users like the ones already through?** No. People who did not verify are, by construction, less motivated than those who did. The 60% downstream rate is set by the compliant group, so applying it to marginal users overstates the gain. Treat the estimate as an optimistic ceiling and validate it rather than banking it. **Is the funnel really linear?** Real users re-enter, skip steps, arrive already signed up, or complete a step days later. A strict ordered funnel counts them as drop-off. Check how many users appear at step 3 without a recorded step 2 — a large number means the funnel definition, not the product, is producing the drop. ## Prioritising in practice Rank steps by expected added conversions, then divide by estimated effort and adjust for confidence in the estimate. Prefer the step where you have a specific, testable hypothesis about **why** people leave — an unclear instruction, a slow email, a required field — over the step with the ugliest number and no diagnosis. And once you pick a step, keep watching the steps after it: a change that pushes more people past verification but attracts weaker intent can raise one step and lower the end-to-end rate. ## The summary The worst step rate tells you where the funnel is leakiest, the absolute loss tells you where the volume is, and neither is the decision. The decision is expected added conversions after checking that the drop is real, that the step is removable, and that the rescued users are worth what the surviving ones are worth.