Funnel Analysis

When a product sells less than expected, the team usually holds two numbers and nothing in between. Forty thousand visits this month, 210 paid conversions. What happened between those two numbers was anybody's guess. A conversion rate of 0.5 percent reports an outcome; it says nothing about which part of the product has to change for that number to move.

So the meeting turned into a contest of hunches. The designer says the signup screen is too busy. The marketer says the traffic from ads is low quality. The engineer says pages load too slowly. All three sound reasonable and none of them has evidence behind it. The loudest voice wins a month of work, the number doesn't move, and the same conversation runs again.

Funnel analysis is the idea of cutting that stretch into pieces. Put four or five screens between the 40,000 and the 210, count the people standing at each one, and it turns out that 40,000 becoming 210 is not one catastrophe but a series of small losses spread across several steps. Once you can see where the steepest loss happens, there is much less left to guess about.

Pick one target action, list the steps that lead to it in order, count people at each step, and compare each step's pass rate against the ones around it.

1. Choose a single target action. This becomes the last box in the funnel — a paid subscription or a first document created, not both. Two targets at once means every step is ambiguous about which population it counts.
2. Write down the steps that lead there, in order. Four or five boxes is typical. For Pagedeck's free signup funnel: homepage visit, signup screen reached, signup completed, first document created. If the team already maps the screens and branches a user crosses on the way to a goal in a user flow, pull the boxes from that map. Slice too finely and every box loses a little, which hides where the real problem sits.
3. Count people at each box. The unit is people, not actions. Someone who opens the signup screen three times counts once in that box.
4. Calculate the pass rate of each box against the one before it. A chain of 40,000, 5,000, 1,200 and 700 gives rates of 12.5 percent, 24 percent and 58 percent. Record the end-to-end rate of 1.75 percent too, but the step-to-step rates are what tell you where to work.
5. Pick the box that is unusually low, not the lowest one. The 12.5 percent at the top is normal, because most visitors arrive with no intention of buying anything. The 24 percent is the one that needs explaining: of the people who already reached the signup screen with an account in mind, three out of four walked away. A box singled out this way is called a funnel leak.
6. Redraw the same funnel by group. One line for everyone shows only an average. Layer on segment analysis, which splits the population into groups and compares them, and if signup completion runs at 8 percent for ad traffic and 34 percent for search traffic, the thing to fix is the ad, not the screen.

It fits flows where the path is defined in advance and users are expected to walk it in order — signup, checkout, onboarding, anywhere the team has already drawn the boxes. It fits poorly when people start anywhere and finish anywhere, such as browsing documentation, because boxes imposed on that behavior produce numbers that don't mean anything.

There are three limits worth knowing.

1. It shows where people left, never why. The reason comes from using the screen yourself or asking the people who left. Actually repairing the spot you found belongs to conversion rate optimization (CRO), the work of getting more conversions out of the same traffic.
2. Anything outside the path you defined is invisible. Someone who bounced between the pricing page and the help center before leaving lands in none of the boxes, so the moment the decision was actually made never appears in the chart.
3. It watches one journey only. Whether that person comes back after the target action is a separate question, answered by retention analysis, which tracks return behavior over time. Raise signup rates without looking at who is still there a month later and you get a quarter where the funnel numbers improve and revenue doesn't.

"I drew the funnel and 76 percent drop off at the signup screen."

Three out of four people who reached that screen left without creating an account, and the speaker is proposing that screen as this cycle's work.

"Is this counting sessions or users?"

Asking how repeat visits by one person were counted. The two bases can move the same funnel's rates by nearly a factor of two.

"Can you break this out by acquisition channel instead of showing the total?"

A request to split the population into groups, because a single average hides the cause.

The chart shows up in Pagedeck's Monday metrics meeting. Paid conversion has been flat for two months, Ravi suggests doubling the ad budget, and Mia puts the funnel on screen. Visits went from 30,000 to 40,000 while completed signups went from 1,200 to 1,210. The extra 10,000 people reached the signup screen and left without an account, so twice the budget leaks at exactly the same spot at the same rate. The conversation moves to that screen instead. Noa counts nine input fields on it out loud, and Theo points out that company name and job title could be collected later. The funnel didn't supply an answer; it turned a meeting about ad spend into a meeting about one screen.

The most common misreading is treating the lowest box as the biggest problem. The smallest number looks like the worst number, which is a natural enough assumption. But the first box of a funnel is supposed to be low. Most visitors arrive with no intent to buy, so 12.5 percent from visit to signup is that step working normally rather than breaking. Teams caught by this spend a quarter pushing on the box that moves least while leaving the one that would move immediately untouched.

Reading the funnel chart as the actual movement of users is the same kind of mistake. Drawn as a tidy staircase, it looks like people really walked it in that order. What it actually shows is records forced into boxes someone chose in advance, so a person who arrived mid-way through a link from a colleague stands in the same line as everyone else. Take it for a real path and you can spend weeks fixing a box nobody ever stepped on.