Conversion and the funnel: where the cats dropped off

A funnel is not four counters

The buyer's path through Kotomarket:

opened app → viewed product → added to cart → paid
  12,000          6,000            1,800        720

The most common beginner mistake is to read those four numbers as four independent counters: "this week there were so many opens, so many views, so many carts, so many payments". You can count that way, but it is not a funnel.

A funnel is ONE group of users passing through an ordered sequence of steps inside a time window.

Let us take that definition apart into three mandatory pieces.

1. One group. We fix a specific set of people — those who took the first step during the chosen week — and follow exactly them. The 6,000 at step two are not "everyone who viewed products", they are "those of our 12,000 who reached a view". A person who opened their cart that week without taking the first step is not in the funnel at all.

2. A sequence. The steps are ordered, and a person reaches step N only by passing steps 1…N−1. Hence the numbers must decrease: 12,000 ≥ 6,000 ≥ 1,800 ≥ 720. If a reported funnel grows somewhere, that is a signal it was built from independent counters rather than from passage.

3. A time window. A funnel always has a deadline: "within the same session", "within 24 hours of the first step", "within a week". Without a window the question "did they make it" has no answer: a person could open the app in March and pay in June. Calling that conversion is self-deception. At Kotomarket the window was seven days from the first step.

Entry #44 in L.'s journal: "A funnel is not columns in a table. It is a claim about people: these particular cats walked the path and stopped right here."

Why counting events gives 400%

Back to the first lesson: the counting unit decides everything. Take the same Kotomarket week and count the steps in events rather than people:

Divide events by events and you get 46,000 / 19,800 = 232%. Divide step-two events by step-one users and you get 46,000 / 12,000 = 383%.

A conversion above 100% means exactly one thing: the numerator and the denominator count different things. People view many products per visit; there are physically more view events than people. There is no "super-conversion" here, only a broken fraction.

The funnel rule:

Every step of a funnel is counted in one unit — as a rule, unique users. Numerator and denominator must be of the same breed.

This does not make events useless. "Views per user" is an excellent engagement metric, and the table above shows it: 46,000 / 6,000 ≈ 7.7 products per person. It is simply a different metric, and it has no place inside a funnel.

A quick report self-check: if any funnel step exceeds the previous one, or a conversion comes out above 100%, then almost certainly events were counted in one place and people in another.

The denominator of every step

Now let us count honestly — in unique users, seven-day window.

Step conversion is the share of people who moved from the previous step to the next. Each step has its own denominator: exactly whoever was standing on the previous rung.

The question from the previous lesson, which should now be automatic: "30% is 1,800 out of which 6,000?" The answer: out of the 6,000 who viewed a product — not out of the 12,000 who entered, and not out of all registered cats.

End-to-end conversion, and why it equals the product

End-to-end conversion goes from the first step to the last: 720 / 12,000 = 6%.

It is often given as "the product of step conversions", which looks like magic. There is no magic — it is fractions cancelling. Written out:

 6000     1800     720      6000 × 1800 × 720       720
────── × ────── × ────── = ─────────────────── = ───────  = 6%
12000     6000     1800    12000 × 6000 × 1800    12000

The numerator holds 6,000 and 1,800, and the very same 6,000 and 1,800 sit in the denominator — they cancel. What remains is 720 / 12,000. The product adds no new knowledge: it is the same path written step by step.

A useful consequence: step conversions multiply. Improve each of the three steps by a tenth and end-to-end conversion grows not by 10% but by roughly 33%. And conversely: one sagging step drags the whole product down even if the others do not move — which is exactly what happens in Case #0.

Three different "leakiest steps"

The question "where is the biggest hole?" sounds unambiguous, but it has three answers — and they disagree.

Drop-off rate is the share who did NOT pass the step: 100% minus the step conversion. The highest drop-off is on "viewed → cart": 70% of arrivals are lost there.

People lost is an absolute: how many humans vanished at the step. The largest absolute outflow is at the first transition: 6,000 people, even though its rate is not the worst.

There is the first discrepancy: the lowest conversion (30%) and the largest loss (6,000 people) are different steps. Reporting "the second one is the leakiest" can be an honest way of getting priorities wrong.

Business significance is the third and most important criterion. It asks not "where do we lose more" but "where is the payoff from fixing bigger". Let us sketch it in money, assuming an average check of 730 ₽ and that an improved step converts downstream as usual:

The same improvement in points yields different money — because the steps sit at different depths and different numbers of people flow through them. Add common sense: five points on the first step and five on the third cost different amounts of work, and that is part of the decision too.

Report not "conversion dropped" but "this transition sagged: the rate went from X% to Y%, that is this many people and roughly this much money per week". Three numbers side by side turn panic into a plan.