Your flows have a P&L. Almost nobody has ever pulled it.

What you’d be able to build

One row of one account, with the derivation.

This is the output shape, not a preview of it. A flow, the people who entered it, what it returned, and the only number that lets you compare it to any other flow. Client codes, never names — these are other people’s businesses.

1. Klaviyo flow report, a six-month window. Revenue is Klaviyo-attributed at the account’s configured window, which is a choice and not a fact — lesson 1 tells you where to find yours and what it does to the ranking.
2. Recipients are unique profiles that entered the flow inside the window, not sends and not deliveries. Dividing revenue by that denominator is the entire trick, and it is the one number the flow report will not give you.
3. This is a derived figure, not a holdout. It says what a flow returned. It does not say what would have happened without it — nothing on a flow report can, and lesson 5 is about the only thing that can.

What’s in it

Five queries, in the order I’d run them.

The rule under each lesson is the grade of number it lets you produce — the same notation this site uses on its own figures. Only the last one earns the heavy rule, and that ordering is the argument.

1. 01

   Revenue per recipient, by flow

   Pull the flow report, divide revenue by unique entrants, and rank. The ranking is almost never the one you’d guess from revenue alone, because revenue rewards whichever flow gets the most traffic and tells you nothing about which one is worth building.

   Produces: a derived figure
2. 02

   The dead-flow count

   A flow that is live, sending and returning nothing is worse than one that’s switched off, because it’s spending your sending reputation to do it. Count them. Then split them: the ones that are dead, and the ones that were never actually built.

   Produces: a derived figure
3. 03

   What a second order is actually worth

   Split your customers at one order versus two or more, and compare revenue per customer between the groups. That ratio is the business case for everything else you want funded, and most brands have never run it.

   Produces: a derived figure · cohort analysis
4. 04

   The window

   Median days between the first order and the second, and where your sends currently land against it. Most programs send on a calendar the business chose. This is the one the customers chose.

   Produces: a derived figure · cohort analysis
5. 05

   Design the holdout before you build

   How to carve a randomly assigned control group before the build starts, pre-register the metric, and set the read window — so that in ninety days you report a number nobody in the room can argue with. This is the only lesson that produces evidence rather than a reading.

   Produces: a figure measured against a holdout

Each lesson is a query and an arithmetic, not an opinion. You run all five against your own account, and at the end you own four derived numbers and a test design. I never see any of it.

What I’m not going to tell you

How many people are on this list.

A subscriber count is an aggregate with no denominator, and this site doesn’t publish those. It’s also the oldest trick in the format: the number goes on the form exactly when it flatters and comes off when it doesn’t.

This list started on 22 September 2026 with nobody on it. When there’s something worth reading about how it’s performing, it’ll be published with the same grade stamps as everything else here — including the months it didn’t work.

Who wrote this

The same arithmetic, already shipped and ungated.

Lesson 3 is already live as a calculator. It runs in your browser, there’s no email gate, and there’s no version of it where I see your numbers. Use it now — the list is optional and it always will be.