Flow Test
by
Lifecycle flow testing for ecommerce teams: pick the variable with room to move, size the test against the volume the flow can actually deliver, and get a plan with subject, offer, timing, content and delay written out.
Most lifecycle teams test subject lines, because subject lines are easy to change, on flows too small to ever resolve the test. A quarter later the variant is still running and nobody can say whether it won. The measurement is usually borrowed from the ESP, which credits the flow with any order that follows an open. This skill reads volume, conversion and revenue for each step of a flow, ranks which variable has genuine room, sizes the sample the flow can deliver and the weeks it will take, then writes a plan with one variable, a hypothesis and a stopping rule.
What It Does
- Reads volume, conversion and revenue for every step of the flow.
- Ranks the variables with room to move: subject, offer, timing, content, delay.
- Sizes the sample each flow can deliver and the weeks needed to call a winner.
- Separates flow revenue from campaign and blended revenue before measuring lift.
- Writes a test plan with one variable, a hypothesis and a stopping rule.
- Flags flows too small to test so effort goes where it can actually resolve.
Use Cases
- Stop running subject line tests on a flow that sends 400 emails a month.
- Decide whether the cart delay or the offer is the variable worth testing first.
- Give a lifecycle manager a queue of tests rather than one idea at a time.
- Keep a record of what was tested so the same variable is not re-run next quarter.
Example Session
What should I test next in the abandoned checkout flow, and how long until I can call it?
The flow sends about 3,100 emails a month across three steps. Two variables have room, one does not.
- Variable | Current | Room | Sample needed | Weeks to call
- Delay on step 1 | 4h | High | 1,800 | 3
- Offer on step 3 | None | High | 2,400 | 4
- Subject on step 1 | 38% open | Low | 9,000 | 14
Test the step 1 delay first: it needs the smallest sample and the flow's drop-off is concentrated in the first four hours. Hold the subject line test, because at 3,100 sends a month it takes a full quarter to resolve and the open rate already sits above the flow's own benchmark. Run the offer test after this one rather than alongside it, so the two do not contaminate each other.
Why This Beats Prompting It Yourself
An Idea vs A Test That Can Resolve: The Hidden Risk
A generic prompt will produce test ideas all day. Ask what to test in a flow and you get five plausible variants, none of them sized against the volume that flow actually sends, and no warning that the first one needs a quarter of traffic to reach a conclusion. The ideas are fine. The problem is that three of them cannot be answered by this flow, and nothing in the prompt is going to say so.
When You Still Need The Sample Math
A test you cannot call is worse than no test, because it occupies the flow and produces a number someone will quote anyway. This skill states 3,100 sends a month, 1,800 needed, three weeks to a call, and refuses the subject line test at fourteen. It also measures on flow revenue rather than the ESP's attributed figure, so the winner is one your finance team would also recognize as a win.
Known Limitations
It needs send and conversion data per step, which means a flow that was rebuilt recently has too little history to size anything and is reported as such. Sample sizing assumes the flow's traffic stays roughly stable, so a test running across a peak trading period will resolve faster or slower than the estimate. It plans and measures tests but does not build them in your ESP.
Install It
- Download the ZIP.
It is free and there is no account to create. - Unzip it into your agent's skills folder.
Claude Code reads ~/.claude/skills/, which is hidden by default: the command in the folder block opens it. Other agents scan their own directory, so drop the same folder there instead. - Ask your agent to use it.
Restart the agent if it was already running, then it picks the skill up with no config.
skills folder
Copy
~/.claude/skills/flow-test/
SKILL.md
references/sample-sizing.md
tests/
# one plan per test, with its stopping rule
# macOS: create the folder and open it in Finder
mkdir -p ~/.claude/skills && open ~/.claude/skills
# Windows: paste in the Explorer address bar
%USERPROFILE%\.claude\skills\Frequently asked questions
How is this different from asking an agent for test ideas?
Ideas are the easy part. This ranks the variables by how much room each one has, sizes the sample your flow can actually deliver, and tells you which tests this flow will never resolve. What comes back is a queue with weeks attached, not a list of suggestions.
Which agents does the skill run in?
Any agent that supports the open SKILL.md format: Claude Code, Cursor, Codex CLI, GitHub Copilot, Gemini CLI, Manus, Grok and others load it unmodified. The format is portable, the location is not. Each agent scans its own skills directory, so you drop the same folder into whichever one yours uses. There is no config file to edit and no API key to provision.
What does it need connected to work?
Send, open, click and conversion data per flow step, plus an order source so lift can be measured on revenue rather than opens. With only ESP data it still plans the test and says clearly that the measurement will be the ESP's attributed revenue.
How does it decide a flow is too small to test?
By comparing the sample a variable needs against what the flow sends in a reasonable window. If the answer runs past a quarter it says so and moves the variable down the queue. A test nobody will wait for is not a test, it is a variant left running.
Can it run more than one test at a time?
It will sequence them rather than stack them. Two variables changed at once in the same flow produce a result you cannot attribute, so the plan orders them and says why. If two flows are independent it will happily run a test in each.
What is included with the skill?
The skill itself, the sample sizing reference, the test plan template with its stopping rule, and the revenue definitions used to measure lift. It is free and security scanned. We re-publish the ZIP when the sizing model changes, so download it again if your estimates start drifting.
A flow test is only as good as the revenue attributed to it.
Polar separates flow revenue from campaign and blended revenue in one semantic layer, so the winning variant is one your finance team would also call a win.
<script type="application/ld+json">{"@context":"https://schema.org","@type":"FAQPage","mainEntity":[{"@type":"Question","name":"How is this different from asking an agent for test ideas?","acceptedAnswer":{"@type":"Answer","text":"Ideas are the easy part. This ranks the variables by how much room each one has, sizes the sample your flow can actually deliver, and tells you which tests this flow will never resolve. What comes back is a queue with weeks attached, not a list of suggestions."}},{"@type":"Question","name":"Which agents does the skill run in?","acceptedAnswer":{"@type":"Answer","text":"Any agent that supports the open SKILL.md format: Claude Code, Cursor, Codex CLI, GitHub Copilot, Gemini CLI, Manus, Grok and others load it unmodified. The format is portable, the location is not. Each agent scans its own skills directory, so you drop the same folder into whichever one yours uses. There is no config file to edit and no API key to provision."}},{"@type":"Question","name":"What does it need connected to work?","acceptedAnswer":{"@type":"Answer","text":"Send, open, click and conversion data per flow step, plus an order source so lift can be measured on revenue rather than opens. With only ESP data it still plans the test and says clearly that the measurement will be the ESP's attributed revenue."}},{"@type":"Question","name":"How does it decide a flow is too small to test?","acceptedAnswer":{"@type":"Answer","text":"By comparing the sample a variable needs against what the flow sends in a reasonable window. If the answer runs past a quarter it says so and moves the variable down the queue. A test nobody will wait for is not a test, it is a variant left running."}},{"@type":"Question","name":"Can it run more than one test at a time?","acceptedAnswer":{"@type":"Answer","text":"It will sequence them rather than stack them. Two variables changed at once in the same flow produce a result you cannot attribute, so the plan orders them and says why. If two flows are independent it will happily run a test in each."}},{"@type":"Question","name":"What is included with the skill?","acceptedAnswer":{"@type":"Answer","text":"The skill itself, the sample sizing reference, the test plan template with its stopping rule, and the revenue definitions used to measure lift. It is free and security scanned. We re-publish the ZIP when the sizing model changes, so download it again if your estimates start drifting."}}]}</script>