Move four

A rule that refuses is a rule you can debug.

Funnel One is a GTM platform for running a B2B funnel under one set of rules, from first touch to closed-won. Move four is how the rules change. They are configuration, not code, so you edit them without a deploy. You can test a change against your own accounts before you commit it, and an experiment will not name a winner until the maths earns it. The funnel gets better because your team improves it, not because the software improves itself.

The test

You don’t create an experiment. You draw a split, and one appears.

There is no manual create path, which is deliberate: an experiment you had to remember to set up is an experiment that mostly does not exist. Draw an A/B split on the canvas and publish, and the experiment creates itself, with sticky bucketing so a contact stays in the arm it was assigned and the variant stamped on the enrolment, the send and the call. Then it waits. Below the sample floor it will not call anything, and the promote button is not there to be clicked. Above it, one click promotes the winner.

When you do promote, the confirmation says what it will do in plain language: the winning branch becomes the only path, the losing branches are removed, and contacts already mid-journey on a losing branch finish the path they are on while everyone new gets the winner. No surprises after the fact, because the blast radius is stated before it happens.

Experiments · Opening line

Opening line — direct vs. context

Running

• A Direct ask Control 31 replied / 384 progressed 8.1%
• B Context first 43 replied / 384 progressed 11.2%

Still collecting — not enough data to call a winner yet. B is ahead. Below 95% two-sided confidence there is no winner to name and nothing here to promote.

Simulate, then promote

See which accounts change tier before anything moves.

A scoring change is the sort of thing that looks fine in a spreadsheet and reshuffles half the pipeline on Monday. So a draft model runs against a real sample of your own companies first, and returns the tier shift: how many move up, how many move down, and which ones. Nothing is written to a company’s score during a simulation — the only thing saved is the result. You look at the list, and then you decide whether to publish it.

The honest shape

You can change the rules and test the work. You cannot yet see what a rule change did to the funnel.

Sequences report. Signals report. Tasks and experiments report. The rulebook itself does not: there is no loss-reason breakdown, no stage-to-stage conversion view and no transition timeline on a record. The data is written — every transition, permanently — and it is reachable through the API. The screen that reads it is not built.

We could leave this out and most readers would not notice. It seemed better to tell you where the edge is, because you would find it in week two and then wonder what else we had not mentioned.

Questions