Ads

An ad is the last link in the chain and the only one that costs money. Everything before it exists so that this step is a test with a question, not a spend with a hope.

The shape of a test

Before market fit, Autopilot does not run “a campaign”. It runs a matrix: one campaign per revenue stream, and inside it, cells that pair an audience with a message.

Revenue stream: bookkeeping for trades
  ┌──────────────────────┬───────────────────────┐
  │ Audience A           │ Audience B            │
  ├──────────────────────┼───────────────────────┤
  │ Message 1  ← cell    │ Message 1  ← cell     │
  │ Message 2  ← cell    │ Message 2  ← cell     │
  └──────────────────────┴───────────────────────┘

Each cell is a question: does this argument work on these people? That only stays answerable if a cell gets enough budget to produce a readable result — so cells have a floor, and a test with too many cells for its budget is cut down rather than run thin.

The rules are written before the money moves

Three things are fixed at launch, not decided afterwards while looking at the numbers:

• The decision window. How long the test runs before it is judged.
• The kill rule. What result means “stop this cell”.
• The scale rule. What result means “give this cell more”.

This is the part people skip, and it is the reason most ad testing produces stories instead of decisions. A rule invented after seeing the data will always confirm what you already hoped.

A phase closes when its window runs out, or when every cell has answered — and a cell that never delivered at all (never built, or built and left paused, zero spend) is recognised as such instead of being waited on forever.

What the result is worth

A cell’s verdict travels back up the chain to the claim it tested. That is the point of building the chain in order: the market’s answer lands on the argument, not just on a picture.

Two honest limits:

• A click is not a customer. Where the funnel is instrumented, judgement uses what happened after the click; where it is not, a cost-per-click verdict is nearly worthless and is treated as such.
• One test answers one question. A cell that loses tells you that this message did not work on these people — not that the claim is dead.

What you decide

Autopilot does	You do
Proposes the test: cells, budget, window, rules	Approve the plan — and the budget
Produces the creative for each cell	Approve each creative before it renders publicly
Runs, measures and closes the phase	Read the verdict and decide the next round

Nothing spends without your release, and a budget change is its own approval — not a footnote inside another one.

Where to read next

• Messages — where the words in an ad come from
• Claims — what a test is actually testing
• What Autopilot costs — the platform’s own cost, separate from your ad spend