Strategy and the learning loop

Strategy is B4ker's learning loop. Instead of running campaigns and reading loose numbers, every campaign the strategist proposes carries a prediction registered before it goes live: and is measured against that prediction afterwards.

The three steps

1. Hypothesis: the strategist reads your brand knowledge and Strategy Memory and proposes a campaign with an explicit prediction (e.g. "a 1% lookalike audience beats broad on CPL"). It is recorded before the campaign goes live, so it cannot be rewritten once the result is in.
2. Verdict: the analysis measures the real outcome against that prediction and closes the experiment: validated, refuted or inconclusive. Inconclusive is a legitimate result: insufficient volume does not become a learning.
3. Learning: what held up enters Strategy Memory with a confidence level, and comes back as input to the next proposal.

Why the prediction comes first

A number read after the fact almost always confirms the story you already had in mind. Registering the prediction first turns the campaign into a real test: the result is allowed to contradict you, and that is where you learn something.

Campaigns you create by hand

Still a valid path. They get no hypothesis: inventing one after the fact would only pollute the memory. Instead, the analysis mines their performance for audience, creative and budget patterns, and records descriptive learnings.

The timeline

On the Strategy tab, time runs left to right. Each experiment is a span whose width is the time it stayed live; it closes on a verdict seal, and a hairline traces from that seal down into the learning it produced. Learnings feed a rail that runs on to now: where the next hypothesis comes from.

Detector signals (CPM anomaly, spend spike, creative fatigue, breakout post…) get their own lane: they are context, not protagonists. The Day / Week / Month scale changes both the axis and how they are grouped.

Start with Analyze performance: the strategist reviews the period, closes the experiments that can already be measured, and records what it learned.