Build a small campaign experiment whose audience evidence, claims, variable and decision rule remain visible.
Separate observed customer language from assumptions before asking AI for a campaign idea.
AI can organize interviews, support questions and search notes, but it cannot turn an invented persona into evidence. Preserve the source beside each theme and keep contradictions visible.
Three support tickets ask who owns payroll after onboarding. That is evidence of a recurring question, not proof that customers will buy a new payroll service.
Collect five permitted observations. Record the source, date, exact wording, interpretation and alternative explanation for each.
Group these audience observations by recurring task or question: [notes]. Preserve source labels. Return evidence, interpretation, contradiction and missing research. Do not create a persona or estimate demand.
Define one audience, approved offer, goal and channel boundary.
A campaign brief prevents an early idea from becoming an unapproved promise. It should state the evidence, offer facts, forbidden claims, deliverables, metric and approver.
The campaign invites existing bookkeeping clients to five research interviews. It does not advertise a payroll service that has not been approved.
Choose one campaign goal and fill the audience evidence, offer facts, call to action, channel, consent rule, deliverables, primary metric and guardrails.
Create a campaign brief from these approved inputs: [inputs]. Mark missing decisions [NEEDS INPUT]. Do not invent urgency, discounts, proof, audience beliefs or channel permission.
Produce two campaign versions that differ in one meaningful variable.
If headline, offer, image and audience all change together, the result cannot explain which decision mattered. Controlled variants keep everything fixed except the selected test variable.
Version A leads with the 30-minute format; version B leads with the weekly-planning outcome. The offer and registration action stay the same.
Choose one variable—headline angle, proof order or call-to-action wording—and create two versions with all other facts held constant.
Create versions A and B from this approved campaign brief: [brief]. Change only [variable]. Return a difference table and flag any other wording that changed. Do not predict a winner.
Distinguish measured outcomes from explanations and choose a defensible next step.
A report should show delivery, metric definitions, data quality and the planned comparison before suggesting why a result changed. Correlation is not automatically causation.
Version B recorded a higher registration rate in this run. The report notes the audience split and sample size before deciding whether to repeat the test.
Enter the planned metric, guardrails, delivery data, observed result, data limitations, alternative explanations and decision rule.
Review this campaign record: [record]. Separate observed results, possible explanations and unsupported claims. Recommend continue, repeat, revise or stop using the supplied decision rule.
Yes. All four lessons exercises and answer explanations are available on this page without an account.
No. The process is platform-neutral and asks you to verify current controls in the platform you use.
No. AI can organize observations but causal conclusions require an experiment and evidence that support them.
