Marketing experiment examples, structured to compound

EXWorked examples

Marketing experiment examples, structured to compound.

Five illustrative experiments across the loop: creative, offer, funnel, campaign and learning. Each declares one variable, sets its decision rule before launch, and reads out on margin rather than clicks. Numbers are deliberately omitted; the structure is the example.

EX-01Creative

Benefit-led headline vs. curiosity-led headline

Hypothesis: A headline that states the outcome plainly will attract buyers with stronger intent than one that withholds it, producing more contribution margin per visitor even if click-through falls.
Declared variable: Headline angle on the primary ad. Everything else — visual, offer, audience, placement — held constant.
Decision rule: Frozen before launch: run to a pre-set sample, then keep whichever arm wins on incremental contribution margin per user. Click-through is observed, not optimized.
Guardrails: Claims registry check: the benefit headline may only use approved claims. Conditional claims require their disclosure in the ad itself.
Readout: Contribution margin per user and allowable CAC by arm, with small samples flagged instead of averaged away.

EX-02Offer

Annual discount vs. bonus-month framing

Hypothesis: Framing the same economic value as two bonus months will convert better than a percentage discount, because the gain is concrete. The margin cost is identical, so any conversion lift is pure win.
Declared variable: Offer framing on the pricing page. Price, term and checkout flow unchanged.
Decision rule: Frozen before launch: decide on paid conversion weighted by 90-day retention of each arm, not day-one conversion. A framing that attracts churn-prone buyers loses.
Guardrails: Both framings map to the same SKU and the same dated contract terms, so finance sees one offer with two presentations, not two offers.
Readout: Payback period and 90-day retention by arm against versioned margin assumptions.

EX-03Funnel

Single-page checkout vs. two-step checkout

Hypothesis: Removing a step will raise completion, but may lower order quality if buyers commit before reviewing terms. The bet is that completion gains outweigh refund gains.
Declared variable: Checkout structure. Traffic source, creative and offer identical across arms.
Decision rule: Frozen before launch: winner decided on contribution margin net of refunds at 30 days, not on checkout completion rate.
Guardrails: Refund and dispute rates monitored as tripwires; a breach pauses the variant automatically pending review.
Readout: Completion, refund rate, dispute rate and net contribution margin by arm.

EX-04Campaign

Broad audience with creative-led targeting vs. interest stack

Hypothesis: A broad audience lets delivery find buyers the interest stack misses, lowering allowable CAC pressure — provided the creative does the qualifying.
Declared variable: Audience definition. Creative, budget ceiling, bid cap and placements held constant.
Decision rule: Frozen before launch: identical budget ceilings per arm; decide on contribution margin per acquired customer after a fixed learning window.
Guardrails: Budget ceilings enforced per policy; campaigns are created paused and only go live after the approved experiment is read back.
Readout: Allowable CAC headroom and margin per acquired customer by arm, with dimension IDs stable for later reuse.

EX-05Learning

Re-test of a prior winner after creative fatigue

Hypothesis: A headline angle that won six months ago has fatigued; the current baseline would lose to a refreshed angle built from the same approved claims.
Declared variable: Refresh of a previously winning angle vs. the fatigued incumbent. Offer, audience and funnel unchanged.
Decision rule: Frozen before launch: same decision rule as the original test, so results are comparable across time instead of confounded.
Guardrails: The learning record links both runs; the refresh reuses only claims still approved and unexpired in the registry.
Readout: Incremental margin vs. incumbent, appended to the original experiment's learning record.

This is what every experiment looks like inside Oddyssey: one declared variable, a decision rule set before launch, guardrails the workspace holds you to, and a readout tied to contribution margin.

Start free Back to the overview