Methodology
Marketing strategy earned from evidence.
Product runs the scientific method on your catalog and channels — every move starts with a hypothesis, runs as a real experiment, and ends with a clear call on what to keep, kill, or scale.
How we decide
Hypothesize. Test. Learn. Scale.
Guesses don’t get budget. Product proposes experiments with success criteria, runs the design you accept, analyzes the result, and turns winners into live campaigns.
01
Start from a hypothesis
Every proposal states what should change, why it might work, and how we’ll know — so you’re never approving a vague idea.
02
Pick the right design
Not every question needs the same test. Product chooses among A/B, holdouts, geo splits, bandits, and more based on what you’re trying to prove.
03
Run it for real
You accept to launch. Product applies the change set, measures against the control or baseline, and keeps the experiment honest through the full duration.
04
Read the result, then act
When the run ends, you get a conclusion — what moved, what didn’t, and what to scale next — so budget follows learning instead of opinion.
Experiment types
The designs Product can run.
Eleven experiment designs cover the questions marketers actually ask — from a simple creative swap to geo rollouts, spend ladders, and adaptive bandits.
A/B
Runs control and treatment campaigns in parallel, then compares them. The default when you can split traffic cleanly and want a head-to-head answer.
Before/after
Applies a change to one scope and compares results to the pre-run baseline. Useful when a parallel control isn’t available and timing is the comparison.
Holdout
Runs treatment against a true no-change holdout so lift isn’t confused with seasonality, promotions, or other market noise.
Multi-arm
Tests several campaign arms at once and crowns a winner — ideal when you have more than two credible options and want one clear pick.
Switchback
Alternates the change on and off over time on the same campaign. Built for cases where you can’t split audiences but can split time.
Geo split
Splits markets into treatment geos vs control geos as separate campaign arms — so you can roll out regionally with a real counterfactual.
Budget ramp
Steps daily budget through a dose ladder and finds the best response. Answers how much spend is enough before returns flatten.
Multivariate (MVT)
Tests factor combinations — like headline × image — and reports main effects so you learn which levers matter, not just which full combo won.
Bandit
Explores arms first, then shifts budget toward the winner automatically. Minimizes regret when you’re willing to adapt mid-flight.
Crossover
Runs treatment A then B in sequence (with optional washout) and compares periods — useful when the same audience must see both treatments over time.
Interrupted time series
Compares post-change results to a seasonal trend, with an optional control difference-in-differences, so before/after isn’t fooled by day-of-week or drift.
Stop guessing. Start proving.
Start free — import a catalog, link a channel, and let Product propose the next experiment with a real design.