CRO Methodology
How the lift gets found

The Experimentation OS

ASTEROI's CRO operating system for DTC subscription brands. Research becomes insights, insights become tests, tests become revenue — and every step traces back to the evidence.

5
Stages from research to test
6
Behavioral levers, one per test
100%
Traceable back to research
Built & run by Juan Cruz
The pipeline

Research becomes a test in five moves

No idea gets tested because it sounds good. Each stage is more specific than the last, and nothing advances unless the stage before it earned it. This is the spine the whole program runs on.

Top: broad & exploratoryBottom: one exact change
01
Research
Heatmaps, session recordings, surveys, analytics audits, competitor teardowns, ad-to-page analysis. The raw look at how customers actually behave.
The inputs
02
Insights
One finding from one method. Atomic — an observation, not a conclusion. Each carries its source and its evidence.
Atomic findings
03
Strategic Themes
Diagnosed problems, not tactics. Each backed by 3+ insights from different methods, written from the customer's point of view.
4–6 themes
04
Ideas
Directional bets — what to improve, never how. Broad enough to spawn several distinct tests, all traced to insights.
10–15 / theme
05
Experiments
One specific change. One behavioral lever. ICE-scored, traceable end-to-end, sized before launch. This is what ships to the split.
3–5 / idea
Behavioral levers

Every test pulls exactly one

A test that changes three things tells you nothing about which one worked. So each experiment commits to a single lever — clean attribution, real learning. Five are about perception; one moves the actual numbers.

01
Co
Cost
Perception of sacrifice. Price framing, shipping visibility, how the subscription commitment feels.
02
Tr
Trust
Confidence in the brand. Social proof, guarantees, clinical evidence, transparency.
03
Us
Usability
Ease of acting. Navigation, page flow, checkout friction, the mobile experience.
04
Cm
Comprehension
Understanding. Product education, ingredient clarity, subscription terms made obvious.
05
Mo
Motivation
Reasons to act now. Urgency, identity, aspiration, picturing the outcome.
numbers, not perception
06
Ec
Economics
The actual unit economics — price level, plan structure, margin. Used sparingly.
If a test pulls two levers, it's two tests. Split it.
The non-negotiables

Rules that keep it honest

A program is only as trustworthy as the discipline behind it. These four hold whether the answer is convenient or not.

01
Traceability
Every output links back: Experiment → Idea → Theme → Insight → Research. If a link is missing, it doesn't ship.
02
No padding
Ten strong ideas beat twenty mediocre ones. When the evidence runs out, we stop — we don't pad a list to hit a number.
03
No generic tactics
"Add social proof" is never an idea. Everything is grounded in this brand's research, audience and category — not best-practice boilerplate.
04
Honest ICE
Impact · Confidence · Ease, scored straight. Confidence stays low when the evidence is indirect — even when the idea sounds great.
Statistical rigor

We size it, then we don't peek

The fastest way to lie with an A/B test is to stop it the moment it looks good. So the math is fixed before launch and the engine — not a hunch — calls the result.

experimentation-os — /test-analyze
✓Power-sized before launch — the sample size N is committed up front, from the minimum effect worth detecting.
✓No peeking — one analysis, at the pre-committed N. No stopping early because the curve looks nice.
✓SRM check on every test — a sample-ratio mismatch flags broken randomization before anyone trusts the number.
✓Frequentist NHST — Evan Miller / Optimizely / VWO-grade math, the same convention serious programs run on.
// a pytest-tested stats engine runs the numbers — never prose, never vibes
Research → Insight → Test → Learn

This is how the lift gets found — every time.

See the system produce: GLAMRDiP Parachute David
JUAN CRUZCRO & Experimentation