CRO — CONVERSION OPTIMIZATION OS (OPERATIONAL)

Built as a no-fluff execution skill for systematic conversion rate optimization.

Structure: Core CRO fundamentals first. Advanced testing in dedicated sections. AI/ML optimization in clearly labeled "Optional: AI / Automation" sections.

Modern Best Practices (January 2026)

• Google Optimize sunset: Use VWO, Optimizely, or PostHog
• Statistical significance: https://www.evanmiller.org/ab-testing/
• CXL Institute: https://cxl.com/
• Baymard Institute UX: https://baymard.com/
• Cookie deprecation + stricter privacy defaults: prefer first-party measurement, validate assignment/tracking, and treat lifts as uncertain without clean instrumentation

When to Use This Skill

• Landing page optimization: Hero, CTA, proof, form optimization
• A/B testing: Hypothesis design, sample size, statistical significance
• Funnel analysis: Drop-off identification, micro-conversion mapping
• Form optimization: Field reduction, multi-step forms, friction removal
• Trust/credibility: Social proof, security signals, guarantees

When NOT to Use

• Brand awareness campaigns → Use marketing-paid-advertising
• User research methodology → Use software-ux-research
• Product analytics setup → Use marketing-product-analytics
• SEO/organic traffic → Use marketing-seo-complete

Expert: CRO Mental Model (Quick Calibration)

Use this to avoid local wins / global losses.

• CRO: Increase the rate of valuable commitments (purchase, qualified lead, activation) while protecting business outcomes (revenue, margin, LTV, support load).
• UX optimization: Reduce friction/errors so users can do what they already intend; good UX does not guarantee better conversions.
• Funnel optimization: Optimize the system across steps and handoffs (traffic quality → intent match → page → form/checkout → sales/onboarding → retention).
• Experimentation: A causal learning method; not every decision belongs in a test.

Do not delegate these to A/B tests (even with infinite traffic): legal/compliance/ethics, dark patterns, misleading claims, and irreversible brand trust decisions.

Core: CRO Framework

The CRO Process

1. ANALYZE → Identify conversion problems (data + qualitative)
2. HYPOTHESIZE → Form testable hypotheses
3. PRIORITIZE → Score by impact/effort (ICE/PIE)
4. TEST → Run A/B tests with statistical rigor
5. LEARN → Document results, iterate
6. IMPLEMENT → Roll out winners, test next

Conversion Rate Benchmarks

Note: Benchmarks vary significantly by industry. Use as directional only.

Core: Landing Page Optimization

Above-the-Fold Checklist

Every landing page needs these elements visible without scrolling:

Headline Formula

[Outcome] + [Timeframe/Ease] + [Without Pain Point]

Examples:
"Get 10 qualified leads per week without cold calling"
"File your tax return in 15 minutes with expert review"
"Double your email conversions without hiring a copywriter"

CTA Button Best Practices

CTA Button Optimization:

• Size: Large enough to tap on mobile (min 44px height)
• Color: Contrasts with page background
• Position: Above fold AND after key sections
• Text: First person ("Get My...") often outperforms second person
• Whitespace: Use spacing to isolate the primary CTA from competing elements; treat big lift claims as case-dependent and verify in your context

Trust Elements Hierarchy

STRONGEST TRUST SIGNALS (use at least 3):
├─ Customer logos (recognizable brands)
├─ Review score (4.5+ stars with count)
├─ Security badges (SSL, payment, compliance)
├─ Money-back guarantee
└─ Phone number visible

SUPPORTING TRUST SIGNALS:
├─ Customer testimonials (with photo, name, company)
├─ Case study snippets (specific metrics)
├─ "As seen in" media logos
├─ Team photos (for services)
├─ Live chat widget
└─ Physical address (for services)

User-Generated Content (UGC)

UGC often increases conversions in SaaS and e-commerce, but lift magnitude varies widely by category, placement, and traffic intent.

Implementation: Pull from G2, Capterra, or in-app feedback. Verify permissions before use.

Core: Form Optimization

Form Field Rules

2026 Benchmark: Average checkout = 5.1 steps, 11.3 fields (Baymard). Target ≤5 fields for lead gen.

Field Priority (Ask Only What You Need)

Multi-Step Form Pattern

Step 1: Low commitment (email)
├─ "What's your email?"
├─ Progress indicator: 1 of 3
└─ CTA: "Continue"

Step 2: Qualifying info
├─ Company size / Industry
├─ Progress indicator: 2 of 3
└─ CTA: "Almost there"

Step 3: Contact info
├─ Name / Phone (optional)
├─ Progress indicator: 3 of 3
└─ CTA: "Get My [Deliverable]"

Multi-step benefits:

• Commitment and consistency principle
• Captures partial data (even if abandoned)
• Feels less overwhelming
• Can qualify leads progressively

Core: A/B Testing Methodology

Hypothesis Template

IF we [change/add/remove X]
THEN [metric] will [increase/decrease] by [estimate]
BECAUSE [reasoning based on data/research]

Example:
IF we add customer logos to the hero section
THEN form conversion will increase by 15%
BECAUSE trust signals reduce perceived risk for new visitors

Sample Size Calculator

Minimum sample size formula (simplified):

n = (16 × p × (1-p)) / MDE²

Where:
- n = sample per variant
- p = baseline conversion rate
- MDE = minimum detectable effect (e.g., 0.10 for 10% lift)

Example:
Baseline CVR: 3% (0.03)
MDE: 20% relative lift (looking for 3.6% or higher)

n = (16 × 0.03 × 0.97) / (0.006)²
n ≈ 12,933 per variant

Total traffic needed: ~26,000 visitors

Quick reference:

Per variant. Multiply by 2 for total traffic needed.

Statistical Significance

Requirements for valid test:

• 95% confidence level (minimum)
• 80% power (default) unless you have a reason to change it
• Run for at least 1-2 full business cycles (7-14 days)
• Don't peek and stop early (increases false positives)
• Document before test: hypothesis, primary metric, guardrails, sample size, duration
• Avoid post-hoc slicing; pre-register segments or adjust for multiple comparisons

Reality check (expert defaults):

• Statistical significance does not mean the change is worth shipping (check practical impact + guardrails)
• Ignore "significant" results when experiment integrity is in doubt (tracking issues, traffic mix shifts, SRM, broken randomization)
• Stop early only for clear harm (guardrail breaches) or invalidity (instrumentation/assignment problems), not for "early wins"

Experiment Integrity (2026 Default Checks)

• Assignment sanity: A/A test periodically; check SRM on day 1 and day 3
• Tracking sanity: confirm event definitions, dedupe, cross-domain, and consent-mode behavior before interpreting results
• Contamination: avoid showing multiple variants to the same user across devices/sessions; prefer stable IDs when possible
• Change control: freeze other major changes to the same flow during the test window

CUPED: Faster Tests via Variance Reduction

CUPED (Controlled-experiment Using Pre-Existing Data) can reduce variance by ~40-60%, allowing tests to reach significance faster.

When to use: High-traffic sites where test velocity matters. See advanced-testing.md for implementation details.

Test Prioritization: ICE Framework

ICE Score interpretation:

• 8-10: High priority, test immediately
• 5-7: Medium priority, add to queue
• 1-4: Low priority, revisit later or skip

Core: Funnel Analysis

Funnel Diagnostic Framework

STEP 1: Map your funnel
Page Visit → Key Action → Form Start → Form Complete → Confirmation

STEP 2: Measure drop-off at each step
├─ Page Visit to Key Action: ___% (bounce rate inverse)
├─ Key Action to Form Start: ___%
├─ Form Start to Complete: ___%
└─ Complete to Confirmation: ___%

STEP 3: Identify biggest drop-off
Biggest percentage drop = highest priority to fix

STEP 4: Diagnose root cause
├─ High bounce? → Relevance, load speed, messaging
├─ Low engagement? → Content, CTA visibility
├─ Form abandonment? → Form friction, trust
└─ Checkout drop? → Pricing, shipping, trust

Expert note: The "biggest drop-off" is not always the best target. Confirm it's a defect (not intentional filtering), not a measurement artifact, and not caused upstream (traffic quality / offer mismatch).

Micro-Conversion Mapping

Heatmap & Recording Analysis

What to look for:

• Click heatmaps: Are users clicking CTAs? Clicking non-clickable elements?
• Scroll maps: Where do users stop scrolling? Key content below fold?
• Session recordings: Where do users hesitate? Rage clicks? Form confusion?
• Form analytics: Which fields cause abandonment? Error patterns?

Reference: Triage, Speed, SOPs

For page speed targets, CRO triage decision tree, operating cadence, and anti-patterns, see references/triage-and-ops.md.

Templates

Expert: Hypothesis Quality (Silent Failure Checklist)

A good CRO hypothesis is not "change X to raise CVR." It must specify mechanism and risk.

Strong hypothesis includes:

• Which constraint it targets: clarity, trust, motivation, friction
• Who it's for: segment/intent/channel/device (at least one)
• What moves: primary metric + guardrails (value, quality, downstream)
• Why it should work: evidence + mechanism (not vibes)

How CRO fails silently (common):

• Conversions go up but value goes down (lower-quality leads, higher refunds/chargebacks, worse retention)
• Overall looks flat but a high-value segment is harmed (mix effects hide damage)
• "Win" is novelty or seasonality; it doesn't repeat

Use assets/ab-test-plan.md to pre-register guardrails and invalidation criteria.

References

International Markets

This skill uses US/UK defaults. For international CRO:

Auto-triggers: When your query mentions regional markets or cultural adaptation, both skills load automatically.

Related Skills

• marketing-geo-localization — International markets, cultural CRO
• marketing-leads-generation — Lead capture strategies
• marketing-paid-advertising — Traffic sources
• marketing-seo-complete — Page speed, Core Web Vitals
• software-ui-ux-design — Design patterns
• software-ux-research — User research methods

Usage Notes (Claude)

• Stay operational: return checklists, audit results, test plans
• Always include statistical significance requirements for testing
• Recommend qualitative research for low-traffic sites
• Use benchmark ranges, not absolute numbers
• Do not invent conversion data; state "varies by industry" when uncertain