ab-testing
bycoreyhaines31· 76 skills
When the user wants to plan, design, or implement an A/B test or experiment, or build a growth experimentation program. Also use when the user mentions "A/B test," "split test," "experiment," "test this change," "variant copy," "multivariate test," "hypothesis," "should I test this," "which version is better," "test two versions," "statistical significance," "how long should I run this test," "growth experiments," "experiment velocity," "experiment backlog," "ICE score," "experimentation program," or "experiment playbook." Use this whenever someone is comparing two approaches and wants to measure which performs better, or when they want to build a systematic experimentation practice. For tracking implementation, see analytics. For page-level conversion optimization, see cro.
How to install ab-testing?
coreyhaines31/marketingskills/ab-testing
$npx -y skills add coreyhaines31/marketingskills --skill ab-testingInstalls into the current project.
›Prefer a prompt? Paste this to your agent
Run `npx skills use "https://github.com/coreyhaines31/marketingskills" --skill "coreyhaines31/marketingskills/ab-testing"` and follow the generated skill instructions now. Read its complete output, redirecting it to a temporary file first if necessary. Resolve relative paths from the supporting-files directory it provides.
Use the skills in "https://github.com/coreyhaines31/marketingskills" that are relevant to the current task. Run `npx skills add "https://github.com/coreyhaines31/marketingskills"` and select the relevant skills, then follow their instructions.
SKILL.md
| 1 | # A/B Test Setup |
| 2 | |
| 3 | You are an expert in experimentation and A/B testing. Your goal is to help design tests that produce statistically valid, actionable results. |
| 4 | |
| 5 | ## Initial Assessment |
| 6 | |
| 7 | **Check for product marketing context first:** |
| 8 | If `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task. |
| 9 | |
| 10 | Before designing a test, understand: |
| 11 | |
| 12 | 1. **Test Context** - What are you trying to improve? What change are you considering? |
| 13 | 2. **Current State** - Baseline conversion rate? Current traffic volume? |
| 14 | 3. **Constraints** - Technical complexity? Timeline? Tools available? |
| 15 | |
| 16 | --- |
| 17 | |
| 18 | ## Core Principles |
| 19 | |
| 20 | ### 1. Start with a Hypothesis |
| 21 | - Not just "let's see what happens" |
| 22 | - Specific prediction of outcome |
| 23 | - Based on reasoning or data |
| 24 | |
| 25 | ### 2. Test One Thing |
| 26 | - Single variable per test |
| 27 | - Otherwise you don't know what worked |
| 28 | |
| 29 | ### 3. Statistical Rigor |
| 30 | - Pre-determine sample size |
| 31 | - Don't peek and stop early |
| 32 | - Commit to the methodology |
| 33 | |
| 34 | ### 4. Measure What Matters |
| 35 | - Primary metric tied to business value |
| 36 | - Secondary metrics for context |
| 37 | - Guardrail metrics to prevent harm |
| 38 | |
| 39 | --- |
| 40 | |
| 41 | ## Hypothesis Framework |
| 42 | |
| 43 | ### Structure |
| 44 | |
| 45 | ``` |
| 46 | Because [observation/data], |
| 47 | we believe [change] |
| 48 | will cause [expected outcome] |
| 49 | for [audience]. |
| 50 | We'll know this is true when [metrics]. |
| 51 | ``` |
| 52 | |
| 53 | ### Example |
| 54 | |
| 55 | **Weak**: "Changing the button color might increase clicks." |
| 56 | |
| 57 | **Strong**: "Because users report difficulty finding the CTA (per heatmaps and feedback), we believe making the button larger and using contrasting color will increase CTA clicks by 15%+ for new visitors. We'll measure click-through rate from page view to signup start." |
| 58 | |
| 59 | --- |
| 60 | |
| 61 | ## Test Types |
| 62 | |
| 63 | | Type | Description | Traffic Needed | |
| 64 | |------|-------------|----------------| |
| 65 | | A/B | Two versions, single change | Moderate | |
| 66 | | A/B/n | Multiple variants | Higher | |
| 67 | | MVT | Multiple changes in combinations | Very high | |
| 68 | | Split URL | Different URLs for variants | Moderate | |
| 69 | |
| 70 | --- |
| 71 | |
| 72 | ## Sample Size |
| 73 | |
| 74 | ### Quick Reference |
| 75 | |
| 76 | | Baseline | 10% Lift | 20% Lift | 50% Lift | |
| 77 | |----------|----------|----------|----------| |
| 78 | | 1% | 150k/variant | 39k/variant | 6k/variant | |
| 79 | | 3% | 47k/variant | 12k/variant | 2k/variant | |
| 80 | | 5% | 27k/variant | 7k/variant | 1.2k/variant | |
| 81 | | 10% | 12k/variant | 3k/variant | 550/variant | |
| 82 | |
| 83 | **Calculators:** |
| 84 | - [Evan Miller's](https://www.evanmiller.org/ab-testing/sample-size.html) |
| 85 | - [Optimizely's](https://www.optimizely.com/sample-size-calculator/) |
| 86 | |
| 87 | **For detailed sample size tables and duration calculations**: See [references/sample-size-guide.md](references/sample-size-guide.md) |
| 88 | |
| 89 | --- |
| 90 | |
| 91 | ## Metrics Selection |
| 92 | |
| 93 | ### Primary Metric |
| 94 | - Single metric that matters most |
| 95 | - Directly tied to hypothesis |
| 96 | - What you'll use to call the test |
| 97 | |
| 98 | ### Secondary Metrics |
| 99 | - Support primary metric interpretation |
| 100 | - Explain why/how the change worked |
| 101 | |
| 102 | ### Guardrail Metrics |
| 103 | - Things that shouldn't get worse |
| 104 | - Stop test if significantly negative |
| 105 | |
| 106 | ### Example: Pricing Page Test |
| 107 | - **Primary**: Plan selection rate |
| 108 | - **Secondary**: Time on page, plan distribution |
| 109 | - **Guardrail**: Support tickets, refund rate |
| 110 | |
| 111 | --- |
| 112 | |
| 113 | ## Designing Variants |
| 114 | |
| 115 | ### What to Vary |
| 116 | |
| 117 | | Category | Examples | |
| 118 | |----------|----------| |
| 119 | | Headlines/Copy | Message angle, value prop, specificity, tone | |
| 120 | | Visual Design | Layout, color, images, hierarchy | |
| 121 | | CTA | Button copy, size, placement, number | |
| 122 | | Content | Information included, order, amount, social proof | |
| 123 | |
| 124 | ### Best Practices |
| 125 | - Single, meaningful change |
| 126 | - Bold enough to make a difference |
| 127 | - True to the hypothesis |
| 128 | |
| 129 | --- |
| 130 | |
| 131 | ## Traffic Allocation |
| 132 | |
| 133 | | Approach | Split | When to Use | |
| 134 | |----------|-------|-------------| |
| 135 | | Standard | 50/50 | Default for A/B | |
| 136 | | Conservative | 90/10, 80/20 | Limit risk of bad variant | |
| 137 | | Ramping | Start small, increase | Technical risk mitigation | |
| 138 | |
| 139 | **Considerations:** |
| 140 | - Consistency: Users see same variant on return |
| 141 | - Balanced exposure across time of day/week |
| 142 | |
| 143 | --- |
| 144 | |
| 145 | ## Implementation |
| 146 | |
| 147 | ### Client-Side |
| 148 | - JavaScript modifies page after load |
| 149 | - Quick to implement, can cause flicker |
| 150 | - Tools: PostHog, Optimizely, VWO |
| 151 | |
| 152 | ### Server-Side |
| 153 | - Variant determined before render |
| 154 | - No flicker, requires dev work |
| 155 | - Tools: PostHog, LaunchDarkly, Split |
| 156 | |
| 157 | --- |
| 158 | |
| 159 | ## Running the Test |
| 160 | |
| 161 | ### Pre-Launch Checklist |
| 162 | - [ ] Hypothesis documented |
| 163 | - [ ] Primary metric defined |
| 164 | - [ ] Sample size calculated |
| 165 | - [ ] Variants implemented correctly |
| 166 | - [ ] Tracking verified |
| 167 | - [ ] QA completed on all variants |
| 168 | |
| 169 | ### During the Test |
| 170 | |
| 171 | **DO:** |
| 172 | - Monitor for technical issues |
| 173 | - Check segment quality |
| 174 | - Document external factors |
| 175 | |
| 176 | **Avoid:** |
| 177 | - Peek at results and stop early |
| 178 | - Make changes to variants |
| 179 | - Add traffic from new sources |
| 180 | |
| 181 | ### The Peeking Problem |
| 182 | Looking at results before reaching sample size and stopping early leads to false positives and wrong decisions. Pre-commit to sample size and trust the process. |
| 183 | |
| 184 | --- |
| 185 | |
| 186 | ## Analyzing Results |
| 187 | |
| 188 | ### Statistical Significance |
| 189 | - 95% confidence = p-value < 0.05 |
| 190 | - Means <5% chance result is random |
| 191 | - Not a guarantee—just a threshold |
| 192 | |
| 193 | ### Analysis Checklist |
| 194 | |
| 195 | 1. **Reach sample size?** If not, result is preliminary |
| 196 | 2. **Statistically significant?** Check confidence intervals |
| 197 | 3. **Effect size meaningful?** Compare to MDE, project impact |
| 198 | 4. **Secondary metrics consistent?** Support the primary? |
| 199 | 5. **Guardrail concerns?** Anything get worse? |
| 200 | 6. **Segment differences?** Mobile vs. desktop? New vs. returning? |
| 201 | |
| 202 | ### Interpreting Results |
| 203 | |
| 204 | | Result | Conclusion | |
| 205 | |--------|------------| |
| 206 | | Significant winner | Implement variant | |
| 207 | | Significant loser | Keep control, learn why | |
| 208 | | No significant difference | Need more traffic or bolder test | |
| 209 | | Mixed signals | Dig deeper, maybe segment | |
| 210 | |
| 211 | --- |
| 212 | |
| 213 | ## Documentation |
| 214 | |
| 215 | Document every test with: |
| 216 | - Hypothesis |
| 217 | - Variants (with screenshots) |
| 218 | - Results (sample, metrics, significance) |
| 219 | - Decision and learnings |
| 220 | |
| 221 | **For templates**: See [references/test-templates.md](references/test-templates.md) |
| 222 | |
| 223 | --- |
| 224 | |
| 225 | ## Growth Experimentation Program |
| 226 | |
| 227 | Individual tests are valuable. A continuous experimentation program is a compounding asset. This section covers how to run experiments as an ongoing growth engine, not just one-off tests. |
| 228 | |
| 229 | ### The Experiment Loop |
| 230 | |
| 231 | ``` |
| 232 | 1. Generate hypotheses (from data, research, competitors, customer feedback) |
| 233 | 2. Prioritize with ICE scoring |
| 234 | 3. Design and run the test |
| 235 | 4. Analyze results with statistical rigor |
| 236 | 5. Promote winners to a playbook |
| 237 | 6. Generate new hypotheses from learnings |
| 238 | → Repeat |
| 239 | ``` |
| 240 | |
| 241 | ### Hypothesis Generation |
| 242 | |
| 243 | Feed your experiment backlog from multiple sources: |
| 244 | |
| 245 | | Source | What to Look For | |
| 246 | |--------|-----------------| |
| 247 | | Analytics | Drop-off points, low-converting pages, underperforming segments | |
| 248 | | Customer research | Pain points, confusion, unmet expectations | |
| 249 | | Competitor analysis | Features, messaging, or UX patterns they use that you don't | |
| 250 | | Support tickets | Recurring questions or complaints about conversion flows | |
| 251 | | Heatmaps/recordings | Where users hesitate, rage-click, or abandon | |
| 252 | | Past experiments | "Significant loser" tests often reveal new angles to try | |
| 253 | |
| 254 | ### ICE Prioritization |
| 255 | |
| 256 | Score each hypothesis 1-10 on three dimensions: |
| 257 | |
| 258 | | Dimension | Question | |
| 259 | |-----------|----------| |
| 260 | | **Impact** | If this works, how much will it move the primary metric? | |
| 261 | | **Confidence** | How sure are we this will work? (Based on data, not gut.) | |
| 262 | | **Ease** | How fast and cheap can we ship and measure this? | |
| 263 | |
| 264 | **ICE Score** = (Impact + Confidence + Ease) / 3 |
| 265 | |
| 266 | Run highest-scoring experiments first. Re-score monthly as context changes. |
| 267 | |
| 268 | ### Experiment Velocity |
| 269 | |
| 270 | Track your experimentation rate as a leading indicator of growth: |
| 271 | |
| 272 | | Metric | Target | |
| 273 | |--------|--------| |
| 274 | | Experiments launched per month | 4-8 for most teams | |
| 275 | | Win rate | 20-30% is common for mature programs (sustained higher rates may indicate conservative hypotheses) | |
| 276 | | Average test duration | 2-4 weeks | |
| 277 | | Backlog depth | 20+ hypotheses queued | |
| 278 | | Cumulative lift | Compound gains from all winners | |
| 279 | |
| 280 | ### The Experiment Playbook |
| 281 | |
| 282 | When a test wins, don't just implement it — document the pattern: |
| 283 | |
| 284 | ``` |
| 285 | ## [Experiment Name] |
| 286 | **Date**: [date] |
| 287 | **Hypothesis**: [the hypothesis] |
| 288 | **Sample size**: [n per variant] |
| 289 | **Result**: [winner/loser/inconclusive] — [primary metric] changed by [X%] (95% CI: [range], p=[value]) |
| 290 | **Guardrails**: [any guardrail metrics and their outcomes] |
| 291 | **Segment deltas**: [notable differences by device, segment, or cohort] |
| 292 | **Why it worked/failed**: [analysis] |
| 293 | **Pattern**: [the reusable insight — e.g., "social proof near pricing CTAs increases plan selection"] |
| 294 | **Apply to**: [other pages/flows where this pattern might work] |
| 295 | **Status**: [implemented / parked / needs follow-up test] |
| 296 | ``` |
| 297 | |
| 298 | Over time, your playbook becomes a library of proven growth patterns specific to your product and audience. |
| 299 | |
| 300 | ### Experiment Cadence |
| 301 | |
| 302 | **Weekly (30 min)**: Review running experiments for technical issues and guardrail metrics. Don't call winners early — but do stop tests where guardrails are significantly negative. |
| 303 | |
| 304 | **Bi-weekly**: Conclude completed experiments. Analyze results, update playbook, launch next experiment from backlog. |
| 305 | |
| 306 | **Monthly (1 hour)**: Review experiment velocity, win rate, cumulative lift. Replenish hypothesis backlog. Re-prioritize with ICE. |
| 307 | |
| 308 | **Quarterly**: Audit the playbook. Which patterns have been applied broadly? Which winning patterns haven't been scaled yet? What areas of the funnel are under-tested? |
| 309 | |
| 310 | --- |
| 311 | |
| 312 | ## Common Mistakes |
| 313 | |
| 314 | ### Test Design |
| 315 | - Testing too small a change (undetectable) |
| 316 | - Testing too many things (can't isolate) |
| 317 | - No clear hypothesis |
| 318 | |
| 319 | ### Execution |
| 320 | - Stopping early |
| 321 | - Changing things mid-test |
| 322 | - Not checking implementation |
| 323 | |
| 324 | ### Analysis |
| 325 | - Ignoring confidence intervals |
| 326 | - Cherry-picking segments |
| 327 | - Over-interpreting inconclusive results |
| 328 | |
| 329 | --- |
| 330 | |
| 331 | ## Task-Specific Questions |
| 332 | |
| 333 | 1. What's your current conversion rate? |
| 334 | 2. How much traffic does this page get? |
| 335 | 3. What change are you considering and why? |
| 336 | 4. What's the smallest improvement worth detecting? |
| 337 | 5. What tools do you have for testing? |
| 338 | 6. Have you tested this area before? |
| 339 | |
| 340 | --- |
| 341 | |
| 342 | ## Related Skills |
| 343 | |
| 344 | - **cro**: For generating test ideas based on CRO principles |
| 345 | - **analytics**: For setting up test measurement |
| 346 | - **copywriting**: For creating variant copy |
Related
6 picksType
- grill-meA relentless interview to sharpen a plan or design.Jul 2026
- triageMove issues and external PRs through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs.Jul 2026
- lark-task飞书任务：管理任务、清单和任务智能体。创建待办任务、查看和更新任务状态、拆分子任务、组织任务清单、分配协作成员、上传任务附件、注册或注销任务智能体、更新任务智能体的主页数据、写入智能体任务记录。当用户需要创建待办事项、查看任务列表、跟踪任务进度、管理项目清单或给他人分配任务、为任务上传附件文件、注册注销任务智能体、更…Jul 2026
- lark-okr飞书 OKR：管理目标与关键结果。查看和编辑 OKR 周期、目标、关键结果、对齐关系、量化指标和进展记录。当用户需要查看或创建 OKR、管理目标和关键结果、查看对齐关系时使用。不负责：待办任务管理（lark-task）、日程/会议安排（lark-calendar）、绩效评估Jul 2026
- brainstormingYou MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior.Jul 2026
- grillingGrill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.Jul 2026
