Marketing Atlas
한국어
% Conversion optimization

Natural-language experiment search

자연어 실험 이력 검색

Find past tests and outcomes with a question before repeating work.

···
html
<div class="history-demo"><div class="history-search">ASK: Shorter checkout tested? <span class="search-cursor"></span></div><div class="history-result"><b>MATCH · FORM STEPS</b><span>3 fields → 2 fields</span><strong>Completion +3.1%</strong></div><div class="history-next">NEXT: Test a different friction point</div></div>
css
.history-demo{width:min(84%,500px);font:700 clamp(10px,3.2vmin,15px)/1.2 sans-serif}.history-search{border:2px solid var(--fg);padding:clamp(7px,1.6vmin,14px);white-space:nowrap}.search-cursor{display:inline-block;width:2px;height:1em;background:var(--accent);vertical-align:middle;animation:blink .8s step-end infinite}.history-result{border-left:4px solid var(--accent);background:var(--surface);padding:clamp(8px,1.8vmin,15px);margin-top:clamp(8px,2vmin,16px);animation:reveal 3s infinite}.history-result b,.history-result span,.history-result strong{display:block}.history-result span{margin:4px 0;color:var(--muted)}.history-result strong{color:var(--accent)}.history-next{margin-top:clamp(7px,1.5vmin,13px);color:var(--muted)}@keyframes blink{50%{opacity:0}}@keyframes reveal{0%,8%{opacity:.15;transform:translateY(6px)}25%,90%{opacity:1;transform:none}100%{opacity:.15}}
···

Experiment hypothesis prioritization실험 가설 우선순위

Rank the next tests by likely impact, confidence, and effort.

···

AI experiment qualityAI 실험 품질 관리

Separate the number of generated tests from validated learning.

···

A/B testingA/B 테스트

Randomly assign two versions to compare their effect on conversion.

···

Conversion funnel analysis전환 퍼널 분석

Compare step completion rates to locate the largest drop-off.

Natural-language experiment search finds previous tests by the meaning of a question, even when exact terms differ. Statsig introduced repository search in 2026 and specified a beta for Warehouse Native customers at the time.

The demo matches “Have we shortened checkout?” to a related test and result note. Review the old record before stating how a new hypothesis differs.

No search hit does not prove the test was never run. Check coverage and permissions, then read the original audience and time window.

When to use

Use it before designing a test to see whether a similar change was already evaluated.

Open as page ↗