# Counterintuitive Tests > Generate ideas that don't sound logical, then validate them with cheap behavioural experiments — Rory Sutherland's 'the opposite of a good idea can be a good idea' plus revealed-preference testing. Use when the obvious/logical solution is saturated and the user wants genuinely different options, asks 'what's a creative/unconventional angle', 'how do we stand out', 'should we run a test/experiment', 'how do we know what people actually want', or is about to trust a survey/focus group. Covers deliberate inversion of the obvious move, why stated preferences lie, and designing small reversible experiments that watch behaviour. Inspired by Rory Sutherland's *Alchemy*. Routed to from psycho-logic. 日本語の相談（「競合と同じ案しか出ない」「アンケートは好評なのに売れない」「思い切った案を安く試したい」）にも使う。中文咨询（“只能想出和竞品一样的方案”“问卷好评却卖不动”）也适用。 - Skill: `meikocho1/counterintuitive-tests` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add meikocho1/counterintuitive-tests` - Raw SKILL.md: https://api.skillmd.com/api/skills/meikocho1/counterintuitive-tests/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: meikocho1 (https://skillmd.com/u/meikocho1) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/meikocho1/counterintuitive-tests --- # Counterintuitive Tests — invert the obvious, then watch behaviour Two moves: (1) deliberately generate the ideas that sound *wrong*, because the sensible ideas are already taken; (2) never trust what people *say* about them — run a cheap, reversible experiment and watch what they *do*. ## Why the obvious idea is the wrong default Logical optimisation converges: every competent competitor finds the same "best practice," so it stops being an advantage. The remaining opportunities are the moves that don't survive a rational business case — which is exactly why nobody has tried them. "Doesn't make sense" is not a reason to reject an idea; it's a reason it might still be available. - **Red Bull** launched a small, expensive can of bad-tasting drink. Every logical signal said fail. The "wrongness" *was* the positioning — it must do something. - **Raising the price** sometimes raises sales (it changes the signal — see costly-signaling). - **Removing** features/options/steps often beats adding them (paradox of choice). ## Why you must test behaviour, not opinions > People don't think what they feel, don't say what they think, and don't do what they say. Surveys and focus groups capture *post-hoc rationalisations*, not causes. Nobody in a focus group asked for Red Bull. Stated preference is cheap talk; **revealed preference** (what they actually clicked, bought, kept) is the only honest data. So your job isn't to *ask* whether the counterintuitive idea works — it's to *expose people to it* and measure behaviour. ## How to apply 1. **Write the obvious solution down.** The thing every competitor would do. This is your "do not just do this" baseline. 2. **Invert it on purpose.** For each obvious move, generate its opposite and ask "under what circumstances would the opposite be *better*?" (Charge more. Offer less. Make it slower but nicer. Hide the price. Add friction. Remove a feature.) Keep the ones that aren't insane on inspection. 3. **Design a cheap, reversible test.** The whole point of a counterintuitive idea is you can't reason your way to the answer — so make the experiment so cheap and reversible that you don't need to. A/B test, a soft launch to one segment, a fake-door, a landing page, one store. Small, fast, undoable. 4. **Measure behaviour, not approval.** Define the behavioural metric *before* running (clicks, purchases, retention, completion). Ignore "would you…?" survey answers about the idea. 5. **Let surprising-but-real results win.** If the illogical version beats the logical one on behaviour, ship it — even if you can't fully explain why. It doesn't have to make sense to work. ## How to think about the experiment - Bias toward **many small reversible bets** over one big committed one (barbell: safe core, cheap weird experiments). - A test that can't *fail* (no clear behavioural metric, no control) isn't a test — it's theatre. - Negative results are wins: a cheap experiment that kills a bad idea saved you the expensive version. ## 日本語の使用例 - 「競合と同じ訴求しか思いつかない」→ まず**当たり前の解**を書き出し、その**逆**を1つずつ生成（"もっと高く" "機能を減らす" "あえて遅く・丁寧に" "価格を最後まで見せない"）。安く・元に戻せる形で検証。 - 「アンケートでは好評だったのに売れない」→ **言うこと≠やること**。アンケート（安い口先）ではなく、ランディングページや少数セグメントへのソフトローンチで**行動（購入・離脱）**を測る。 - 「値上げしたら売れなくなりそう」→ 直感で却下せず、1セグメント・期間限定の**可逆なA/B**で行動を見る。論理ではなく結果に判断させる。 - 葬儀/シニア領域：「分かりやすさ＝情報を増やす、が定石」→ 逆に**選択肢と情報を削る**版を作り、問い合わせ完了率で比較する。 ## Cross-check against the canon Sanity-check inversions and experiment designs against these; if one contradicts your plan, name it and justify the exception. When a claim is load-bearing or the user asks for evidence or fresh examples, web-search the named theory or case and cite what you find. - **Paul Samuelson, revealed preference** — the economics behind "watch what they do": choices, not statements, define preference. - **Barry Schwartz, *The Paradox of Choice*** — why "offer less" is a serious inversion, not a gimmick. - **Nassim Taleb, the barbell** — keep the core safe and make the experiments small and reversible; never bet the company on an inversion. - **Byron Sharp, *How Brands Grow*** — the empirical brake: some counterintuitive moves (e.g. abandoning reach to court loyalists) reliably fail in his data. Test against evidence, not just cleverness. ## Related skills - **psycho-logic** — the router; use it to decide which obvious move to invert - **reframing** / **costly-signaling** — common sources of the counterintuitive idea you'll test - **experience-design** — many counterintuitive wins are journey changes; test them the same way - **ab-testing** / **marketing-psychology** — mechanics of running tests and the biases involved (separate plugins — only if installed)