2nd/10_Wiki/Topics/AI_and_ML/손실 회피.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-손실-회피	손실 회피 (Loss Aversion)	10_Wiki/Topics	verified	self	Loss Aversion 손실 회피 Endowment Effect Adjacent Prospect Theory Loss	none	A	0.9	applied	behavioral-economics ux game-design prospect-theory kahneman		2026-05-10	pending	language framework typescript react

손실 회피 (Loss Aversion)

매 한 줄

"매 loss 의 매 psychological weight 의 매 gain 의 ~2배". Kahneman & Tversky 의 1979 Prospect Theory 핵심 발견 — $100 잃는 pain ≈ $200 얻는 pleasure. 매 modern UX, game economy, pricing, behavioral nudge 의 매 ubiquitous lever.

매 핵심

매 Prospect Theory 의 가치 함수

• Value function: 매 origin (status quo) 기준 의 asymmetric. Gain 측 concave, loss 측 convex + steeper.
• Loss aversion coefficient λ ≈ 2.0–2.5 (typical empirical range).
• Reference-dependent: 매 absolute wealth 가 아닌 reference point 기준.
• Diminishing sensitivity: $100→$200 의 gain 이 $1100→$1200 보다 큰 impact.

매 Cognitive 메커니즘

• Endowment effect: 소유한 것 의 valuation 이 동일 item 의 acquire price 보다 높음.
• Status quo bias: change 의 potential downside 가 upside 보다 무겁게 느껴짐.
• Sunk cost fallacy: 이미 잃은 것 을 회복하려는 irrational continuation.

매 응용

1. UX framing: "Save $20" (gain) vs "Lose $20 if you don't act" (loss) — loss frame 의 conversion 우수.
2. Game retention: streak / lose-progress 위협 (Duolingo streak freeze).
3. Pricing anchoring: trial → paid 의 loss aversion lever (already-using 의 endowment).

💻 패턴

Loss-framed CTA copy A/B test

const variants = {
  control: { headline: "Save 30% on Pro", cta: "Upgrade now" },
  loss:    { headline: "Don't lose Pro features tomorrow", cta: "Keep Pro access" },
};

function trackVariant(userId: string, variant: keyof typeof variants) {
  analytics.track("cta_view", { userId, variant, ts: Date.now() });
}

Streak system (Duolingo pattern)

interface Streak {
  count: number;
  lastActiveDay: string; // YYYY-MM-DD
  freezesAvailable: number;
}

function updateStreak(s: Streak, today: string): Streak {
  const days = daysBetween(s.lastActiveDay, today);
  if (days === 1) return { ...s, count: s.count + 1, lastActiveDay: today };
  if (days === 2 && s.freezesAvailable > 0)
    return { ...s, freezesAvailable: s.freezesAvailable - 1, lastActiveDay: today };
  if (days === 0) return s;
  return { ...s, count: 0, lastActiveDay: today }; // streak lost
}

Endowment via free trial → paid

// User accumulates value during trial; cancellation = loss frame
function trialSummary(usage: Usage) {
  return {
    headline: `You've created ${usage.docCount} documents`,
    losses: [
      `${usage.collaborators} teammates will lose access`,
      `${usage.gbStored}GB of files become read-only`,
    ],
    cta: "Keep your work — upgrade now",
  };
}

Game economy — escrow / forfeit on disconnect

// Loss aversion as engagement: forfeit currency if you abandon ranked match
function onMatchAbandon(player: Player, match: Match) {
  const penalty = match.entryFee * 1.5; // > entry fee — sharper sting
  player.currency -= penalty;
  notify(player, `You forfeited ${penalty} for leaving`);
}

Loss-aversion-aware notification timing

// Trigger reminder when user is closest to losing accrued value
function shouldNotifyAboutStreak(s: Streak): boolean {
  const hoursLeft = hoursUntilEndOfDay();
  return s.count >= 7 && hoursLeft < 6 && !s.notifiedToday;
}

Prospect-theory utility (decision sim)

def prospect_value(outcome: float, ref: float = 0, lam: float = 2.25,
                   alpha: float = 0.88) -> float:
    delta = outcome - ref
    if delta >= 0:
        return delta ** alpha
    return -lam * (-delta) ** alpha

매 결정 기준

상황	Loss aversion 활용?
Onboarding (no prior value)	약함 — gain frame 우수
Retention (existing value)	강함 — loss frame 우수
Pricing anchor	Endowment + price reference
Subscription cancel flow	Inventory of "what you'll lose"
Dark pattern territory?	매 ethical line — manipulative if value 가 없음

기본값: 매 user 의 actual accrued value 가 있을 때 만 loss frame. 매 fabricated scarcity 는 매 dark pattern.

🔗 Graph

• 부모: Behavioral-Economics · Prospect-Theory · Cognitive Biases
• 변형: Endowment-Effect · Sunk Cost Fallacy

🤖 LLM 활용

언제: copy A/B variant generation, retention email subject lines, game design lever brainstorm. 언제 X: 매 ethical line 의 judgment — manipulative copy 의 detection 은 human review 필수.

❌ 안티패턴

• Manufactured loss: 가짜 limited-time scarcity → 매 dark pattern, brand trust 손상.
• Loss frame on cold acquisition: prospect 가 endowment 없음 → backfire.
• Over-using λ=2.5 mental model: 매 individual variance 큼 (λ 0.5–4.0). Test, don't assume.
• Sunk cost induction: 매 player 의 deeper investment 강요 — 매 ethical issue.

🧪 검증 / 중복

• Verified (Kahneman & Tversky 1979 "Prospect Theory", "Thinking Fast and Slow" 2011, Duolingo retention experiments published 2020-2024).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — Prospect Theory + UX/game application patterns