Korean Appraisal Tech — Market Research

Overview

Research project investigating whether viable software products exist for the Korean residential appraisal market. The central question: can Step 4 of the legally mandatory appraisal procedure — 사례자료 수집 및 정리 (comparable transaction data collection and cleaning) — be automated for solo and small-firm 감정평가사 who lack access to the proprietary platforms built by large 감정평가법인? The research applied an invalidation-first method across four distinct hypotheses, mapping competitors, verifying API access, and characterizing buyer segments before building anything. The methodology directly produced one working product (fraud-screen, from Hypothesis A) and one structured research block with a clear go/no-go gate (Hypothesis B).

Problem

The Korean appraisal market is structurally split. Large firms (태평양감정평가법인 and peers) have proprietary platforms — 태평양 spent 18+ months building PAAS internally — that automate data collection workflows inaccessible to solo practitioners. Solo and small-firm 감정평가사 must manually execute Step 4 of the 감정평가에 관한 규칙 §9 procedure for every engagement: collect comparable transaction records from 실거래가 data, clean and normalize them, and document the selection rationale. Landvisor and other AVM platforms serve investors and real estate agents — not appraiser workflows. The gap between the proprietary large-firm tools and what solo practitioners actually have access to is the potential opportunity.

Constraints

• No way to determine actual practitioner pain level without talking to practitioners — all published information describes the procedure (감정평가규칙 §9), not the experienced friction
• The 감정평가사 market is small: approximately 5,000 licensed practitioners total, with solo/small-firm tier estimated at 30–40% of that — a small buyer pool for a niche tool
• AVM competitors (Landvisor, 부동산R114, 직방 데이터) serve adjacent use cases and could expand into appraiser-facing tools; their presence raises the build-vs-wait question
• Hypothesis A (forensic anomaly detection) and Hypothesis B (appraiser workflow) serve completely different buyers — conflating them in one project would have obscured both

Approach

Six research rounds total, structured as three rounds per cluster. Invalidation-first method: assume each hypothesis is wrong and actively search for evidence that kills it before looking for evidence that supports it. Competitive landscape mapped first (7 AVM and appraisal-adjacent competitors), then API access verified, then buyer segment characterized. Each hypothesis was assigned a go/no-go gate before any build work. Hypothesis A failed the 'no private buyer' kill condition — and revealed a different buyer (bank fraud prevention, not appraiser workflow) — which became fraud-screen. Hypothesis B cleared the data access and legal framework gates but is blocked on the practitioner interview gate.

Key Decisions

Tech Stack

• Research methodology (6-round adversarial deep research)
• Competitive analysis (7 AVM + appraisal-adjacent competitors)
• API assessment (실거래가, 공동주택공시가격, Landvisor)
• 감정평가에 관한 규칙 §9 (statutory framework)

Result & Impact

• 6
  Research rounds completed
• 4
  Hypotheses evaluated
• 7
  Competitors mapped
• 1 (fraud-screen)
  Products produced

Hypothesis A produced a working demo — fraud-screen — that cross-validated its findings against government investigation records and identified a buyer segment (bank fraud prevention) that the initial research had missed. Hypothesis B produced a structured research block: a clear go/no-go gate (one practitioner interview), specific interview questions, a WTP target range (₩50,000–80,000/month), and an abort date (2026-06-30). The research methodology that produced these outcomes — invalidation-first, buy/build gate sequencing, hypothesis spin-off discipline — is transferable across the portfolio.

Learnings

• The invalidation-first method surfaces wrong assumptions faster than a confirmatory approach. The ABANDON verdict on Hypothesis A was based on two assumptions that both turned out to be wrong (detection requires non-public data; government does this for free). Running the invalidation pass rigorously uncovered both before any build work started.
• Enforcement and prevention are different products with different buyers, even when they use the same underlying data. This distinction transformed a hypothesis with no buyer into a hypothesis with a clear buyer (bank fraud prevention). The distinction is obvious in retrospect but invisible when the research question is framed as 'who does anomaly detection?'
• A hypothesis spin-off is the right response when a secondary finding has a cleaner buyer and clearer product shape than the original hypothesis. Forcing Hypothesis A to remain part of a valuation research project would have distorted both.
• Research that produces a clear go/no-go gate (interview one practitioner by 2026-05-31, abort by 2026-06-30) is more valuable than research that produces a vague 'maybe.' The gate is the deliverable.