How we measure
buyability.

Overview

The Buyability Score is the share of agent-driven product impressions that translate into a transaction a U.S. shopper can actually complete. It combines two layers — discovery (which products surface in agent results) and verification (whether each surfaced product satisfies six purchasability conditions) — into one weighted percentage per brand, per platform, per quarter.

1 — Query design

We maintain a curated query corpus of 2,400 commercial questions a U.S. shopper might plausibly ask an AI agent for Korean beauty products. Queries are weighted by U.S. AI shopping volume using the Bain Q1 2026 benchmark and refreshed monthly.

• Recommendation queries — "best Korean retinol for sensitive skin"
• Branded queries — "where to buy Beauty of Joseon sunscreen in the U.S."
• Comparison queries — "COSRX snail mucin versus Mixsoon bean essence"
• Price-aware queries — "Anua heartleaf toner cheapest U.S. seller"
• Variant-specific queries — "Glow Recipe watermelon dew drops 60 ml"

2 — Five surfaces, equal weight

Each query is run against every major U.S. AI shopping surface: ChatGPT, Gemini, Claude, Perplexity, and Microsoft Copilot. We use the public consumer-facing endpoint for each surface with U.S. geo-context and an English-language prompt. Rendered output is captured, parsed for product recommendations, and normalized to canonical SKUs via the Korean specialization layer.

We weight all five surfaces equally per query. We do not adjust for surface market share, because market share shifts faster than the measurement cadence. Brands receive both blended and per-surface scores in their dashboard.

3 — The Verifier

Every recommended SKU is sent into a headless-browser verification fleet that visits the canonical U.S.-facing PDP and runs six checks per page:

• Inventory — variant in stock, available to a U.S. shipping ZIP
• Price truth — price quoted by the agent matches PDP within ±2%
• Page load — full PDP render under 5 seconds, no error states
• Schema validity — schema.org Product markup complete and well-formed
• Variant match — recommended size, shade, or scent exists at the listed SKU
• Checkout path — add-to-cart and shipping calculation succeed for U.S. address

Failures are dual-classified: by failure type (inventory, pricing, page, schema, variant, fulfillment) and by funnel stage (discoverability, click-through alignment, prompt-matched landing, on-site assistance, frictionless checkout). The dual classification turns the score from a number into a remediation roadmap.

4 — Korean specialization layer

Generic verifiers cannot measure Korean brands accurately. The Tigerlayer Korean layer adds:

• Hangul + romanization normalization — 조선미녀 / Joseon / Beauty of Joseon collapse to one canonical brand entity
• Multi-listing reconciliation — the same SKU across YesStyle, Soko Glam, Sephora.com, and Amazon resolves to one Buyability number per brand
• KFDA → FDA ingredient mapping — regulatory cross-reference for compliance-aware scoring
• Korean-to-U.S. unit conversion — mL ↔ fl oz with consumer-context heuristics
• Counterfeit + grey-market scoring — pattern detection on Amazon and TikTok Shop listings against known brand authorization data

5 — Recoverable revenue

Recoverable revenue is computed per brand from four observable inputs:

Recoverable = AI impressions × broken-impression rate × AI conversion rate × average order value

AI impressions are measured per Tigerlayer query corpus weighting. Broken-impression rate is the Verifier output. AI conversion rate uses the industry Q1 2026 benchmark of 12.3% unless the customer provides their own measured value. Average order value is brand-supplied. We market the figure as directionally accurate within 15%, not invoice-grade. Customers do not need to believe the model is perfect; only that it is directionally correct and materially undercounting the problem.

6 — The protocol layer

Tigerlayer is built against the four protocol standards that have stabilized AI commerce since late 2025:

• OpenAI Product Feed Specification — September 2025
• Stripe Agentic Commerce Protocol (ACP) — September 2025
• Google Universal Commerce Protocol (UCP) — version 2026-04-08 in production
• Anthropic Model Context Protocol (MCP) — donated to the Linux Foundation December 2025

These protocols give us standardized surfaces from which to extract what each AI agent ecosystem actually shows shoppers. Building this product against the fragmented APIs of 2024 would not have been viable.

7 — Legal and platform posture

Tigerlayer respects robots.txt and conservative crawler etiquette across all platform surfaces. For brand customers, we offer an authenticated mode in which the brand grants us scoped API or MCP access, in exchange for accuracy guarantees and pricing discount. We expect 60–80% of brand customers to opt into authenticated mode within 18 months because it materially improves score reliability.

On the platform side, we structure our buyability data as a quality-control feed that AI surfaces can license directly to maintain user trust — turning the platforms from a risk into a customer.

About Tigerlayer

Tigerlayer Inc. is a U.S.-incorporated company building the buyability layer for AI commerce, with a Korean specialization wedge. We believe AI shopping has crossed the threshold from emerging channel to primary discovery surface, and that no system currently exists to measure whether the recommendations AI surfaces actually convert. Tigerlayer is that system.

Headquartered in the United States, with operations in Seoul. For partnership inquiries: hello@tigerlayer.com.

Citing this work

Tigerlayer Inc. (2026). K-Beauty Buyability Index, Q2 2026. Retrieved from tigerlayer.com/buyability-index. Updated quarterly.