Korean Cases: Advertising and Commerce

Key takeaways

• Naver logs associate advertising stores with +118.1% transaction value versus nonadvertisers in log-coefficient notation, approximately +226% after transformation. Stores able to spend more self-select into advertising.
• Vendor store-conversion aggregates give a 2.06% mean and 0.72–5.05% by industry. The top-100-store sample from 2021 may overstate new-store performance.
• No source for the 5–8% live-commerce conversion claim discloses its sample; 20% refers to top Chinese wanghong. Naver cumulative buyers divided by views is about 1%.
• We found two Korean advertiser conversion-lift cases, but neither discloses enough sample information to provide an adjustment multiplier.
• Lift depends on the metric and measurement scope. The same Kurly feature produced +0.3 percentage points purchase conversion in an A/B test and +595% in a narrower before/after comparison.

Grades follow How to Read the Evidence and appear at the end of the first table cell. Select inputs in the Korean Conversion Atlas.

Paid Channels

Naver Transaction-Log Analysis

The D-Ad Report and AI RIDE Report use Naver data analyzed by university researchers. They are not peer-reviewed, and Naver sells the ads; it also co-designed AI RIDE campaigns. Still, few Korean advertising sources disclose both samples and models.

ADVoost's +165% attributed conversions versus +66% transaction value, a 2.5-times ratio of lifts, is not a reported-to-incremental multiplier. Treatment classification uses purchase conversions, mechanically favoring that outcome. The D-Ad Report provides a less distorted illustration of attribution versus total outcomes: four products yield about twice the conversion revenue of three, yet total store transaction values are effectively equal. Different stores use different combinations, so this is not a direct comparison; it is only consistent with the warning in Reported and Incremental Conversions.

AI RIDE reports spending elasticities of 0.31% for purchase conversions, 0.09% for transaction value, and 0.17% for ROAS per 1% spend increase. Rising ROAS with spend contradicts usual diminishing-return expectations and warrants scrutiny of sample selection.

Increasing Budgets

Naver's advertiser-center Place article compares 90 days before/after, with undisclosed sample (D).

Elasticity is ln(click multiple) ÷ ln(spend multiple). The source describes rising performance, but the figures imply rapidly increasing CPC. No conversions/revenue are reported, and only budget-increasing advertisers are included. See Creative, Retargeting, and Saturation for curve assumptions.

Consumer Recall

Nasmedia 2025 NPR reports recalled purchases following ad exposure: search 21%, short-form video 15%, Kakao messenger 12%, social feeds 11% (January 2025, 1,900 respondents; media-representation firm, C). Social feeds moved from 25% to 11% in one year, short-form from 6% to 15%. Such survey variation makes neither levels nor rankings suitable for channel selection.

A Korea Press Foundation 2024 survey interviewed 3,000 people at home. Among social users, 63.7% had clicked ads and 28.2% had purchased; blog users reported 39.4% and 12.8%. These are experience/recall responses with an unverified recall window, not conversion rates (C via press).

Search-Ad Papers

Platform-Published Cases

All are platform or agency sales material, grade D. Keep each result tied to its stated comparison. Cases with clear interpretive lessons come first.

Akei and Catalic are the only Korean advertiser conversion-lift cases found; absent samples prevent deriving calibration multipliers. A pricing reference is CHZZK 15-second instream list CPM of KRW 13,000, with 30–60% discounts by monthly spend. This is a rate card, not realized cost.

Stores and D2C

Purchase Conversion Baselines

BIGIN Ecommerce Trends Vol. 04 reports a 2.06% mean, with a confirmed sample of the top 100 stores by event count and about 237.17 million sessions, February–July 2021. These appear to be the CRM vendor's customers, potentially skewed toward mature stores. The summary does not specify whether the denominator is sessions or visitors, or how the average was calculated. The full-report link is broken; this relies on the summary.

• Children / baby5.05%
• Pets3.21%
• Food2.95%
• Beauty2.93%
• Fashion1.87%
• Accessories1.74%
• Sports1.34%
• Furniture0.72%

Vertical line: Overall mean 2.06%Source: BIGIN

• Organic search2.88%
• Direct2.85%
• Email2.59%
• Paid search2.38%
• Referral1.61%
• Other1.55%
• Affiliates1.30%
• Social1.13%
• Display0.34%
• Other ads0.17%

Source: BIGIN

AI-referred visits to Cafe24 D2C stores converted at 0.85%, up from 0.50% a year earlier (Q2 2026, about 850,000 visits, C via press). ChatGPT: 0.96%; Gemini: 0.57%; Perplexity: 0.5%. Cafe24's difference-in-differences analysis of paid features reports conversion gains of 29.9% for SEO plus AI product pages and 13.5% for SEO alone, about 2,300 stores (C via press). Adoption is self-selected and baseline rates absent.

Repeat Purchase and Channels in Filings

Definitions differ across companies; rows are not directly comparable.

Company Experiments and Before/After Comparisons

Kurly's no-results search improvement measures one feature in two ways. A roughly two-week A/B test gave purchase conversion +0.3 percentage points and purchase amount +0.7%; four weeks after full rollout, purchased units in no-result cases were +595% versus before. The latter concerns only the 6–7% of searches returning no results, with a different outcome and denominator as well as design.

For conversion windows, Zigzag's mean search-to-payment time of 3.4 days offers a reference: beauty 2.4, lifestyle 4.4 days in 2023 (D). Searches never leading to purchase are excluded.

Live Commerce

Tracing widely quoted “live-commerce conversion rates” reveals different or undisclosed denominators. The chart groups them to clarify provenance, not to compare performance.

• Top Chinese wanghong20%D
• Best Musinsa episode14%D
• Brokerage / solution vendor5% ~ 8%D
• Solution vendor 20253.5%D
• Grip average3%D
• One TMON broadcast2.2%D
• Naver cumulative, calculated0.89% ~ 1%C

The three data sources yield around 1%, but the first two cover overlapping Naver periods and use buyers versus units sold. They are not three independent confirmations. Follow this book's rule on ratios of averages and use them only for scale. Kakao reportedly did not disclose purchase conversion; no later official figure was found.

Naver D-Commerce 2021 reports before/after unit-sales +49% and sales value +48% among live-broadcast stores, versus +7% and +16% among nonbroadcast stores (four categories, 5,026 stores, 827 live stores, B). Excluding sales during broadcasts still gives +39% and +37%. Sellers choose whether to broadcast, and baseline daily unit sales already differ, 77 versus 30; this is not causal.