條碼覆蓋率分析報告

Analysis date: 2026-03-31 Source: channel_master.csv (2,490,077 records, 67 channels) Purpose: Difficulty coefficient for barcode collection contract pricing

Executive Summary | 重點摘要

Critical finding: The majority of Taiwan's largest e-commerce platforms (PChome 980K, ETmall 425K, Poya 28K, 7-ELEVEN 13K) have ZERO barcode data in their crawled product listings. Barcode collection for these channels requires fundamentally different approaches (image OCR, manual lookup, cross-channel matching by product name) rather than direct scraping.

Section 1: Per-Channel Barcode Coverage | 各通路條碼覆蓋率

Tier 1: Channels with Excellent GTIN Coverage (>80%)

Tier 2: Channels with Partial Coverage (5-80%)

Tier 3: Channels with Zero Valid GTIN

Section 2: Key Difficult Channels Deep Dive | 重點困難通路詳析

A. PXGo (全聯線上購 - 純量販通路)

Assessment: PXGo is the EASIEST among the key channels. Over 92% of products have valid GTINs. The remaining 7.6% are store-internal codes (likely private label or fresh food). This channel is a strong candidate for cross-referencing other channels' products.

Top categories with coverage: Condiments (99.7%), Hair care (100%), Biscuits (100%), Oral care (100%), Frozen meals (100%).

B. PXBox (全聯線上購 - 綜合電商)

Assessment: PXBox is a mixed marketplace. Its FMCG grocery products (biscuits 88.9%, hand tools 51.9%) have decent barcode rates, but its non-grocery products (phone cases 0.8%, furniture 0.3%, fashion 2.6%) have nearly zero. The 50,943 valid GTINs overlap heavily with PXGo (6,086 shared = 66.5% of PXGo's catalog) and PXMart_Mega (8,167 shared = 84.0% of PXMart_Mega).

C. PXMart Mega (全聯量販 Mega)

Assessment: Similar to PXGo -- physical-store-oriented FMCG data with excellent barcode coverage. Almost all categories at 100%. Near-complete overlap with PXBox GTINs (84% of Mega's barcodes found in PXBox).

D. Carrefour (家樂福)

Assessment: Carrefour is moderately difficult. Over 20% of products have barcodes that failed validation (likely internal SKUs formatted differently or source data quality issues). The valid 28,420 GTINs are valuable -- they overlap 23.7% with PXBox. Category-wise, stationery (13.9%), kitchenware (21.3%), pet supplies (23.3%) and food categories tend to have better coverage than electronics/furniture (0-5%).

Special note on invalid_format barcodes (24,441): These may contain recoverable barcode data that just needs reformatting. Investigating this could unlock an additional ~20% coverage.

E. Cosmed (康是美)

Assessment: Cosmed's crawled data has ZERO product GTINs. The 44,005 "barcodes" are all ISBNs from a books section. The top categories are luxury brands (Michael Kors, Hermes, Burberry, Chanel) and books -- suggesting the scrape captured a marketplace/department store view rather than the core drugstore assortment. Barcode collection would require entirely new crawling strategy or cross-channel matching.

F. Poya (寶雅)

Assessment: Zero barcode data. However, brand names and detailed category hierarchy (3 levels) are available, which could support fuzzy matching to other channels. Categories are heavily beauty/personal care focused -- the right FMCG categories for a barcode project, but data must be sourced externally.

G. 7-ELEVEN

Assessment: Zero barcode AND zero brand data. Maximum difficulty for matching. The product catalog appears to be ibon/online shop (jewelry, home appliances, gourmet food) rather than in-store convenience items.

H. Momo / Watsons (屈臣氏)

No data in channel_master.csv. These channels were not crawled or are under different channel codes. Need to verify if they exist under alternative names (e.g., shopee_watsons has 2,164 products but also zero GTINs).

Section 3: Cross-Channel Barcode Overlap | 跨通路條碼重疊

Overlap Statistics

Key Pairwise Overlaps (shared unique GTINs)

Key insight: The PX family (PXBox + PXGo + PXMart_Mega) forms a highly overlapping barcode cluster. Their combined unique GTINs cover the broadest domestic FMCG barcode set. Carrefour adds ~20K additional unique GTINs. iherb operates in a nearly independent barcode universe (supplements/health foods).

Section 4: Category-Level Barcode Coverage | 品類層級條碼覆蓋率

Best-Covered FMCG Categories (from category_stats.md)

Worst-Covered FMCG Categories

Section 5: Difficulty Coefficient | 困難度係數

Methodology

Difficulty Score (0-100) = 
    Barcode_Gap * 0.40        (% without valid barcode)
  + GTIN_Gap * 0.20           (% without valid GTIN specifically)  
  + Brand_Gap * 0.15          (% without brand data - affects matching ability)
  + Category_Diversity * 0.15 (normalized 0-100)
  + Dedup_Issues * 0.10       (100 - dedup_ratio)

Difficulty Rankings for Key Channels

Global Difficulty Distribution (all 67 channels)

Section 6: Strategic Recommendations | 策略建議

For Contract Pricing

1. Tier-based pricing is essential. A flat per-channel rate would severely underprice hard channels and overprice easy ones. The difficulty score spans 27 to 100.

2. The "Big 4" problem: PChome (980K), ETmall (425K), Cosmed (125K), and TRPlus (72K) collectively represent 1.6M products with ZERO barcode data. Any contract promising barcode coverage for these channels requires non-scraping methods (image recognition, manufacturer database, manual lookup, or cross-channel name matching).

3. PX family is the best starting point: PXGo + PXMart_Mega combined give ~15K unique GTINs at >92% valid rate -- an excellent seed dataset for FMCG barcode matching against other channels.

4. Carrefour's 24K invalid-format barcodes are a potential "quick win" -- these may be recoverable with format normalization, adding ~20% more coverage at minimal cost.

5. Cross-channel matching potential: 85% of GTINs exist in only 1 channel. The overlap between PX-family and Carrefour is ~22%. This means each channel adds substantial unique barcode inventory.

Suggested Difficulty Multipliers for Contract

Report generated from barcode_coverage_analysis.py Output files: barcode_coverage_analysis.json, channel_difficulty_scores.csv