AI-driven marketing portraits coincide with noticeably better conversion: across 50 Chinese cross-border merchants average ad click-through rose from 1.21% to 2.07% and 90‑day repurchase from 6.23% to 11.46% after deployment. The reported gains are promising for practitioners but are based on descriptive pre/post comparisons and may reflect concurrent investments and selection effects rather than a clean causal impact.

Main Finding

AI-powered user marketing portraits built from commercial big data materially improve cross-border e‑commerce conversion across the full funnel. After deploying AI portraits, the 50 sampled Chinese exporters saw meaningful increases in ad click-through, add-to-cart, order conversion, and 90‑day repurchase rates. The paper also identifies five practical optimization paths and four implementation barriers (data silos, weak algorithm adaptability, cross‑border compliance risk, fragmented conversion chains).

Key Points

• Definition and scope

  • Commercial big data: in‑platform behavior, overseas social media/ad data, transaction/support data, and basic user attributes.
  • AI marketing portrait: dynamic label system (basic attributes; consumption behavior; interest/preferences; customer value) generated by ML clustering and collaborative filtering.
  • Conversion funnel: Ad exposure → Product click → Store visit → Add‑to‑cart & order → Repurchase.

• Modeling & engineering workflow

  • Data ingestion from four channels → preprocessing (cleaning, standardization, desensitization, fusion) → feature extraction → weight assignment → automatic segmentation (K‑means, collaborative filtering) → exportive interfaces to ad backends, recommender systems, private‑domain tools, after‑sales.
  • Portraits must be integrated with marketing systems to realize conversion gains.

• Standardized indicator system (four layers)

  • Basic attributes: country, time zone, device, demographics.
  • Consumption behavior: traffic behavior, transaction stats, repurchase behavior.
  • Interest/preferences: category, content, campaign format preferences.
  • Customer value: lifecycle segment, revenue contribution, CAC.

• Empirical outcomes (50 enterprises; 2025–H1 2026)

  • Average ad click‑through rate: 1.21% → 2.07% (+0.86 pp; ≈71% relative)
  • Average add‑to‑cart rate: 4.35% → 7.12% (+2.77 pp; ≈64% relative)
  • Average store order conversion rate: 1.08% → 1.95% (+0.87 pp; ≈81% relative)
  • Average 90‑day repurchase rate: 6.23% → 11.46% (+5.23 pp; ≈84% relative)
  • Heterogeneous gains by firm type (order conversion growth):
    • Small & medium sellers: +0.52% (limited infra, budgets CNY 5k–30k/yr)
    • Premium independent sites: +0.91% (CNY 30k–150k/yr)
    • Leading brand/platform stores: +1.38% (>CNY 150k/yr)

• Key implementation barriers

  • Data silos and poor governance (manual integration, missing unstructured sources).
  • Weak algorithm adaptability to diverse platforms and markets.
  • Overseas data compliance risks (GDPR, CCPA, local rules; consent, cross‑border limits).
  • Fragmented conversion chain (portraits not connected to full range of marketing tools).

• Recommended optimization paths

  • Build unified data middle platform and governance.
  • Iteratively update and localize AI models (market/language adaptation).
  • Design hierarchical/tiered marketing strategies linked to portrait segments.
  • Implement full‑link conversion transformation (ads → onsite → private domain → after‑sales).
  • Establish cross‑border data compliance management (consent flows, desensitization, regional controls).

Data & Methods

• Methods
  • Literature review on big data, user portraits, cross‑border e‑commerce.
  • Empirical analysis using operational data from 50 domestic cross‑border enterprises (three firm types; data drawn 2025–H1 2026).
  • Case analyses of three representative firms and comparative analysis vs. manual marketing.
• Data sources and preprocessing
  • Four channels: in‑site, overseas social media ads, transaction support, basic attributes.
  • Preprocessing: deduplication, cleaning, format standardization, desensitization/encryption, fusion of multi‑channel records per user.
• Modeling
  • Feature extraction (price sensitivity, access windows, category preference, repurchase cycle, etc.), feature weighting (emphasize order/repurchase predictors), clustering/segmentation (K‑means, collaborative filtering), daily updates to labels.
• Empirical limitations (noted by authors / inferable)
  • Observational before‑after comparisons without explicit randomized controls; potential confounders (macro trends, advertising budget changes) could bias estimates.
  • Short follow‑up window and sample of 50 firms concentrated in one origin country; heterogeneity across destination markets may be underexplored.
  • Reported investment ranges and gains are averages; firm‑level ROI depends on margin, ad spending, and infrastructure cost.

Implications for AI Economics

• Measurable productivity gains and ROI potential
  • The reported uplifts—especially in repurchase—imply notable improvements in lifetime value (LTV) per customer and lower wasted ad spend. Economists can use the reported deltas as plausible ranges for modeling adoption payoffs and payback periods.
• Heterogeneous returns and scale effects
  • Returns increase with firm digital maturity and budget. This suggests adoption may exacerbate concentration: larger firms with data platforms capture larger efficiency gains, raising entry and competition implications.
• Policy and regulation interactions
  • Data protection regimes materially affect implementation design and costs. Compliance imposes additional fixed costs (governance, desensitization, legal) that can disadvantage smaller sellers; policy choices (e.g., standardized consent frameworks) will influence market structure and welfare.
• Labor and organizational implications
  • Demand shifts from manual marketing to data engineering/model ops and localized AI adaptation skills. Investments in lightweight tools for SMEs could democratize gains.
• Research opportunities & recommended empirical strategies
  • Use randomized controlled trials (A/B tests) or difference‑in‑differences with matched controls to isolate causal effects of portrait deployment.
  • Study long‑run effects on LTV, churn, and price/margin dynamics across destination markets.
  • Quantify compliance costs and how different regulatory regimes alter adoption speed and market concentration.
• Practical guidance for economists advising firms
  • Estimate expected conversion uplifts by firm type using the paper’s average deltas, then compute incremental revenue vs. implementation + annual operating cost (given provided CNY bands).
  • Model scenario analyses: (a) small seller with limited spend—focus on lightweight portrait tools and vendor integrations; (b) platform leader—invest in multilingual, multi‑regional models and strict compliance tooling.

If you want, I can: - Convert the paper’s average conversion improvements into a simple ROI template (inputs: avg order value, monthly visitors, margin) to estimate payback time; or - Draft an outline for a causal evaluation (RCT or DiD) to strengthen the evidence on the effectiveness of AI portraits.