Analysis · Business
An AI Merchant Ranking Cannot Tell You Which Sales Channel Wins
By AWEI · AI-compiled · Published · Analysis prepared · 1 source · doi.org
The three merchants at the top of one AI-generated performance ranking all used overseas warehouses alongside their own websites or platforms. That is an intriguing pattern for cross-border sellers, but it does not establish that copying those arrangements would improve another business.
The ranking came from a 2025 study by Rundong Liao and Yuxi Chai, published in Scientific Reports. They developed a performance evaluation method using records from 10 small and medium-sized cross-border e-commerce businesses in eastern coastal China. The methods section describes quarterly data spanning 2020–2023, covering profitability, debt, liquidity, asset turnover and revenue growth.
Their approach paired an LSTM neural network, designed to learn patterns across time, with an artificial bee colony algorithm. The bee-inspired component searched for useful model settings, such as learning rate and hidden-unit count. Its role was to tune the prediction system; it did not simulate shoppers or test merchant decisions.
The authors reported a mean absolute prediction error of 0.016 for the combined model, compared with 0.035 for the ordinary LSTM. These are model evaluation figures, not percentage increases in sales or profit. The comparison suggests that tuning improved accuracy within the reported analysis, without demonstrating that using the model improves business outcomes.
There are important limits. The paper describes a random training/test split and later a stratified split, leaving the validation procedure unclear. It does not clearly establish a test that trains only on earlier quarters and predicts later ones. Ten firms from one broad region also offer limited evidence for other markets, and the inputs exclude external shocks such as logistics disruption.
For merchants, the useful distinction is between predicting a performance score and explaining its causes. The warehouse pattern is descriptive; the proposed benefits of channel control remain explanations rather than tested effects. This method offers a candidate evaluation tool, not a validated recipe for choosing a sales channel.
Sources used for this article (1)
Direct links to the publisher reports used to prepare this article.
- Source 1
- Research on the business performance evaluation method for small and medium-sized enterprises in cross-border e-commerce based on artificial bee colony optimized LSTM model Rundong Liao, Yuxi Chai
