How Unified AI Interaction Boosted Merchant Efficiency in Baidu’s E‑Commerce Backend

Background

AI capabilities have been integrated into Baidu’s merchant backend across product creation, operation management, and customer‑service scenarios. Merchants initially faced a “don’t know how, don’t trust” problem caused by inconsistent interaction patterns and confusing visual cues, which increased the cognitive load of using AI tools.

Three Interaction Paradigms

2.1 Embedded Interaction – Flexible Assistant for Complex Forms

The embedded paradigm places AI‑generated results directly beside the user’s input fields, providing lightweight assistance for dense forms such as product‑listing creation.

Fill‑recognition type : When the target field has clear, rule‑based content (e.g., packaging image → “dry‑cured”, “boxed”), the AI extracts the key attribute and auto‑populates the corresponding module. A visible AI pre‑fill tag indicates that the value was generated automatically and invites the merchant to verify it.

Recommendation‑optimization type : For content that requires creative improvement (e.g., main product image, title), the AI proposes an enhanced version but does not auto‑fill. Merchants can review the suggestion, edit it, and manually apply the result, preserving control over creative decisions.

After deployment, average product‑listing time decreased by 8 minutes , the success‑rate of listings rose by 0.7 % , and merchant satisfaction with the module grew by 25 % .

2.2 Accompanying Interaction – Proactive Diagnostic Assistant Across All Scenarios

The accompanying paradigm continuously monitors the merchant’s workflow and pushes context‑relevant insights before the merchant asks a question.

Proactive outreach : An animated entry point with rotating messages displays key metrics (e.g., “store visits up 20 % today”). When a merchant dwells on a task page, the assistant analyses page content and business logic, detects potential issues (e.g., sub‑optimal image size), and issues a lightweight reminder.

Continuous insight : Each suggestion includes a clear reasoning chain and a source summary, enabling explainable AI. The assistant also predicts the next logical step and recommends actions (e.g., “investigate why experience score dropped”), turning isolated queries into a systematic problem‑solving workflow.

This design shifts AI from a passive Q&A bot to an “active business partner”.

2.3 Managed Interaction – Implicit, Trustworthy Service Butler

The managed paradigm achieves the highest degree of automation by letting merchants pre‑configure response rules that the system executes without real‑time human input. It is applied primarily to customer‑service scenarios.

Visual configuration builds trust : Merchants use a strategy panel to define reply rules for pre‑sale inquiries, order reminders, refunds, etc. A preview page shows simulated AI responses, so merchants can see exactly how the AI will behave before activation.

Clear status for seamless handoff : Global and per‑session status indicators update in real time. If the AI cannot answer a query (e.g., complex complaint), the conversation is automatically moved to a “awaiting human reply” queue with visual highlighting and audio alerts. Once an agent replies, the AI pauses, generates a concise conversation summary, and can resume handling subsequent messages.

Post‑deployment metrics show a 15.8 % reduction in response time , a 14 % increase in merchant satisfaction , and a 7.4 % rise in buyer satisfaction**, delivering 24/7 scalable service.

Intelligent Perception System

Based on the three paradigms, a unified visual language and perception system was built to span the entire product lifecycle. The system inherits Baidu APP’s AI branding colors, adapts them to a low‑saturation palette suitable for dense B‑side interfaces, and adds state animations and sound cues. These multimodal feedback mechanisms ensure that merchants instantly perceive AI activity even on complex screens.

Core Principle: Merchant‑Centric Efficiency Enablement

The upgrade follows the principle of “unified AI cognition to improve merchant efficiency”. By selecting and innovating the three interaction paradigms, constructing a visual‑plus‑multisensory perception layer, and addressing the “don’t know how, don’t trust” pain points, the system delivers measurable business outcomes:

Faster product listing (‑8 min per item)

Higher issue‑resolution rates through proactive diagnostics

Improved service satisfaction for both merchants and buyers

The design methodology and component assets are reusable for other AI‑enabled scenarios such as live‑stream sales and intelligent ad placement.