Sales Automation
Shanghai Xinzhi Software needed a scalable way to help automotive sales teams cope with high staff turnover, expensive talk-track training, fragmented customer-profile information, and weak use of historical call data in coaching and conversion planning.
Reported impact
The sales copilot improved lead conversion, raised salesperson scores, reduced training time, lowered manual quality-inspection work, and gave managers a data-backed way to optimize coaching and market strategy.
Shanghai Xinzhi Software built an automotive-sales copilot on Baidu AI Cloud using Qianfan ModelBuilder, Qianfan AppBuilder, and ERNIE-4.0-Turbo. The system ingests sales-conversation content and related customer information, then uses prompt-tuned large-model workflows to extract customer tags, concerns, resistance points, and intent labels. Qianfan AppBuilder organizes interview analysis, scoring, and recommendation delivery, while Qianfan ModelBuilder and ERNIE-4.0-Turbo handle talk-track quality checks, improvement suggestions, customer-insight analysis, and strategy guidance. The source also notes automatic connection to Baidu Search plugins to improve sales Q&A. The result turns each interaction into reusable data for manager dashboards, salesperson review, training, and standardized best-practice replication, helping the business shift from experience-led management to data-driven coaching and conversion optimization.
Sales-conversation content and customer information are fed into Qianfan workflows, where ERNIE-4.0-Turbo and tuned prompts extract customer tags, concerns, and intent signals. The system then produces scores, talk-track improvement suggestions, and strategy recommendations that flow into manager monitoring, salesperson review, and training loops.
Baidu AI Cloud provided ERNIE-4.0-Turbo and prompt-tuning services for customer-tag extraction, talk-track quality inspection, improvement suggestions, reception scoring, and customer-insight analysis (original: 百度智能云提供ERNIE-4.0-Turbo基座大模型…提供Prompt提示词调优服务…如客户标签提取、话术质量检测、话术改进建议、接访评分生成、客户洞察分析等).
Sales conversion increased 30%, average sales scores improved 25%, and the training cycle was shortened by 50% (original: 30% 销售转化率提升；25% 销售得分平均提升；50% 培训周期缩短).
The project targeted three core problems: high turnover among sales consultants, high learning costs for talk-track knowledge, and weak use of past call data for strategy analysis (original: 销售顾问流动大，话术知识学习成本高…策略分析支撑少，过往通话数据难以发挥数据价值).
Through the Qianfan platform and prompt tuning, the system can accurately extract customer tags, concerns, resistance information, and intent labels to support intelligent Q&A and strategy recommendation (original: 通过百度智能云千帆大模型平台和提示词调优，能精准提取客户标签、关注点、抗性信息及意向标签，构建多维客户画像，并基于大模型实现智能问答与策略推荐).
Xinzhi built an automotive-sales intelligent assistance system on Baidu AI Cloud's Qianfan large-model platform and ERNIE-4.0-Turbo to enable precise customer profiling, intelligent strategy recommendations, and real-time talk-track optimization (original: 新致软件基于百度智能云千帆大模型平台与ERNIE-4.0-Turbo能力，打造汽车销售智能辅助系统，实现客户画像精准构建、销售策略智能推荐与话术实时优化).
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Google Cloud Sample Modified · Oct 9, 2026 Sansan Sansan needed to unify fragmented customer and sales data so teams could move from static reporting to immediate, data-driven sales actions without relying on manual aggregation or hallucination-prone AI outputs. BigQuery · Gemini · LookML 
MongoDB Sample Sep 30, 2026 MongoDB MongoDB needed to stop sales reps from spending 4 to 6 hours per account or deal cycle manually stitching together customer context from Salesforce, email, meeting transcripts, and documents before they could plan… Holly currently runs on 30 agents. Atlas Agent Engine 
OpenAI Sample Sep 25, 2026 Proaction Proaction needed a way to create highly personalized fleet-management demos and operational follow-ups without consuming scarce engineering time, while also extending automation into day-to-day fleet workflows for… 30 to 45 minutes to build each demo ChatGPT-5.6 Sol · Codex · GPT-6 Astra 
