K-文化的一切——从回归到 K-美妆，发送到您的邮箱订阅邮件

METAL MEDIA

Bridging Search and CRM: Productionizing AI Product Research Agents for Customer Re-Engagement

arXiv:2608.185432026-08-20

AI智能体追着离场用户发WhatsApp,把逛而不买的顾客拉回来

针对像“最好的智能手机”这类含糊搜索后就跑去外部查资料的顾客,Flipkart搭建了一套多智能体协作系统,自动做产品调研、核实信息,再通过WhatsApp发送个性化推荐,已在生产环境正式上线。为期23天、约发送1.5万条WhatsApp消息的实测显示,点击率比以往的WhatsApp营销活动高出约285%,还观察到用户自发转发消息的现象,并带动了后续实际购买和成交额(GMV)增长。

METAL MEDIA 解读图

AI智能体追着离场用户发WhatsApp,把逛而不买的顾客拉回来

  1. 01用大规模数据处理框架PySpark对搜索日志做筛选,锁定零点击、高消费力用户、含“最好的”“最新的”等主观词汇的查询,并聚焦手机品类
  2. 02查询分析智能体先理解用户意图,发现智能体通过网络搜索、评测和视频等外部信息挖掘候选产品,Flipkart搜索智能体再把候选商品对接到真实库存、配送范围和个性化价格
  3. 03审核智能体重新核对产品参数和上市日期是否与目录及外部信息一致,过滤掉不准确的推荐,减少AI胡编乱造(幻觉)的情况
  4. 04采用统一“指挥官”智能体协调各环节的分层架构,比按固定顺序依次执行的架构指令违规率低得多(基于2200条推荐样本的测试)
  5. 0523天实测中点击率比历史WhatsApp手机品类营销活动高约285%,多天访问量甚至超过发送消息数(说明用户自发转发),后续还带来了实际购买和成交额提升
这是 METAL MEDIA 制作的解读图，并非论文作者提供的原图。

他们做了什么

  1. 用大规模数据处理框架PySpark对搜索日志做筛选,锁定零点击、高消费力用户、含“最好的”“最新的”等主观词汇的查询,并聚焦手机品类
  2. 查询分析智能体先理解用户意图,发现智能体通过网络搜索、评测和视频等外部信息挖掘候选产品,Flipkart搜索智能体再把候选商品对接到真实库存、配送范围和个性化价格
  3. 审核智能体重新核对产品参数和上市日期是否与目录及外部信息一致,过滤掉不准确的推荐,减少AI胡编乱造(幻觉)的情况
  4. 采用统一“指挥官”智能体协调各环节的分层架构,比按固定顺序依次执行的架构指令违规率低得多(基于2200条推荐样本的测试)
  5. 23天实测中点击率比历史WhatsApp手机品类营销活动高约285%,多天访问量甚至超过发送消息数(说明用户自发转发),后续还带来了实际购买和成交额提升
Figure 1. Proposed architecture of the Research Agent for Product Recommendations.
Figure 1. Proposed architecture of the Research Agent for Product Recommendations.
Table 1. Comparison of Historical WhatsApp (WA) Metrics for mobile product campaigns and AI Agent Campaign Interaction Statistics (23-Day Average).
% WA message readsCTR
AI Agent campaign~+8%~+285%
Table 2. Quantitative manual evaluation of factual accuracy
No. of productsspec accuracylaunch date accuracy
221899.1 %99.2 %
Table 3. Effect of the choice of the agent architecture. It is observed that hierarchical supervisor-worker architecture provides superior performance.
ArchitectureInstruction Violation Rate (%)
Centralized orchestration8.5%
Sequential Agents35.4%

为什么重要

这展示了一个把搜索、AI推理和CRM消息串联起来、真正上线运行的案例,能把原本要跑去外部研究再流失的用户重新拉回并促成购买,而不只是停留在实验阶段。对于想把智能体AI用在营销转化而不仅是聊天机器人的团队,这是一个附带点击率和GMV等实际业务数据的参考范例。

本文术语

  • PySpark · 用于快速处理海量日志数据的分布式计算框架
  • 点击率(CTR) · 收到消息的人中实际点击链接的比例
  • 指挥官智能体(orchestrator) · 负责协调多个AI智能体执行顺序和信息传递的控制角色
  • GMV · 一定时期内平台上商品交易的总金额
  • UTM标签 · 附加在网址上用来追踪点击来源于哪条消息或活动的标记

论文原文摘要（英文）

Modern e-commerce platforms often operate search, recommendation, personalization, and CRM systems independently, limiting opportunities for proactive customer re-engagement. This is particularly challenging for exploratory intents such as best smartphones or latest 5G phones, where users may leave the platform for external research before purchasing. We present a scalable, production-deployed framework that bridges search and CRM workflows through AI-powered Product Research Agents. The system identifies users with exploratory purchase intent and low engagement, conducts grounded multi-agent product research using behavioral signals, external knowledge, and enterprise catalog data, and delivers personalized recommendations through WhatsApp. We evaluate the framework in a 23-day production deployment involving approximately 15K WhatsApp notifications for mobile product discovery. The campaign achieved substantial CTR improvements over traditional WhatsApp recommendation campaigns, with evidence of secondary engagement through message forwarding and sharing. The deployment also generated downstream purchases and GMV impact, demonstrating the practical effectiveness of AI Product Research Agents for proactive customer re-engagement and end-to-end customer journey optimization.

作者 · Mandar Kulkarni, Pooja A., Samir Shah

在 arXiv 阅读

最新论文

全部论文 →

METAL MEDIA 最新报道

图片来源： Mandar Kulkarni et al., arXiv:2608.18543, CC BY 4.0